Sébastien Piérard

My Publications

Here is the list of my publications (provided by orbi). If you prefer, you can also find another list with bibtex entries at the website of the "Telecommunications and Imaging Laboratory" (here). Moreover, my profile provided by Google Scolar can be found here.

Full Text
See detailImproving pedestrian detection using motion-guided filtering
Wang, Yi; Pierard, Sébastien; Su, Song-Zhi; Jodoin, Pierre-Marc

in Pattern Recognition Letters (2017), 96

In this letter, we show how a simple motion-guided nonlinear filter can drastically improve the accuracy of several pedestrian detectors. More specifically, we address the problem of how to pre-filter an image so almost any pedestrian detector will see its false detection rate decrease. First, we roughly identify moving pixels by cumulating their temporal gradient into a motion history image (MHI). The MHI is then used in conjunction with a nonlinear filter to filter out background details while leaving untouched foreground moving objects. We also show how a feedback loop as well as a merging procedure between the filtered and the unfiltered frames can further improve results. We tested our method on 26 videos from 6 categories. The results show that for a given miss rate, filtering out background details reduces the false detection rate by a factor of up to 69.6 times. Our method is simple, computationally light, and can be implemented with any pedestrian detector. Code is made publicly available at: https://bitbucket.org/wany1601/pedestriandetection.

Full Text
See detailA two-step methodology for human pose estimation increasing the accuracy and reducing the amount of learning samples dramatically
Azrour, Samir; Pierard, Sébastien; Geurts, Pierre; Van Droogenbroeck, Marc

in Advanced Concepts for Intelligent Vision Systems (2017, September)

In this paper, we present a two-step methodology to improve existing human pose estimation methods from a single depth image. Instead of learning the direct mapping from the depth image to the 3D pose, we first estimate the orientation of the standing person seen by the camera and then use this information to dynamically select a pose estimation model suited for this particular orientation. We evaluated our method on a public dataset of realistic depth images with precise ground truth joints location. Our experiments show that our method decreases the error of a state-of-the-art pose estimation method by 30%, or reduces the size of the needed learning set by a factor larger than 10.

Full Text
See detailSemantic Background Subtraction
Braham, Marc; Pierard, Sébastien; Van Droogenbroeck, Marc

in IEEE International Conference on Image Processing (ICIP), Beijing 17-20 September 2017 (2017, September)

We introduce the notion of semantic background subtraction, a novel framework for motion detection in video sequences. The key innovation consists to leverage object-level semantics to address the variety of challenging scenarios for background subtraction. Our framework combines the information of a semantic segmentation algorithm, expressed by a probability for each pixel, with the output of any background subtraction algorithm to reduce false positive detections produced by illumination changes, dynamic backgrounds, strong shadows, and ghosts. In addition, it maintains a fully semantic background model to improve the detection of camouflaged foreground objects. Experiments led on the CDNet dataset show that we managed to improve, significantly, almost all background subtraction algorithms of the CDNet leaderboard, and reduce the mean overall error rate of all the 34 algorithms (resp. of the best 5 algorithms) by roughly 50% (resp. 20%). Note that a C++ implementation of the framework is available at http://www.telecom.ulg.ac.be/semantic.

Full Text
See detailLaBGen: A method based on motion detection for generating the background of a scene
Laugraud, Benjamin; Pierard, Sébastien; Van Droogenbroeck, Marc

in Pattern Recognition Letters (2017), 96

Given a video sequence acquired with a fixed camera, the generation of the stationary background of the scene is a challenging problem which aims at computing a reference image for a motionless background. For that purpose, we developed our method named LaBGen, which emerged as the best one during the Scene Background Modeling and Initialization (SBMI) workshop organized in 2015, and the IEEE Scene Background Modeling Contest (SBMC) organized in 2016. LaBGen combines a pixel-wise temporal median filter and a patch selection mechanism based on motion detection. To detect motion, a background subtraction algorithm decides, for each frame, which pixels belong to the background. In this paper, we describe the LaBGen method extensively, evaluate it on the SBI 2016 dataset and compare its performance with other background generation methods. We also study its computational complexity, the performance sensitivity with respect to its parameters, and the stability of the predicted background image over time with respect to the chosen background subtraction algorithm. We provide an open source C++ implementation at http://www.telecom.ulg.ac.be/labgen.

Full Text
See detailLaBGen-P: A Pixel-Level Stationary Background Generation Method Based on LaBGen
Laugraud, Benjamin; Pierard, Sébastien; Van Droogenbroeck, Marc

in 2016 International Conference on Pattern Recognition Contest Proceedings (2016, December)

Estimating the stationary background of a video sequence is useful in many applications like surveillance, segmentation, compression, inpainting, privacy protection, and computational photography. To perform this task, we introduce the LaBGen-P method based on the principles of LaBGen and the conclusions drawn in the corresponding paper. It combines a pixel-wise median filter and a pixel selection mechanism based on a motion detection performed by the frame difference algorithm. By working with pixels instead of patches, as originally done in LaBGen, it avoids some discontinuities between different spatial areas and generates better visual results. In this paper, we describe the LaBGen-P method, study its performance on the sequences of the SBMnet dataset, and compare it to that of LaBGen and other methods on the same dataset. Both algorithms emerged as the best ones during the IEEE Scene Background Modeling Contest (SBMC) organized in 2016. However, as there is not yet a good understanding of the recommended metrics, and due to the small amount of video sequences provided with the corresponding ground truth, we have performed a subjective evaluation. More precisely, 35 human experts were asked to compare background images estimated by LaBGen-P and LaBGen, and select the best one. From these experiments, it turns out that the results of LaBGen-P are preferred for about two thirds of the video sequences. Note that we provide an open-source C++ implementation at http://www.telecom.ulg.ac.be/labgen.

Full Text
See detailWhat are the optimal walking tests to assess disability progression?
Pierard, Sébastien; Van Droogenbroeck, Marc

in Multiple Sclerosis Journal (2016, September 16), 22(S3), 452-453

Background. Therapy success is assumed when there is no evidence of disease activity. Clues to show it include an MRI, the relapses history, questionnaires, and clinical measures to assess the disability progression. Especially gait analysis plays a major role as gait impairment is considered by patients as the most disabling symptom. Too often only the walking speed is measured. New technologies (eg GAIMS, see ECTRIMS 2012-15) measure many spatiotemporal gait parameters, even during long tests (\eg 6min, 500m), without equipping patients with markers or sensors. Moreover, various tests can be done, depending on the length and type of walk (comfortable pace --C--, as fast as possible --F--, tandem gait --T--). Objective. Determine if there is an advantage to perform various walking tests, and which test or combination of tests brings the higher amount of information about the patient state in a reasonable amount of acquisition time. Methods. The system GAIMS provided 434 recordings of the gait parameters of healthy people and 60 recordings of MS patients with EDSS<= 4. They performed 12 tests (25ft C+F+T each twice, 20m C+F+T, 100m C+F, 500m F). To assess the ability of these clinical outcome measures to detect disability progression, we evaluate the possibility of differentiating the persons below a given EDSS threshold (0.25) from those above it based only on the measured gait parameters. For individual tests, we use the classifier of Azrour (ESANN 2014). All subsets of the tests are also considered, by combining the individual classifiers and determining automatically the optimal relative importance of the tests with the linear support vector machine (SVM) technique. The ability to detect the disability progression is quantified by the performance (area under the ROC curve --AUC-- and the maximum achievable balanced accuracy --MBA--) of the corresponding classifiers. Results. The best test alone is the 500m F (note that the walking speed measured during it is the gait parameter best correlated with the EDSS). Combining several tests leads to a better performance. A performance (MBA=95.7%, AUC=0.983) close to the best achievable one can be obtained with 6 tests only (25ft C twice, 25tf F twice, 20m C, 20m T). Conclusions. The clinical gait analysis can help to detect disability progression. While considering different types of walking tests improves the ability of taking decisions, we showed that performing 6 tests for a total of 70.48m suffices.

Full Text
See detailBoosting shape classifiers accuracy by considering the inverse shape
Pierard, Sébastien; Lejeune, Antoine; Van Droogenbroeck, Marc

in Journal of Pattern Recognition Research (2016), 11(1), 41-54

Many techniques exist for describing shapes. These techniques almost exclusively consider the contour or the inside of the shape; the major problem for describing the outside of a shape, or inverse shape, being that it has an infinite extension. In this paper, we show how to adapt two shape descriptors, one region based, the Cover By Rectangles, and one transform based, the Zernike moments, to be applicable to the inverse shape. We analyze their properties, and show how to deal with the infinite extension of the inverse shape. Then, we apply these descriptors to shape classification and compare representations that use the shape, its inverse, or both. Our experiments establish that, for shape classification, a representation integrating the inverse shape often outperforms a representation restricted to the shape. This opens the path for better techniques that could use, as a rule of thumb, both the representations of a shape and its inverse for the purpose of classification.

Full Text
See detailLeveraging orientation knowledge to enhance human pose estimation methods
Azrour, Samir; Pierard, Sébastien; Van Droogenbroeck, Marc

in Articulated Motion and Deformable Objects AMDO 2016 (2016, July)

Predicting accurately and in real-time 3D body joint positions from a depth image is the cornerstone for many safety, biomedical, and entertainment applications. Despite the high quality of the depth images, the accuracy of existing human pose estimation methods from single depth images remains insufficient for some applications. In order to enhance the accuracy, we suggest to leverage a rough orientation estimation to dynamically select a 3D joint position prediction model specialized for this orientation. This orientation estimation can be obtained in real-time either from the image itself, or from any other clue like tracking. We demonstrate the merits of this general principle on a pose estimation method similar to the one used with Kinect cameras. Our results show that the accuracy is improved by up to 45.1 %, with respect to a method using the same model for all orientations.

Full Text
See detailImproving pose estimation by building dedicated datasets and using orientation
Azrour, Samir; Pierard, Sébastien; Van Droogenbroeck, Marc

Poster (2016, May 19)

Markerless pose estimation systems are useful for various applications including human- computer interaction, activity recognition, security, gait analysis, and computer-assisted medical interventions. They have attracted much interest since the release of low-cost depth cameras such as Microsoft’s Kinect camera. Shotton et al. and Girshick et al. pioneered tractable methods that infer a full-body pose reconstruction in real-time. Despite this technological breakthrough, the accuracy of human pose estimation from single depth images remains insufficient for some applications. Our work aims at building a simulation environment to create images databases suited for any camera position and improving the mainstream machine learning-based pose estimation algorithms.

Full Text
See detailSlicing the 3D space into planes for the fast interpretation of human motion
Pierard, Sébastien; Azrour, Samir; Van Droogenbroeck, Marc

Poster (2016, May 19)

Full Text
See detailDefining a score based on gait analysis for the longitudinal follow-up of MS patients
Azrour, Samir; Pierard, Sébastien; Van Droogenbroeck, Marc

in Multiple Sclerosis Journal (2015, October 09), 23(S11), 408-409

BACKGROUND. The project GAIMS [ECTRIMS 2013 P800] aims at developing a gait measuring system particularily suited for the clinical routine, and providing a reference database with the gait characteristics of many MS patients (MSP) and healthy people (HP). As the gait impairments are related to the disease progression, defining an objective and quantitative score based on the gait characteristics would be useful for the longitudinal follow-up. Based on the dataset of GAIMS and machine learning techniques (MLT), a score, well correlated with the EDSS, can be defined [Azrour et al. ESANN 2014]. OBJECTIVE. Burggraaff et al. [ECTRIMS 2014 P033] showed that paired comparisons can help human raters to better judge the state of the patients. In the same spirit, we aim at predicting the difference of EDSS between two persons or between two visits of a same person, based on clinical gait measures. We show that the pairwise comparison strategy leads to a score (Gait-Score) well correlated with the EDSS and sensitive to small modifications of the gait. METHODS. The gait of 162 HP and 72 MSP (44 with EDSS>3) has been recorded and analyzed with GAIMS. The Gait-Score is defined using the MLT of [Geurts et al. 2006]. We can compute the Gait-Score of a person by comparing him to others with known EDSS, and compute the difference of Gait-Score of a same person at two different moments. We measure the merits of the Gait-Score by the correlation between the predicted Gait-Score and the EDSS, as well as the ability to detect subtle gait deteriorations among people with ataxia induced by a low dose of alcohol (data of [Piérard et al. ESANN 2014]). RESULTS. The Gait-Score is well correlated with the EDSS (Pearson’s correlation=0.8743). Moreover, it manages to detect a gait deterioration after a small alcohol intake for 19 persons out of 24 (79% correct) which is much better than what was obtained by visual inspection of neurologists (62% according to [Piérard et al. ESANN 2014]). CONCLUSIONS. Based on the accurate gait measures provided by GAIMS, we are able to derive a Gait-Score, automatically, that is well correlated with the EDSS. Moreover, this score is able to detect subtle deteriorations of the gait caused by a low dose of alcohol. These results reinforce our conviction that the use of an automatic method based on gait analysis is very promising for the longitudinal follow-up of MS patients and the assessment of the impact of new drugs and rehabilitation programs.

Full Text
See detailUnderstanding how people with MS get tired while walking
Pierard, Sébastien; Phan-Ba, Rémy; Van Droogenbroeck, Marc

in Multiple Sclerosis Journal (2015, October 09), 23(S11), 406

BACKGROUND. Walking impairment is frequent, appears early in the disease course of MS patients (MSP), and is perceived as the most disabling symptom. When walking, patients get tired more and differently than healthy people (HP) [Phan-Ba et al PLOS 2012]. This limits their walking perimeter. Understanding this phenomenon is thus important to suggest adequate therapies at the right time. OBJECTIVE. Our aim is to understand how MSP get tired while walking compared to HP. Two groups of MSP are considered: those with a low disability level (MSPL) and those with a high one (MSPH). We consider two criteria to measure the disability: the EDSS and the deceleration index (DI) [Phan-Ba et al PLOS 2012]. The limit between the groups is set at DI=0.8 and EDSS=3 (inclusive for MSPL). METHODS. Many gait characteristics (GC) have been measured with the system GAIMS along a 500m path walked as fast as possible. The dataset gathers 464 visits of HP and 70 of MSP. Some people have been assessed several times. There are 33 visits in the group MSPL with the EDSS criterion, and 25 with the DI criterion. Statistical tests (Welch) were performed on the differences and relative differences of the GC measured during the first and last 100m of the test to detect differences between HP and MSPL, and between MSPL and MSPH, as in [ECTRIMS 2012 P755]. RESULTS. Both criteria for defining the groups lead to similar conclusions. For many GC, the distributions of the variations are significantly different between MSPL and MSPH. The largest difference is for the relative difference of speed (p=0.000119 for EDSS and p=0.000021 for DI). In contrast, only the variation of the average lateral distance between the feet, which is related to the size of the base of support (and thus to the balance) shows a very significant difference between HP and MSPL (p=0.000116 for EDSS and p=0.000120 for DI). The balance does not seem to change much from MSPL to MSPH. Besides, we note that the variance decreases slightly from HP to MSPL and increases a lot from MSPL to MSPH. CONCLUSIONS. Statistically, from the motor fatigue point of view, it seems that the course of the MS disease is divided in two different stages. In the first one, MSP get more tired than HP because of a deterioration of the balance. Then, in the second one, their fatigue becomes related to a faster decrease of the walking speed. This suggests that physical therapy exercises focused on the balance could be given to MSP in the early stage of the disease.

Full Text
See detailA perfect estimation of a background image does not lead to a perfect background subtraction: analysis of the upper bound on the performance
Pierard, Sébastien; Van Droogenbroeck, Marc

in International Conference on Image Analysis and Processing (ICIAP) (2015, September)

The quest for the “best” background subtraction technique is ongoing. Despite that a considerable effort has been undertaken to develop flexible and efficient methods, some elementary questions are still unanswered. One of them is the existence of an intrinsic upper bound to the performance. In fact, data are affected by noise, and therefore it is illusory to believe that it is possible to achieve a perfect segmentation. This paper aims at exploring some intrinsic limitations of the principle of background subtraction. The purpose consists in studying the impact of several limiting factors separately. One of our conclusions is that even if an algorithm would be able to calculate a perfect background image, it is not sufficient to achieve a perfect segmentation with background subtraction, due to other intrinsic limitations.

Full Text
See detailSimple Median-Based Method for Stationary Background Generation Using Background Subtraction Algorithms
Laugraud, Benjamin; Pierard, Sébastien; Braham, Marc; Van Droogenbroeck, Marc

in New Trends in Image Analysis and Processing - ICIAP 2015 Workshops (2015, September)

The estimation of the background image from a video sequence is necessary in some applications. Computing the median for each pixel over time is effective, but it fails when the background is visible for less than half of the time. In this paper, we propose a new method leveraging the segmentation performed by a background subtraction algorithm, which reduces the set of color candidates, for each pixel, before the median is applied. Our method is simple and fully generic as any background subtraction algorithm can be used. While recent background subtraction algorithms are excellent in detecting moving objects, our experiments show that the frame difference algorithm is a technique that compare advantageously to more advanced ones. Finally, we present the background images obtained on the SBI dataset, which appear to be almost perfect. The source code of our method can be downloaded at http://www.ulg.ac.be/telecom/research/sbg.

Full Text
See detailNonlinear Background Filter to Improve Pedestrian Detection
Wang, Yi; Pierard, Sébastien; Su, Song-Zhi; Jodoin, Pierre-Marc

in New Trends in Image Analysis and Processing - ICIAP 2015 Workshops (2015, September)

In this paper, we propose a simple nonlinear filter which improves the detection of pedestrians walking in a video. We do so by first cumulating temporal gradient of moving objects into a motion history image (MHI). Then we apply to each frame of the video a motion-guided nonlinear filter whose goal is to smudge out background details while leaving untouched foreground moving objects. The resulting blurry-background image is then fed to a pedestrian detector. Experiments reveal that for a given miss rate, our motion-guided nonlinear filter can decrease the number of false positives per image (FPPI) by a factor of up to 26. Our method is simple, computationally light, and can be applied on a variety of videos to improve the performances of almost any kind of pedestrian detectors.

Full Text
See detailIndividual variation of gait characteristics along a 500 meter walk in people with multiple sclerosis and healthy volunteers
Phan-Ba, Rémy; Pierard, Sébastien; Giet, Amaury; Van Droogenbroeck, Marc

in Multiple Sclerosis Journal (2015, April 10), 21(4), 532-533

Background: we previously demonstrated the usefulness of the Deceleration Index (DI, the ratio between the last 100m of the Timed 500-Meter Walk test –T500MW – and the walking speed - WS – of the Timed 25-Foot Walk Test with a propelled start – T25FW+) to evaluate motor fatigue over a long walking distance in people with multiple sclerosis (pwMS). We also recently designed and internally validated a new gait analysis tool for pwMS (GAIMS) that can measure other relevant gait characteristics than the sole WS, such as ataxia, asymmetry and perhaps spasticity. Aims: (i) To compare various gait characteristics between the last and the first 100m of the T500MW in a population of pwMS and healthy volunteers (HV), (ii) to compare the ratio between the last and the first 100m of the T500MW with the DI, and (iii) their relationship with the EDSS. Methods : Subjects were asked to perform the T25FW+ and the T500MW as part of a multimodal evaluation at the MS Clinic of the CHU of Liège. Their gait characteristics were measured using GAIMS. (i) Paired Student’s t-tests were performed on various gait characteristics extracted during the last and first 100m of the T500MW with .05 as a level of significance, (ii) Spearman correlation coefficient (ρ) was calculated (ii) between these ratio and (iii) subject’s EDSS. Results: Seventy-one pwMS and 129 were enrolled in our study. (i) Significant differences were observed for speed related gait characteristics between the last and first 100m of the T500MW, but also for gait characteristics related to ataxia and precision of foot placement. (ii) A moderate positive correlation was observed between the WS ratio of the last and first 100m of the T500MW and the DI. (iii) The correlation between the DI and the EDSS was weakly negative, while the one between the last and first 100m of the T500MW ratio and the EDSS was moderatly negative. Conclusion: (i) As previously demonstrated, we here confirm that alongside to WS, there are other gait features affected by locomotor fatigue over a long walking distance, (ii) the moderate positive correlation between the DI and the last/first 100m of the T500MW indicates that these measures are not the same and that next to a long distance walking test such as the T500MW, a short one such as the T25FW+ remains useful. (iii) The last/first 100m of the T500MW is better correlated to the EDSS and might be a better predictive tool of pwMS’ neurologic state than the DI.

Full Text
See detailIndividual variation of gait parameters along a 500 meter walk in people with multiple sclerosis and healthy volonteers
Phan-Ba, Rémy; Giet, Amaury; Pierard, Sébastien; Van Droogenbroeck, Marc

Poster (2015, April 10)

Background: we previously demonstrated the usefulness of the Deceleration Index (DI, the ratio between the last 100m of the Timed 500-Meter Walk test –T500MW – and the walking speed - WS – of the Timed 25-Foot Walk Test with a propelled start – T25FW+) to evaluate motor fatigue over a long walking distance in people with multiple sclerosis (pwMS). We also recently designed and internally validated a new gait analysis tool for pwMS (GAIMS) that can measure other relevant gait characteristics than the sole WS, such as ataxia, asymmetry and perhaps spasticity. Aims: (i) To compare various gait characteristics between the last and the first 100m of the T500MW in a population of pwMS and healthy volunteers (HV), (ii) to compare the ratio between the last and the first 100m of the T500MW with the DI, and (iii) their relationship with the EDSS. Methods : Subjects were asked to perform the T25FW+ and the T500MW as part of a multimodal evaluation at the MS Clinic of the CHU of Liège. Their gait characteristics were measured using GAIMS. (i) Paired Student’s t-tests were performed on various gait characteristics extracted during the last and first 100m of the T500MW with .05 as a level of significance, (ii) Spearman correlation coefficient (ρ) was calculated (ii) between these ratio and (iii) subject’s EDSS. Results: Seventy-one pwMS and 129 were enrolled in our study. (i) Significant differences were observed for speed related gait characteristics between the last and first 100m of the T500MW, but also for gait characteristics related to ataxia and precision of foot placement. (ii) A moderate positive correlation was observed between the WS ratio of the last and first 100m of the T500MW and the DI. (iii) The correlation between the DI and the EDSS was weakly negative, while the one between the last and first 100m of the T500MW ratio and the EDSS was moderatly negative. Conclusion: (i) As previously demonstrated, we here confirm that alongside to WS, there are other gait features affected by locomotor fatigue over a long walking distance, (ii) the moderate positive correlation between the DI and the last/first 100m of the T500MW indicates that these measures are not the same and that next to a long distance walking test such as the T500MW, a short one such as the T25FW+ remains useful. (iii) The last/first 100m of the T500MW is better correlated to the EDSS and might be a better predictive tool of pwMS’ neurologic state than the DI.

Full Text
See detailGAIMS: a tool specifically developed for the clinical gait analysis of patients with multiple sclerosis
Pierard, Sébastien; Phan-Ba, Rémy; Giet, Amaury; Van Droogenbroeck, Marc

in Multiple Sclerosis Journal (2015, April), 21(4), 498-499

Full Text
See detailAssessment of physical therapy and activity in multiple sclerosis (APTAIMS): development, reliability and validity of a new questionnaire
Giet, Amaury; Pierard, Sébastien; Foidart-Dessalle, Marguerite; Van Droogenbroeck, Marc; Phan-Ba, Rémy

in Multiple Sclerosis Journal (2015, April), 21(4), 531

Full Text
See detail3D human motion analysis based on ultra-low resolution sensors
Pierard, Sébastien

Scientific conference (2015, February 27)

Full Text
See detailThe marauder's map or the use of non-intrusive range laser scanners in the context of smart rooms
Pierard, Sébastien

Scientific conference (2014, October 24)

Dans cette présentation, je vais expliquer comment des capteurs laser peuvent être utilisés pour réaliser différentes fonctionnalités importantes pour les environnements intelligents. Je montrerai comment créer une carte précise et y représenter le mouvement de toutes les personnes, en particulier les trajectoires de leurs pieds. Celles-ci peuvent servir à identifier la personne observée, car chacune a sa propre démarche. Ceci ouvre des voies dans les domaines de la domotique, des environnements intelligents et de la sécurité. Les trajectoires de pieds ont également de nombreuses applications dans le domaine médical, en particulier pour la gériatrie, la kinésithérapie et la neurologie, ce que je détaillerai. Je démontrerai également que cette technologie permet de détecter les situations de piggybacking et de tailgating. Tout ceci étant rendu possible par une chaîne de traitement de signal minutieusement étudiée et par des techniques d'apprentissage automatique.

Full Text
See detailDiagnosing multiple sclerosis with a gait measuring system, an analysis of the motor fatigue, and machine learning
Pierard, Sébastien; Azrour, Samir; Phan-Ba, Rémy; Delvaux, Valérie; Maquet, Pierre; Van Droogenbroeck, Marc

in Multiple Sclerosis Journal (2014, September 11), 20(S1), 171

Full Text
See detailOverview and Benchmarking of Motion Detection Methods
Jodoin, Pierre-Marc; Pierard, Sébastien; Wang, Yi; Van Droogenbroeck, Marc

in Bouwmans; Porikli; Hoferlin; Vacavant (Eds.) Background Modeling and Foreground Detection for Video Surveillance (2014)

In this chapter, we provide an overview of the most highly cited motion detection meth- ods. We identify the most commonly used background models together with their features, the kind of updating scheme they use, some spatial aggregation models as well as the most widely used post-processing operations. We also provide an overview of datasets used to validate motion detection methods. Please note that this literature review is by no means exhaustive and thus we provide a list of surveys that the reader can rely on for further details.

Full Text
See detailIntroduction to Bayes' classifier and to the on-the-fly bayesian domain adaptation
Pierard, Sébastien

Scientific conference (2014, June 19)

Full Text
See detailOn-the-fly domain adaptation of binary classifiers
Pierard, Sébastien; Marcos Alvarez, Alejandro; Lejeune, Antoine; Van Droogenbroeck, Marc

in 23rd Belgian-Dutch Conference on Machine Learning (BENELEARN) (2014, June 06)

This work considers the on-the-fly domain adaptation of supervised binary classifiers, learned off-line, in order to adapt them to a target context. The probability density functions associated to negative and positive classes are supposed to be mixtures of the source distributions. Moreover, the mixture weights and the priors are only available at runtime. We present a theoretical solution to this problem, and demonstrate the effectiveness of the proposed approach on a real computer vision application. Our theoretical solution is applicable to any classifier approximating Bayes' classifier.

Full Text
See detailDesign of a reliable processing pipeline for the non-intrusive measurement of feet trajectories with lasers
Pierard, Sébastien; Azrour, Samir; Van Droogenbroeck, Marc

in International Conference on Acoustics, Speech, and Signal Processing (ICASSP) (2014, May)

Reliable measurements of feet trajectories are needed in some applications, such as biomedical applications. This paper describes the data processing pipeline used in GAIMS, which is a non-intrusive system that measures feet trajectories based on multiple range laser scanners. Our processing pipeline relies on a new tracking paradigm, and it is based on two innovative algorithms: the first algorithm localizes the feet directly from the observed point cloud without any clustering, and the other algorithm identifies the feet. After reviewing the various types of noise affecting the point cloud, this paper explains the limitations of the classical processing approach and gives an overview of our new pipeline. The effectiveness of the proposed approach is established by discussing the results that have been obtained in several studies based on GAIMS.

Full Text
See detailMachine learning techniques to assess the performance of a gait analysis system
Pierard, Sébastien; Phan-Ba, Rémy; Van Droogenbroeck, Marc

in European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN) (2014, April 24)

This paper presents a methodology based on machine learning techniques to assess the performance of a system measuring the trajectories of the lower limbs extremities for the follow-up of patients with multiple sclerosis. We show how we have established, with the help of machine learning, four important properties about this system: (1) an automated analysis of gait characteristics provides an improved analysis with respect to that of a human expert, (2) after learning, the gait characteristics provided by this system are valuable compared to measures taken by stopwatches, as used in the standardized tests, (3) the motion of the lower limbs extremities contains a lot of useful information about the gait, even if it is only a small part of the body motion, (4) a measurement system combined with a machine learning tool is sensitive to intra-subject modifications of the walking pattern.

Full Text
See detailData normalization and supervised learning to assess the condition of patients with multiple sclerosis based on gait analysis
Azrour, Samir; Pierard, Sébastien; Geurts, Pierre; Van Droogenbroeck, Marc

in European Symposium on Artificial Neural Networks, Computational Intelligence and Machine Learning (ESANN) (2014, April)

Gait impairment is considered as an important feature of disability in multiple sclerosis but its evaluation in the clinical routine remains limited. In this paper, we assess, by means of supervised learning, the condition of patients with multiple sclerosis based on their gait descriptors obtained with a gait analysis system. As the morphological characteristics of individuals influence their gait while being in first approximation independent of the disease level, an original strategy of data normalization with respect to these characteristics is described and applied beforehand in order to obtain more reliable predictions. In addition, we explain how we address the problem of missing data which is a common issue in the field of clinical evaluation. Results show that, based on machine learning combined to the proposed data handling techniques, we can predict a score highly correlated with the condition of patients.

Full Text
See detailIntroduction à $BASH (Bourne-again shell)
Pierard, Sébastien

Learning material (2014)

Full Text
See detailStreamlining cutaneous melanomas in young women of the Belgian Mosan region.
Hermanns-Lê, Trinh; Pierard, Sébastien

in BioMed Research International (2014), 2014

Sporadic cutaneous melanoma (SCM) has shown a dramatic increase in incidence in Caucasian populations over the past few decades. A particular epidemiological increase was reported in women during their childbearing age. In the Belgium Mosan region, a progressive unremitting increase in SCM incidence was noticed in young women for the past 35 years. The vast majority of these SCMs were of the superficial type without any obvious relationship with a large number of melanocytic nevi or with signs of frequent and intense sunlight exposures as disclosed by the extent in the mosaic subclinical melanoderma. A series of investigations pointed to a possible relationship linking the development of some SCM to the women hormonal status including the effect of hormonal disruptors. These aspects remain, however, unsettled and controversial. It is possible to differentiate and clearly quantify the SCM shape, size, scalloped border, and varegated pigmentation using computerized morphometry as well as fractal and multifractal methods

Full Text
See detailLes avatars de la 3D: le divertissement et au-delà . . .
Pierard, Sébastien

Conference given outside the academic context (2013)

Full Text
See detailMeasuring feet trajectories: challenges and applications
Pierard, Sébastien; Azrour, Samir; Van Droogenbroeck, Marc

Conference (2013, November 07)

Measuring reliable feet trajectories is needed in many applications. This paper provides the principles used in GAIMS, which is a non-intrusive system that measures feet trajectories based on multiple range laser scanners. We present the technical challenges that we had to address, as well as an overview of the implemented processing pipeline of GAIMS.

Full Text
See detailMetrics to evaluate binary classifiers for video background extraction: weaknesses, strengths, and perspectives
Laugraud, Benjamin; Pierard, Sébastien; Van Droogenbroeck, Marc

Conference (2013, November 07)

Full Text
See detailUsing GAit Measuring System (GAIMS) to discriminate patients with multiple sclerosis from healthy person
Azrour, Samir; Pierard, Sébastien; Van Droogenbroeck, Marc

Poster (2013, November 07)

Among voluntary movements, gait is the most affected by multiple sclerosis. Gait impairment is also a good indicator of the disease progression. However, measurement of gait character- istics made by neurologists is usually limited to the use of a stopwatch. The GAit Measuring System (GAIMS), provides a wider range of measurements that allow the definition of several relevant gait descriptors. The work presented here shows the effectiveness of these gait descriptors and machine learning techniques to discriminate between healthy persons and patients with multiple sclerosis.

Full Text
See detailGAIMS: A Reliable Non-Intrusive Gait Measuring System
Pierard, Sébastien; Azrour, Samir; PHAN BA, Remy; Van Droogenbroeck, Marc

in ERCIM News (2013), 95

Gait observation and analysis can provide invaluable information about an individual [1]. Studies that have interpreted gait using traditional imaging devices have demonstrated that it is difficult to make reliable measurements with colour cameras. GAIMS, our new system resulting from a multidisciplinary project born from collaboration between engineers and neurologists, aims at developing non-intrusive and reliable tools to provide quantitative measures of gait and interpretations of the acquired data. Following a current trend in imaging, it takes advantage of imaging sensors that measure distance instead of colour. While its principles are general, GAIMS is currently used for the diagnosis of multiple sclerosis (MS) and the continued evaluation of disease progression [2]. It is the first available system to fully satisfy the clinical routine and its associated constraints.

Full Text
See detailGAIMS: a powerful gait analysis system satisfying the constraints of clinical routine
Pierard, Sébastien; PHAN BA, Remy; DELVAUX, Valérie; Maquet, Pierre; Van Droogenbroeck, Marc

in Multiple Sclerosis Journal (2013, October), 19(S1), 359

Full Text
See detailRelative Contribution of Walking Speed, Ataxia and Gait asymmetry to the Composition of Gait in Multiple Sclerosis
PHAN BA, Remy; Pierard, Sébastien; LOMMERS, Emilie; DELRUE, Gaël; CALAY, Philippe; Van Droogenbroeck, Marc; DELVAUX, Valérie; MAQUET, Pierre

in Multiple Sclerosis Journal (2013, October), 19(S1), 261

Introduction - Objective: Walking speed measured according to the T25FW is the most widely used descriptor of gait in MS clinical research and practice but other dimensions influencing gait variance exist according to alternative gait analysis methods. The relative importance of these different dimensions of gait relatively to its variance is unknown. Methods: We measured the performances of persons with MS and healthy subjects on the T25FW and the Timed 20-Meter Walk (T20MW) performed in tandem with a new gait analysis system (GAIMS). We performed a factorial analysis of variance to underline the main dimensions influencing gait variance and observed their composition. Findings - Conclusion: The main factor influencing gait variance in conventional walk tests is mostly composed of features related to walking speed. Balance, gait asymmetry and variability also participate to this variance but to a lesser extent. The inverse is observed in tests performed in tandem gait.

Full Text
See detailInfluence of the mode of walk on walking speed in multiple sclerosis: are you walking comfortably?
PHAN BA, Remy; DELRUE, Gaël; Pierard, Sébastien; LOMMERS, Emilie; CALAY, Philippe; Van Droogenbroeck, Marc; DELVAUX, Valérie

Poster (2013, June 10)

Introduction : Walking speed (WS) is the most frequent gait variable taken into account when measuring gait dysfunction in neurological diseases. Influences of the mode of walk instructed to the subject, i.e. « as fast as possible » (AFAP) or « at a comfortable pace » (PrP) have not been well characterized in multiple sclerosis (MS). Objectives : to compare those 2 mode of walk in a population of persons with MS (pMS) and healthy volunteers (HV). Methods: WS was measured with a new automated device along a 25 foot distance (T25FW) as part of a multimodal evaluation of gait in an MS ambulatory department. Results: Baseline demographics between HV and pMS were comparable. Our first results demonstrate that (i) WS is obviously significantly higher in AFAP than in PrP both for pMS and HV (p < 0.001 for all comparisons) and (ii) the relative difference between AFAP and PrP WS is significantly higher in HV than in pMS (p < 0.001). The AFAP-PrP WS correlation is higher in pMS (r = 0.87, p < 0.001) than in HV (r = 0.51, p < 0.001). Finally, the relative difference between AFAP and PrP WS is significantly and negatively correlated with the PrP WS in HV (r = -0.41, p < 0.001) and pMS with mild to moderate disability (EDSS 0-3.5, r = -0.49, p < 0.01) but not in pMS with high disability (EDSS 4-5.5, r = 0.008). Conclusions : these results suggests that heatlhy subjects have access to a higher range of PrP WS than pMS and questions the regulation of PrP WS that might be under psychological or behavioural influences. The demonstration of a lower PrP-AFAP difference in MS suggests that pMS are either adopting a natural WS closer to their maximum WS, or alternatively that they can’t reach their maximum WS because of neurological impairments. Our results also emphasize the importance of the instructed mode of walk in the quantification of gait disorders both for routine clinical practice and clinical trials.

Full Text
See detailEfficient database pruning for large-scale cover song recognition
Osmalsky, Julien; Pierard, Sébastien; Van Droogenbroeck, Marc; Embrechts, Jean-Jacques

in International Conference on Acoustics, Speech, and Signal Processing (ICASSP) (2013, May)

This paper focuses on cover song recognition over a large dataset, potentially containing millions of songs. At this time, the problem of cover song recognition is still challenging and only few methods have been proposed on large scale databases. We present an efficient method for quickly extracting a small subset from a large database in which a correspondence to an audio query should be found. We make use of fast rejectors based on independent audio features. Our method mixes independent rejectors together to build composite ones. We evaluate our system with the Million Song Dataset and we present composite rejectors offering a good trade-off between the percentage of pruning and the percentage of loss.

Full Text
See detailI-see-3D! An Interactive and Immersive System that dynamically adapts 2D projections to the location of a user’s eyes
Pierard, Sébastien; Pierlot, Vincent; Lejeune, Antoine; Van Droogenbroeck, Marc

in International Conference on 3D Imaging (IC3D) (2012, December)

This paper presents a system that gives the illusion of a 3D immersive and interactive environment with 2D projectors. The user does not need to wear glasses, nor to watch a (limited) screen. The virtual world is all around him, drawn on the floor. As the user is himself immersed in the virtual world, there is no need for a proxy like an avatar; he can move inside the virtual environment freely. Moreover, the Isee-3D system allows a user to manipulate virtual objects with his own body, making interactions with the virtual world very intuitive. Giving the illusion of 3D requires to render images insuch a way that the deformation of the image projected on thefloor is taken into account, as well as the position of the user’s “eye” in its virtual world. The resulting projection is neither perspective nor orthographic. Nevertheless, we describe how thiscan be implemented with the standard OpenGL pipeline, without any shader. Our experiments demonstrate that our system is effective in giving the illusion of 3D. Videos showing the results obtained with our I-see-3D system are available on our website http://www.ulg.ac.be/telecom/projector

Full Text
See detailDeciphering Distance-Induced Deceleration of Gait and Ataxia in People with Multiple Sclerosis
PHAN BA, Remy; Pierard, Sébastien; Moonen, Gustave; Van Droogenbroeck, Marc; Belachew, Shibeshih

in Multiple Sclerosis Journal (2012, October 12), 18(S4), 338

Full Text
See detailDetection and Quantification of Efficiency and Quality of Gait Impairment in Multiple Sclerosis through Foot Path Analysis
PHAN BA, Remy; Pierard, Sébastien; MOONEN, Gustave; Van Droogenbroeck, Marc; Belachew, Shibeshih

in Multiple Sclerosis Journal (2012, October 11), 18(S4), 110

Introduction: Walking speed is generally considered as the best outcome measure in trials for people with multiple sclerosis (pMS). We recently designed a device based on range laser scanner capable to track feet paths of walking subjects. Our purpose was to explore gait descriptors of pMS and compare them with those of healthy volunteers (HV). Methods: Fourty-four pMS (considered as moderatly or highly disabled according to a cut-off EDSS value of 3.0) and 28 HV performed 4 walking tasks along 2 trajectories in 3 walking modes. Twenty-six gait descriptors crudely dichotomized in « efficiency» and « quality » of gait were compared in the 2 populations using unpaired t-tests. Results: (i) apart from an older age in pMS, the two populations were comparable, (ii) efficiency of gait descriptors including walking speed distinguished HV from pMS, and pMS with moderate from pMS with high disability, (iii) quality of gait descriptors were also significantly altered in pMS, including in walking tasks where their walking speed was comparable to that of HV. Conclusions: RLS technology can distinguish pMS from HV according to (i) more efficiency of gait descriptors than the sole walking speed and (ii) quality of gait descriptors, including in subjects with a « normal » walking speed.

Full Text
See detailUtilisation de la Kinect
Lejeune, Antoine; Pierard, Sébastien; Van Droogenbroeck, Marc; Verly, Jacques

in Linux Magazine France (2012), 151

Fin 2010, Microsoft lançait la Kinect pour Xbox 360, la première caméra 3D destinée au grand public. Une semaine plus tard sortait la première librairie permettant d'utiliser l'appareil sur un ordinateur personnel. Depuis lors, des centaines d'applications ont vu le jour utilisant l'information de profondeur capturée par la Kinect pour analyser le mouvement humain ou guider des robots. Dans cette article, nous allons voir comment développer une application utilisant la Kinect sous GNU/Linux.

Full Text
See detailNeural networks for musical chords recognition
Osmalsky, Julien; Embrechts, Jean-Jacques; Van Droogenbroeck, Marc; Pierard, Sébastien

in Journées d'informatique musicale (2012, May)

In this paper, we consider the challenging problem of music recognition and present an effective machine learning based method using a feed-forward neural network for chord recognition. The method uses the known feature vector for automatic chord recognition called the Pitch Class Profile (PCP). Although the PCP vector only provides attributes corresponding to 12 semi-tone values, we show that it is adequate for chord recognition. Part of our work also relates to the design of a database of chords. Our database is primarily designed for chords typical of Western Europe music. In particular, we have built a large dataset filled with recorded guitar chords under different acquisition conditions (instruments, microphones, etc), but also with samples obtained with other instruments. Our experiments establish a twofold result: (1) the PCP is well suited for describing chords in a machine learning context, and (2) the algorithm is also capable to recognize chords played with other instruments, even unknown from the training phase.

Full Text
See detailRecognition of emotions in facial images
Pierard, Sébastien; Lejeune, Antoine; Van Droogenbroeck, Marc

Report (2012)

This document reports about some work done in the field of emotions classification in facial images.

Full Text
See detailOn the human pose recovery based on a single view
Pierard, Sébastien; Van Droogenbroeck, Marc

in International conference on pattern recognition applications and methods (ICPRAM) (2012, February)

Estimating the pose of the observed person is crucial for a large variety of applications including home entertainment, man-machine interaction, video surveillance, etc. Often, only a single side view is available, but authors claim that it is possible to derive the pose despite that humans evolve in a 3D environment. In addition, to decrease the sensitivity to color and texture, it is preferable to rely only on the silhouette to recover the pose. Under these conditions, we show that there is an intrinsic limitation: at least two poses correspond to the observed silhouette. We discuss this intrinsic limitation in details in this short paper. To our knowledge, this issue has been overlooked by authors in the past. We observe that this limitation has an impact on the way previous reported results should be interpreted, and it has clearly to be taken into account for designing new methods.

Full Text
See detailEstimation of human orientation based on silhouettes and machine learning principles
Pierard, Sébastien; Van Droogenbroeck, Marc

in International conference on pattern recognition applications and methods (ICPRAM) (2012, February)

Estimating the orientation of the observed person is a crucial task for home entertainment, man-machine interaction, intelligent vehicles, etc. This is possible but complex with a single camera because it only provides one side view. To decrease the sensitivity to color and texture, we use the silhouette to infer the orientation. Under these conditions, we show that the only intrinsic limitation is to confuse the orientation q with the supplementary angle (that is 180°-theta), and that the shape descriptor must distinguish between mirrored images. In this paper, the orientation estimation is expressed and solved in the terms of a regression problem and supervised learning. In our experiments, we have tested and compared 18 shape descriptors; the best one achieves a mean error of 5:24°. However, because of the intrinsic limitation mentioned above, the range of orientations is limited to 180°. Our method is easy to implement and outperforms existing techniques.

Full Text
See detailMultimodal evaluation of gait and stride dynamics in relapsing and progressive forms of multiple sclerosis
Belachew, Shibeshih; Pierard, Sébastien; PHAN BA, Remy; Van Droogenbroeck, Marc

in Proceedings of the Belgian Royal Academies of Medicine (2012), 1

Ambulation measures are being increasingly recognized as highly relevant to the quantification of multiple sclerosis (MS) severity and response to treatment. Feet paths are highly informative for gait analysis and we have recently designed a new system, which captures the position of the feet in real time. We use several range laser scanners (RLS) to analyze a horizontal slice of the scene in which each foot is considered as a point, and the vertical movements are ignored. Neat ambulation measures may be easily extracted such as walking speed, distance between feet over time, swing phase duration, and gait asymmetry in specific settings of walking recommendations. Our RLS platform is much cheaper than existing sensor-based and motion capture systems and may be more convenient for the development of multicentric clinical trials settings since patients can be easily and rapidly assessed without tags or sensors in the hallway of an outpatient clinic. We use 4 BEA LZR-i100 RLS arranged in a corridor of at least 10m long and 4m width, devoid of obstacle. The scanned plane is chosen to be located at 15 cm above the floor, which is right above the tibio-tarsal joint of the ankle in a barefoot configuration for adult individuals in stance phase. We expect further studies to validate and empower the meaning of non-intrusive RLS-derived gait measures that should pave the ground for major improvements in the way we will assess the efficacy of disease-modifying treatments (DMTs), physical therapy and symptomatic interventions on walking impairment, ataxia and fatigability in MS. RLS-derived gait measures may also reveal to be crucial in the near future for the development of treatments that would specifically target progressive forms of MS.

Full Text
See detailA new jump edge detection method for 3D cameras
Lejeune, Antoine; Pierard, Sébastien; Van Droogenbroeck, Marc; Verly, Jacques

in International Conference on 3D Imaging (IC3D) (2011, December)

Edges is a fundamental clue for analyzing, interpreting, and understanding 3D scenes: they describe objects boundaries. Available edge detection methods are not suited for 3D cameras such as the Microsoft Kinect or a time-of-flight camera: they are slow and do not take into consideration the characteristics of the cameras. In this paper, we present a fast jump edge detection technique for 3D cameras based on the principles of Canny’s edge detector. We first analyze the characteristics of the range signal for two different kinds of cameras: a time-of-flight camera (the PMD[vision] CamCube) and the Microsoft Kinect. From this analysis, we define appropriate operators and thresholds to perform the edge detection. Then, we present some results of the developed algorithms for both cameras.

Full Text
See detailA new low-cost and non-intrusive feet tracker
Pierard, Sébastien; PHAN BA, Remy; Van Droogenbroeck, Marc; Belachew, Shibeshih

in Workshop on Circuits, Systems and Signal Processing (ProRISC) (2011, November)

Capturing gait is useful for many applications, including video-surveillance and medical purposes. The most common sensors used to capture gait suffer from significant drawbacks. We have therefore designed a new low-cost and nonintrusive system to capture gait. Our system is able to track the feet on the horizontal plane in both the stance and the swing phases by combining measures of several range laser scanners. The number of sensors can be adjusted according to the target application specifications. The first issue addressed in this work is the calibration: we have to know the precise location of the sensors in a plane, and their orientations. The second issue addressed is how to calculate feet coordinates from the distance profiles given by the sensors. Our method has proven to be robust and precise to measure gait abnormalities in various medical conditions, especially neurological diseases (with a focus on multiple sclerosis).

Full Text
See detailEstimation of human orientation in images captured with a range camera
Pierard, Sébastien; Leroy, Damien; Hansen, Jean-Frédéric; Van Droogenbroeck, Marc

in Advanced Concepts for Intelligent Vision Systems (ACIVS) (2011, August)

Estimating the orientation of the observed person is a crucial task for some application fields like home entertainment, man-machine interaction, or intelligent vehicles. In this paper, we discuss the usefulness of conventional cameras for estimating the orientation, present some limitations, and show that 3D information improves the estimation performance. Technically, the orientation estimation is solved in the terms of a regression problem and supervised learning. This approach, combined to a slicing method of the 3D volume, provides mean errors as low as 9.2° or 4.3° depending on the set of considered poses. These results are consistent with those reported in the literature. However, our technique is faster and easier to implement than existing ones.

Full Text
See detailObject descriptors based on a list of rectangles: method and algorithm
Van Droogenbroeck, Marc; Pierard, Sébastien

in Soille (Ed.) International Symposium on Mathematical Morhology (ISMM) (2011, July)

Most morphological operators use a unique structuring element, possibly at different scales, to describe an object. In addition, morphological algorithms are often restricted to 1D structuring elements, combinations of 1D elements, or isotropic structuring elements (like circles), because of the lack of methods directly applicable to 2D structuring elements. While these descriptors have proved useful in the past, we propose an alternative that uses the list of maximal rectangles contained in a set X. In particular, we focus on an opening that preserves large rectangles contained in a set X and on its companion 2D algorithm that builds a list of all the maximal rectangles that fit inside an arbitrary set X. This list is the base of new descriptors that have been used successfully for machine learning tasks.

Full Text
See detailA probabilistic pixel-based approach to detect humans in video streams
Pierard, Sébastien; Lejeune, Antoine; Van Droogenbroeck, Marc

in International Conference on Acoustics, Speech, and Signal Processing (ICASSP 2011) (2011, May)

Human detection in video streams is an important task in many applications including video surveillance. Surprisingly, only few papers have been devoted to this topic. This paper presents a new approach to detect humans in video streams. Our approach is based on the temporal information present in videos. A background subtraction algorithm is first used to segment the silhouettes of the users and the moving objects. Then a classification process in two steps determines for each connected component if it corresponds to the silhouette of a human or not. During the first step, a probabilistic information is computed for each pixel independently. The information from a subset of pixels is then gathered to predict the class of the observed silhouette. This paper presents the principles and some results obtained on real silhouettes. It is shown that our approach is efficient for the detection of humans in video streams.

Full Text
See detailA virtual curtain for the detection of humans and access control
Barnich, Olivier; Pierard, Sébastien; Van Droogenbroeck, Marc

in Advanced Concepts for Intelligent Vision Systems (ACIVS) (2010, December)

Biometrics has become a popular field for the development of techniques that aim at recognizing humans based upon one or more intrinsic physical or behavioral traits. In particular, many solutions dedicated to access control integrate biometric features like fingerprinting or face recognition. This paper describes a new method designed to interpret what happens when crossing an invisible vertical plane, called virtual curtain hereafter, at the footstep of a door frame. It relies on the use of two laser scanners located in the upper corners of the frame, and on the classification of the time series of the information provided by the scanners after registration. The technique is trained and tested on a set of sequences representative for multiple scenarios of normal crossings by a single person and for tentatives to fool the system. We present the details of the technique and discuss classification results. It appears that the technique is capable to recognize many scenarios which may lead to the development of new commercial applications.

Full Text
See detail3D information is valuable for the detection of humans in video streams
Pierard, Sébastien; Lejeune, Antoine; Van Droogenbroeck, Marc

in 3D Stereo Media (2010, December)

In this paper, we propose a technique based on 3D information (also called depth or range) for the detection of humans. First, a background subtraction technique operates to detect the silhouettes of humans and objects moving in the scene. Then, a machine learning algorithm is used to predict if a silhouette annotated with depth matches a human silhouette or not. The complete method is designed to cope with defects introduced during the segmentation step. Results, obtained on computer generated data, show that 3D depth data is a valuable information for detecting humans in that it improves over techniques based on binary silhouettes. In our experiments, we have reached an accuracy of 99.9% thanks to the depth information.

Full Text
See detailA platform for the fast interpretation of movements and localization of users in 3D applications driven by a range camera
Pierard, Sébastien; Pierlot, Vincent; Barnich, Olivier; Van Droogenbroeck, Marc; Verly, Jacques

in Proceedings of the 3DTV Conference (2010, June)

Interactivity is one of the key challenges for immersive applications like gaming. Manufacturers have been working towards interfaces that are driven by a device (e.g. a Wiimote) or interfaces that are controlled by a camera with a subsequent computer vision module. Both approaches have unique advantages, but they do not permit to localize users in the scene with an appropriate accuracy. Therefore, we propose to use both a range camera and accurate range sensors to enable the interpretation of movements. This paper describes a platform that uses a range camera to acquire the silhouettes of users, regardless of illumination, and to improve the pose recovery with range information after some image processing steps. In addition, to circumvent the difficult process of calibration required to map range values to physical distances, we complete the system with several range laser sensors. These sensors are located in a horizontal plane, and measure distances up to a few centimeters. We combine all these measurements to obtain a localization map, used to locate users in the scene at a negligible computational cost. Our method fills a gap in 3D applications that requires absolute positions.

Full Text
See detailTechniques to improve the foreground segmentation with a 3D camera and a color camera
Pierard, Sébastien; Van Droogenbroeck, Marc

(2009, November)

Nowadays, techniques for real-time interpretation of video scenes are widespread. Amongst these techniques, the foreground segmentation is one of the favorite. It can be applied to color images as well as depth maps. The point of using depth maps is straightforward as a single color camera is not able to provide depth information. Technologies capable to acquire 3D informations are thus adequate to complement color cameras in consumer products. Practice has shown that 3D or RGB signals, taken alone, are unreliable to extract the foreground under arbitrary conditions. Therefore we combine both modalities to counter the intrinsic limitations of both modalities, which is only possible if the problems specific to a technology are handled appropriately. This paper presents a new global approach for enhanced foreground segmentation that handles limitations to 3D and RGB in a combined way.

Full Text
See detailA technique for building databases of annotated and realistic human silhouettes based on an avatar
Pierard, Sébastien; Van Droogenbroeck, Marc

in Annual Workshop on Circuits, Systems and Signal Processing (ProRISC) (2009, November)

The real-time interpretation of video scenes and human motion is a research field with a lot of applications. Awidely used pre-processing step is the background subtraction, which highlights silhouettes. In addition, example-based and learning-based approaches are attractive for the interpretation of complex scenes. But to be effective, such techniques need to be trained on samples of realistic human silhouettes. This paper presents a method to automatically build databases of artificial, realistic, and annotated human silhouettes. Our method, based on a 3D human avatar, allows to save additional data with the silhouettes, and to highlight the different body parts. These supplementary data may be application-driven. To help in evaluating the relevance of automatically generated databases, we suggest some practical solutions based on a visual signature of the databases.

Full Text
See detailReal-time processing of depth and color video streams to improve the reliability of depth maps
Pierard, Sébastien; Leens, Jérôme; Van Droogenbroeck, Marc

Poster (2009, November)

Depth is a useful information in vision to understand the geometrical properties of an environment. Depth is traditionally computed in terms of a disparity map acquired by a stereoscopic system but, over the last few years, several manufacturers have released single-lens cameras that directly capture depth information (also called range). This is an important technological breakthrough although range signals remain difficult to handle in practice, due to many reasons (low resolution, noise, low framerate, . . . ). Practitioners still struggle to use range data in their applications. The purpose of this paper is to give a brief introduction to range data (captured with a camera), discuss common limitations, and propose techniques to cope with difficulties typically encountered with range cameras. These techniques are based on a simultaneous view of the scene by a color and a depth camera that are combined to improve their interpretation in real time.

Full Text
See detailCombining Color, Depth, and Motion for Video Segmentation
Leens, Jérôme; Pierard, Sébastien; Barnich, Olivier; Van Droogenbroeck, Marc; Wagner, Jean-Marc

in Computer Vision Systems (2009)

This paper presents an innovative method to interpret the content of a video scene using a depth camera. Cameras that provide distance instead of color information are part of a promising young technology but they come with many diff culties: noisy signals, small resolution, and ambiguities, to cite a few. By taking advantage of the robustness to noise of a recent background subtraction algorithm, our method is able to extract useful information from the depth signals. We further enhance the robustness of the algorithm by combining this information with that of an RGB camera. In our experiments, we demonstrate this increased robustness and conclude by showing a practical example of an immersive application taking advantage of our algorithm.

Full Text
See detailGénération automatique d'une base de données de silhouettes humaines au moyen d'un avatar humain tridimensionnel
Pierard, Sébastien

Master's dissertation (2007)

Pour la conception de systèmes de vidéo-surveillance, il est avantageux d’avoir un système de détection automatique d’êtres humains. Certains de ces systèmes sont basés sur la reconnaissance de la forme de silhouettes humaines. L’utilisation de méthodes d'apprentissage pour la reconnaissance nécessite une base de données de silhouettes humaines, dont il est intéressant d’automatiser la génération. Dans ce travail, nous utilisons un avatar humain tridimensionnel, celui du projet open-source MakeHuman, afin de générer automatiquement la base de données. Le choix des configurations (positions, orientations et poses) à donner à l’avatar pour obtenir une base de données adéquate est central. Dans un premier temps, nous tentons une étude de l’impact de modifications dans le mode tridimensionnel sur la silhouette. Cet impact est quantifié par la distance de Hausdorff. Parmi les modifications envisagées se trouvent le déplacement, la rotation et le changement de pose de l’avatar. Nous espérions déterminer, à partir de ces impacts, les pas par lesquels doivent varier les différents paramètres. Les configurations que doit prendre l’avatar, pour obtenir une base de données suffisamment riche sans être trop volumineuse, seront obtenues en prenant toutes les combinaisons des valeurs des différents paramètres. Le nombre de configurations est exponentiel en le nombre de degrés de liberté, ce qui rend l’approche inutilisable en pratique, même si nous sommes indifférents à la translation et à la mise à l’échelle des silhouettes. Dans un deuxième temps, nous avons opté pour une approche radicalement différente. Cette fois, nous n’espaçons plus les silhouettes, mais les poses. La position de l’avatar est fixée et il ne peut tourner qu’autour de l’axe vertical. Les poses sont choisies en effectuant des tirages aléatoires des différents paramètres. Cette approche aléatoire du problème a mené à l’élaboration d’un nouveau logiciel, Silhouette. Il est écrit en Java et une interface écrite en C lui permet, grâce à Java Native Interface d’exploiter efficacement le moteur de pose de MakeHuman. Parmi les fonctionnalités de Silhouette, notons la possibilité de restreindre les poses à un ensemble voulu, de ne générer que des silhouettes correspondant à des poses licites et d’annoter les silhouettes. Grâce au logiciel mis en place, nous avons été en mesure de générer diverses bases de données. Nous avons étudié l'impact de l'utilisation de celles-ci sur les performances d'un système de détection de personnes existant. Nous obtenons des résultats encourageants, 73% des silhouettes humaines étant correctement détectées. Un raffinement et un paramétrage soigneux de la méthode devraient mener à des résultats excellents.