La thèse porte sur la détection d’objets à partir d’une caméra embarquée sur un véhicule mobile en exploitant l’approche monoculaire « c-vélocité ». Cette méthode s’inspire de la méthode appelée « v-disparité » utilisée en stéréovision : toutes deux ont pour objectif la détection d’objets en les approximant par des plans d’orientations différentes, ce qui permet d’éviter, en monoculaire, d’estimer la profondeur. Ces deux approches, monoculaires et binoculaires, permettent de transformer le problème complexe de la détection d’objets en un problème plus simple de détection de formes paramétriques simples (droites, paraboles) dans un nouvel espace de représentation où la détection peut être réalisée à l’aide d’une transformée de Hough. La « c-vélocité », pour être efficace, requiert un calcul assez précis du flot optique et une bonne estimation de la position du Foyer d’expansion (FOE). Dans cette thèse, nous avons étudié les approches existantes de calcul de flot optique et sommes arrivés à la conclusion qu’aucune n’est vraiment performante notamment sur les régions homogènes telle que la route dans les scènes qui correspondent à l’application que nous visons à savoir : les véhicules intelligents. Par ailleurs, les méthodes d’estimation du flot optique peinent également à fournir une bonne estimation dans le cas de déplacement importants dans les régions proches de la caméra. Nous proposons dans cette thèse d’exploiter à la fois un modèle 3D de la scène et une estimation approximative de la vitesse du véhicule à partir d’autres capteurs intégrés. L’utilisation de connaissances a priori permet de compenser le flot dominant pour faciliter l’estimation de la partie résiduelle par une approche classique. Par ailleurs, trois approches différentes sont proposées pour détecter le foyer d’expansion. Parmi elles, nous proposons une méthode novatrice permettant d’estimer le FOE en exploitant la norme du flot et la structure de la scène à partir d’un processus « c-vélocité » inversé. En plus d’améliorer ces étapes préliminaires, nous proposons aussi l’optimisation et l’accélération de l’algorithme « c-vélocité » par une implémentation multithread. Enfin, nous proposons une modification de l’approche c-vélocité d’origine afin d’anticiper une éventuelle coopération mouvement/stéréo, proposée en perspective, à travers un jumelage avec la v-disparité. / This thesis is based on the detection of objects from an onboard moving camera by exploiting the monocular approach "c-velocity". This method is inspired by the method called "v-disparity" used in stereovision: both methods aim at detecting objects by approximating objects into plans with different orientations. Such approximation can avoid to estimate the depth in monocularvision. These two approaches, monocular and binocular, allow to transform the complex objet détection problem into a more simple parametric forms (eg. lines) detection in a new space, where these formes can be easily extracted using Hough Transform.The “c-velocity”, to make it effective, requires an accurate computation of optical flow and a good estimation of the focus of expansion (FOE) location. Therefore, we have studied the existing approaches of optical flow estimation and arrived at the conclusion that none of them is really powerful especially on the homogeneous regions such as road surface. In addition, the optical flow estimation methods also struggle to provide a good estimate in the case of huge displacement in the areas close to the camera. We propose in this thesis to exploit both a 3D model of the scene and a rough estimate about the vehicle speed from other integrated sensors. Using a priori knowledge allows to compensate the dominant optical flow and to facilitate the estimation of the rest part by a classical approach. In addition, three different approaches are proposed to detect the focus of expansion. Among them, we propose a novel method for estimating FOE by leveraging the flow norm and the scene structure from an inverse “c-velocity“ process. In addition to improve these preliminary steps, we also propose an acceleration and optimization of the “c-velocity“ algorithm by a multi-thread implementation. Finally, we propose a modification to the original “c-velocity“ approach in order to anticipate a possible cooperation motion/stereo, proposed in perspective, with the “v-disparity“ approach.

The objective of this dissertation was to examine the action profiles of elite field hockey players in relationship to the 2015 FIH rule change. The following are major findings of the dissertation: The following are major findings of the dissertation: Study 1 – Relative action profiles before the rule change revealed that defenders work at a lower meter per minute (m/min) when compared with all other positions, and that forwards, midfielders, and screens perform similar m/min during a game. Examination of pre rule change difference from the 1st to the 2nd half play showed that elite level field hockey players are able maintain high-intensity actions in zone 6 throughout the game by increasing actions in zones 1 and 2, and decreasing actions in zones 4 and 5. Study 2 – Action profiles after the rule formatting change revealed the team was unable to match the percent of distance covered in zones 4 and 5 during the 1st quarter all in subsequent quarters. The low intensity actions in zone 1 and 2 gradually increased, while m/min gradually declined. However the percent of distance covered in zone 6 showed no statistically significant change. When positional differences were examined forwards covered the greatest percent of distance in zones 5 and 6, followed by midfielders, screens, and defenders. This pattern varies for zone 4, within which the midfielders possesses the greatest percent distance covered. 3 Study 3 – Relative action profile comparisons for the team, pre to post the 2015 rule change did not indicate a significant change in zones 2, 3, 4, 5, and 6. However zone 1 experience a statistically significant decrease. Positional analysis showed statistically significant changes for midfielders only. The changes were a decrease in zone 1, and increase in zone 5 and 6 during the first half of the game, and decrease in zone 1 and m/min during the second half of the game. A major focus of the US Women’s National Team is to develop the athletes’ physical capacity to maintain and repeat high intensity actions. The combination of physical preparation and tactical strategies allow the team to express high m/min and numerous high intensity actions throughout a match.

Visual analysis of human motion is one of the active research areas in computer vision. The trend shifts from computing motion fields to understanding actions. In this thesis, an action coding scheme based on trajectories of the features calculated with respect to a part based coordinate system is presented. The part based coordinate system is formed using an axis based representation. The features are extracted from images segmented in the form of silhouettes. We present some preliminary experiments that demonstrate the potential of the method in action similarity analysis.

Parent-child social games, such as peek-a-boo and patty-cake, are a key element of an infant's earliest social interactions. The analysis of children's behaviors in social games based on video recordings provides a means for psychologists to study their social and cognitive development. However, the current practice in the use of video for behavioral research is extremely labor-intensive, involving many hours spent extracting and coding relevant video clips from a large corpus. From the standpoint of computer vision, such real-world video collections pose significant challenges in the automatic analysis of behavior, such as cluttered backgrounds, the effect of varying camera angles, clothing, subject appearance and lighting. These observations motivate my thesis work - automatic retrieval of social games from unstructured videos. The goal of this work is both to help accelerate the research progress in behavioral science and to take the initial steps towards the analysis of natural human interactions in natural settings. Social games are characterized by repetitions of turn-taking interactions between the parent and the child, with variations that are recognizable by both of them. I developed a computational model for social games that exploits the temporal structure over a long time-scale window as quasi-periodic patterns in a time series. I presented an unsupervised algorithm that mines the quasi-periodic patterns from videos. The algorithm consists of two functional modules: converting image sequences into discrete symbolic sequences and mining quasi-periodic patterns from the symbolic sequences. When this technique is applied to video of social games, the extracted quasi-periodic patterns often correspond to meaningful stages of the games. The retrieval performance on unstructured, lab-recorded videos and real-world family movies is promising. Building on this work, I developed a new feature extraction algorithm for social game categorization. Given a quasi-periodic pattern representation, my method automatically selects the most relevant space-time interest points to construct the feature representation. Our experiments demonstrate very promising classification performance on social games collected from YouTube. In addition, the method can also be used to categorize TV videos of sports rallies, demonstrating the generality of this approach. In order to support and encourage more research on human behavior analysis in realistic contexts, a video database of realistic child play in natural settings has been collected and is published on our project website (http://www.cc.gatech.edu/cpl/projects/socialgames), along with annotations. The unsupervised quasi-periodic pattern mining method represents a substantial generalization of conventional periodic motion analysis. Its generality is evaluated by retrieving motions of a range of quasi-periodicity from unstructured videos. The performance was compared with that of a periodic motion detection method based on motion self-similarity. Our method demonstrates superior retrieval performance with a 100% precision when the recall is up to 92.04%, with much fewer parameters than that of the other method.

This work focuses on the use of 3D motion capture data to create and optimize a robotic human body model (RHBM) to predict the inverse kinematics of the upper body. The RHBM is a 25 degrees of freedom (DoFs) upper body model with subject specific kinematic parameters. The model was developed to predict the inverse kinematics of the upper body in the simulation of a virtual person, including persons with functional limitations such as a transradial or transhumeral amputation. Motion data were collected from 14 subjects: 10 non-amputees control subjects, 1 person with a transradial amputation, and 3 persons with a transhumeral amputation, in the University of South Florida's (USF) motion analysis laboratory. Motion capture for each subject consisted of the repetition of a series of range of motion (RoM) tasks and activities of daily living (ADLs), which were recorded using an eight camera Vicon (Oxford, UK) motion analysis system. The control subjects were also asked to repeat the motions while wearing a brace on their dominant arm. The RoM tasks consisted of elbow flexion & extension, forearm pronation & supination, shoulder flexion & extension, shoulder abduction & adduction, shoulder rotation, torso flexion & extension, torso lateral flexion, and torso rotation. The ADLs evaluated were brushing one's hair, drinking from a cup, eating with a knife and fork, lifting a laundry basket, and opening a door. The impact of bracing and prosthetic devices on the subjects' RoM, and their motion during ADLs was analyzed. The segment geometries of the subjects' upper body were extracted directly from the motion analysis data using a functional joint center method. With this method there are no conventional or segment length differences between recorded data segments and the RHBM. This ensures the accuracy of the RHBM when reconstructing a recorded task, as the model has the same geometry as the recorded data. A detailed investigation of the weighted least norm, probability density gradient projection method, artificial neural networks was performed to optimize the redundancy RHBM inverse kinematics. The selected control algorithm consisted of a combination of the weighted least norm method and the gradient projection of the null space, minimizing the inverse of the probability density function. This method increases the accuracy of the RHBM while being suitable for a wide range of tasks and observing the required subject constraint inputs.

[Truncated abstract] Normal and pathological upper limb movement assessments rely on the valid and reliable identification of the glenohumeral joint centre of rotation (GHJ). However, clarifying the most suitable techniques to identify and reference this location has proved a challenge, and performing a variety of methods that lack validation is commonplace. This may not only be erroneous, but also prevents the standardised collection of upper limb biomechanical information. The principle aim of this research was to clarify the accuracy and reliability of various methods of GHJ identification, including both predictive and functional techniques, as well as the error associated with referencing the GHJ location during dynamic movement trials. Predictive methods of GHJ identification rely on a generic relationship between the GHJ position and predetermined anatomical distances or locations. The ISB recommended predictive method was developed and validated using cadavers, and it appears that a number of convenient, yet to be validated methods are routinely performed in preference of this recommended technique. In the present study, magnetic resonance imaging (MRI) was utilised to validate, in vivo, the accuracy of various predictive approaches; the ISB recommended method and a representative sample of commonly used techniques. A new multiple linear regression model and simple 3D offset method, were developed from the MRI identified locations of the GHJ and the surface markers. The results indicated that the new multiple linear regression model (13 ±4.6) mm and simple 3D offset (12 ±4.6 mm) found an average GHJ location closer to the MRI determined location than any of the established predictive methods (14-50 mm), including the ISB recommended method (32 ±8.2 mm), and a recently publicised amended 2nd version (16 ±8.4 mm). ... For instance when the optimal algorithm (geometric sphere fit), marker set and movement trial were used in the functional approach, average in vivo accuracy errors of 27 ±8.6 mm were reported, around half the error reported by the most accurate and reliable predictive method (13 ±4.6 mm). A further investigation aimed to determine the most suitable location to reference the GHJ during dynamic motion analysis trials. The GHJ was referenced in a number of upper arm and acromion technical coordinate systems (TCSs) in a series of static MRIs. This permitted the error associated with each set of markers to be calculated in vivo. The results indicated that a combination of TCSs defined from two sets of markers; one placed on the acromial plateau and one located proximally on the upper arm, produced the most accurate results, recording an average of 18 ±4 mm of error following a large humeral elevation (up to 180°). Furthermore, a distal upper arm set of markers proved to be inappropriate for GHJ referencing, reporting average errors greater than 30 cm in two large humeral elevations. Therefore, following the identification of the GHJ, its 3D location should be referenced in the average of two TCSs determined from respective sets of markers placed on the acromion and proximal upper arm, during dynamic trials.

[Truncated abstract] This thesis consists of five experimental studies from seven data collection periods. The first two studies quantitatively analyse children with and without cerebral palsy using upper limb three dimensional (3D) motion analysis. Upper limb angular kinematics and sub-structures were measured and analysed, both of which were utilised during subsequent studies. The final three studies assess the efficacy of lycra® arm splints using clinical assessments, 3D dimensional upper limb kinematics and 3D sub-structures. Study 1 analysed 3D movement sub-structures in children with and without cerebral palsy ... The aim of the study was to quantitatively analyse movement sub-structures in children with and without cerebral palsy during four functional tasks taken from the Melbourne Assessment of Unilateral Upper Limb Function (Melbourne Assessment - Randall, Johnson & Reddihough, 1999) ... Results demonstrated significant differences in angular kinematics in children with and without cerebral palsy, while the methodology developed in this study provided improved insight into the movement of the upper limb and trunk during functional tasks. Study 3 reported a randomised controlled trial of lycra® arm splints in children with cerebral palsy across all levels of the International Classification of Functioning Disability and Health (ICF) ... Lycra® arm splints were shown to have a statistically significant impact at the level of participation, whereas no significant difference was seen at the level of impairment and activity. Study 4 reported a randomised controlled trial of the effects of lycra® arm splints on 3D movement sub-structures during functional tasks in children with cerebral palsy ... This research demonstrated that movement sub-structures (including movement time) can be quantified and are amenable to change with intervention. Study 5 reported a randomised controlled trial of the effects of lycra® arm splints on angular kinematics (thorax, shoulder and elbow) during functional tasks in children with cerebral palsy ... The benefits of the splint on angular kinematics were only apparent when worn for the 3 month period, as minimum evidence was established for the short-term (1hour) and long term (3 month post splint wear) carry-over effects.

This thesis comprises one review paper and five experimental studies, all of which are presented in the form of journal article submissions. These six research papers attempt to further our understanding of the physiological and metabolic requirements of repeated-sprint activity, specific to field-based team sports. Although coaches and sport scientists have suggested that repeated-sprint ability is an important fitness component of team sports, this area of investigation has only become more common in the past 10 years.

Considerando que a duração do combate de judô pode variar entre poucos segundos até um período superior a 5 min, a caracterização da demanda fisiológica do combate na sua duração regulamentar e fracionada, assim como a relação dessas respostas com o nível de aptidão aeróbia e anaeróbia dos atletas é relevante para o aperfeiçoamento dos métodos de treinamento. Assim, o objetivo do presente estudo foi analisar a aptidão física, ações técnicas e respostas fisiológicas durante a luta de judô com diferentes durações. Para tanto, 12 atletas de judô foram submetidos a 10 sessões de avaliação. Nas duas primeiras sessões foram realizados quatro testes físicos para caracterização da aptidão aeróbia e anaeróbia dos membros superiores e inferiores. Nas demais sessões, os atletas foram submetidos a 15 combates simulados de judô (3 condições x 5 durações - 1, 2, 3, 4 e 5 min) com o mesmo oponente, em ordem aleatória e vendados quanto à duração do combate. Em uma das condições, os atletas simularam um combate real. Na segunda condição, os atletas lutaram usando um analisador de gases para mensurar o consumo de oxigênio, para a estimativa da contribuição dos sistemas de transferência de energia. Nessa condição foram impostas algumas adaptações de movimento. Para verificar se o uso do analisador de gases modificou as respostas obtidas, na terceira condição, os atletas lutaram com as mesmas adaptações, porém não usaram o equipamento. Em cada sessão foram conduzidos dois combates, intercalados por 60 min de recuperação. Foram quantificadas as respostas fisiológicas, perceptivas, de desempenho, de estrutura temporal e ações técnicas em todos combates. A comparação das variáveis dependentes foi realizada com uma análise de modelos mistos para medidas repetidas, seguida do post hoc de Bonferroni. As associações entre a aptidão aeróbia e anaeróbia com as variáveis do combate foram verificadas por meio da correlação de Pearson. Para todas as comparações pareadas significantes foi calculado o tamanho do efeito (d de Cohen). Em todas as análises foi adotado 5% como nível de significância. Os principais resultados demonstraram aumento da contribuição aeróbia com concomitante redução da contribuição dos sistemas anaeróbios com o transcorrer do combate, sendo a contribuição aeróbia superior à contribuição dos sistemas anaeróbios a partir do primeiro min de combate. O sistema anaeróbio lático contribuiu com a menor proporção do custo energético do combate de judô. As taxas de incremento das respostas fisiológicas e perceptivas por min de combate foram reduzidas com o transcorrer do combate, sendo as principais modificações observadas no início do combate, sugerindo que os atletas podem apresentar estratégia para modular seus esforços. Com o transcorrer do combate observou-se manutenção do tempo médio de esforço e ações aplicadas, porém houve aumento do tempo médio dos períodos de pausa. Após a realização do combate observou-se declínio da resistência muscular dinâmica dos membros superiores, manutenção da força isométrica máxima de preensão manual e melhora da potência muscular dos membros inferiores. As variáveis de aptidão aeróbia e anaeróbia relacionaram-se com menor queda do desempenho, maior taxa de golpes de perna e maior tempo médio de combate. Assim, observou-se predominância do sistema aeróbio para suprir a energia necessária para a realização do combate de judô. Além disso, maiores níveis de aptidão aeróbia e anaeróbia estão associados positivamente com o desempenho durante a luta / Considering that judo combat can last from a few seconds to up to more than 5 minutes, characterization of the physiological demands of combat in different time frames, and the relationship of these responses to an athlete\'s level of aerobic and anaerobic fitness is relevant to improving training methods. Thus, the objective of the present study was to describe the physiological demands of judo combat. Twelve judo athletes performed 10 experimental sessions. In the first and second sessions, four physical tests to characterize the aerobic and anaerobic fitness of the upper and lower limbs were performed. In the other sessions, the athletes performed 15 judo combat (involving 3 conditions x 5 durations of time: 1, 2, 3, 4 and 5 minutes) randomly determined, against the same opponent, on different days and blinded for time duration. In one condition, the athletes simulated actual combat. In the second condition, the athletes fought while using a gas analyzer to measure oxygen consumption in order to estimate the contribution of the energy systems. In this condition some motion adjustments were imposed. To verify if the use of the gas analyzer modified the responses obtained, in the third condition the athletes fought with the same adjustments, but did not use the equipment. These combat periods were divided into 8 sessions with two different conditions, separated by an interval of 60 minutes. The physiological, perceptual, performance, time-motion, and technical action responses were quantified in all matches. The comparison of the dependent variables was performed with an analysis of mixed models for repeated measures, followed by a post hoc Bonferroni. The associations between aerobic and anaerobic fitness to combat variables were verified using Pearson\'s correlation. For all significant pairwise comparisons the effect size (d Cohen) was calculated. A 5% level of significance was adopted for all analyses. The main results showed an increase in aerobic contribution with a concomitant reduction of the anaerobic contribution in the course of combat, given that the aerobic contribution was greater than the anaerobic contribution from the first minute of combat. The lactic anaerobic system contributed to the smaller proportion of energy cost in judo combat. The increment in rate of the physiological and perceptual responses per minute of combat decreased throughout the course of the combat, with the highest changes occurring at the beginning of combat, which suggest that judo athletes seem to modulate their effort. In the course of combat, maintenance of the average time of effort and actions occurred, but there was an increase in the average time of pause. After the combat, a decline in the dynamic strength endurance of the upper limbs was observed, while maximal isometric handgrip strength was maintained, and improvement of muscle power in the lower limbs was noted. The aerobic and anaerobic fitness variables were associated with lower fall performance, higher rate of leg attacks and a higher average time of standing combat. Thus, a predominance of the aerobic system in supplying the energy needed for the realization of judo combat was observed. In addition, higher levels of aerobic and anaerobic fitness are positively associated with performance during the combat

Ces travaux avaient pour objectif la preuve de concept et le développement d’un dispositif embarqué sur un T-shirt visant à terme à se substituer au corset afin de pouvoir améliorer la prise en charge des patients scoliotiques. Il s’agissait d’orienter les soins de la scoliose vers un traitement permettant au patient de conserver l’ensemble de sa mobilité et de rééquilibrer l’activité des muscles de son tronc. Tout d’abord, une étude clinique comparative entre sujets sains, patients scoliotiques lombaires et patients scoliotiques thoraciques a été menée afin de mettre en évidence des différences cinématiques et posturales potentiellement corrigeables, à travers l’étude de mouvements simples standardisés. Ensuite, le dispositif baptisé « kinésithérapeute embarqué » a été développé. Son but est de détecter des anomalies cinématiques ou de mauvaises postures chez le patient afin de les lui signaler pour lui permettre une autocorrection de ses défauts. Le T-shirt permet alors une mesure précise des mouvements de l’utilisateur en temps-réel et sans dispositif externe, à l’aide de centrales inertielles. Deux feedbacks ont été conçus répondant à deux utilisations différentes, l’un vibro-tactile pour une correction posturale en ambulatoire au cours de la journée, et l’autre visuel permettant d’apprendre et d’entraîner des mouvements spécifiques lors de séances de kinésithérapie à domicile. / The purpose of this thesis was the proof of concept and the development of an embedded device on a T-shirt aiming in the future to replace the brace for improving the medical care of scoliotic patients. The aim was to orient scoliosis care toward a treatment allowing a patient to keep his/her full mobility and rebalancing his trunk muscles. First, a comparative clinical study between healthy subjects, lumbar scoliotic patients and thoracic scoliotic patients was conducted in order to highlight kinematic and postural differences potentially correctable through the study of simple and standardized movements. Then, the so-called “embedded physiotherapist” device has been developed. Its aim is to detect patient’s kinematic singularities or bad postures and to alert him/her for an auto-correction of the defect. Thus, the T-shirt is able to accurately measure user’s motion in real-time and without any external device, using inertial measurement units. Two feedback have been designed for satisfying two different purposes, first a vibro-tactile one for postural corrections during the day, and a visual one allowing the user to learn and train specific motions during physiotherapy sessions at home.