Human Motion

While research on articulated human motion and pose estimation has progressed rapidly in the last few years, there has been no systematic quantitative evaluation of competing methods to establish the current state of the art. Current algorithms make many different choices about how to model the human body, how to exploit image evidence and how to approach the inference problem. We argue that there is a need for common datasets that allow fair comparison between different methods and their design choices. Until recently gathering ground-truth data for evaluation of results (especially in 3D) was challenging. In this report we present a novel dataset obtained using a unique setup for capturing synchronized video and ground-truth 3D motion. Data was captured simultaneously using a calibrated marker-based motion capture system and multiple high-speed video capture systems. The video and motion capture streams were synchronized in software using a direct optimization method. The resulting HumanEva-I dataset contains multiple subjects performing a set of predefined actions with a number of repetitions. On the order of 50,000 frames of synchronized motion capture and video was collected at 60 Hz with an additional 37,000 frames of pure motion capture data. The data is partitioned into training, validation, and testing sub-sets. A standard set of error metrics is defined that can be used for evaluation of both 2D and 3D pose estimation and tracking algorithms. Support software and an on-line evaluation system for quantifying results using the test data is being made available to the community. This report provides an overview of the dataset and evaluation metrics and provides pointers into the dataset for additional details. It is our hope that HumanEva-I will become a standard dataset for the evaluation of articulated human motion and pose estimation.

see: https://epubs.scu.edu.au/theses/658/  >>>>>>  Harvesting electricity from human motions using piezoelectric energy harvesters is attracting the attention of many researchers in recent years. The energy harvested can potentially power portable electronic devices such as mobile phones as well as some medical devices like
pacemakers without the need of external power source or battery.
The frequency of human movements is low and thus the design of energy harvesting systems needs to be improved to harvest high power. The overarching aim of this thesis was to improve the efficiency of piezoelectric energy harvesting from human motions. To achieve this,
optimising orientation of conventional piezoelectric cantilever beam investigated; the new mechanism consisting of double pendulum system was studied to make several impacts in each walking step and to reduce the idle time among voltage peaks and finally the new shape design of cantilever was proposed to generate multi resonance peaks in the low frequency range. The outcome of this research showed that (1) maximum power (2779 micro Watts) was achieved when the PEH was orientated at 70º with reference to a coordinate system attached to the leg; (2) using double pendulum system, a dramatic increase in maximum output voltage and power (2100 micro Watts) was achieved with the proposed system on both the well-controlled mechanical shaker test and practical tests on human arm and leg motion in walking and jogging
modes, in comparison to the conventional harvester and a similar system with only one pendulum and (3) the proposed shape design of cantilever with two branched triangular beams was able to produce several resonance peaks under 20 Hz which can increase the output power.
These achievements may help to improve the efficiency of piezoelectric energy harvesting systems in the future.

Autonomous video surveillance and monitoring has a rich history. Many deployed systems are able to reliably track human motion in indoor and controlled outdoor environments. However, object detection and tracking at night remain very important problems for visual surveillance. The objects are often distant, small and their signatures have low contrast against the background. Traditional methods based on the analysis of the difference between successive frames and a background frame will do not work. In this paper, a novel real time object detection algorithm is proposed for night-time visual surveillance. The algorithm is based on contrast analysis. In the first stage, the contrast in local change over time is used to detect potential moving objects. Then motion prediction and spatial nearest neighbor data association are used to suppress false alarms. Experiments on real scenes show that the algorithm is effective for night-time object detection and tracking. ᭧

This survey reviews advances in human motion capture and analysis from 2000 to 2006, following a previous survey of papers up to 2000 [T.B. Moeslund, E. Granum, A survey of computer vision-based human motion capture, Computer Vision and Image Understanding, 81(3) (2001) 231-268.

This paper deals with the imitation of human motions by a humanoid robot based on marker point measurements from a 3D motion capture system. For imitating the human's motion, we propose a Cartesian control approach in which a set of control points on the humanoid is selected and the robot is virtually connected to the measured marker points via translational springs. The forces according to these springs drive a simplified simulation of the robot dynamics, such that the real robot motion can finally be generated based on joint position controllers effectively managing joint friction and other uncertain dynamics. This procedure allows to make the robot follow the marker points without the need of explicitly computing inverse kinematics. For the implementation of the marker control on a humanoid robot, we combine it with a center of gravity based balancing controller for the lower body joints. We integrate the marker control based motion imitation with the mimesis model, which is a mathematical model for motion learning, recognition, and generation based on hidden Markov models (HMMs). Learning, recognition, and generation of motion primitives are all performed in marker coordinates paving the way for extending these concepts to task space problems and object manipulation. Finally, an experimental evaluation of the presented concepts using a 38 degrees of freedom humanoid robot is discussed.

Predictive dynamics is a novel approach for simulating human motion. It avoids direct integration of differential-algebraic equations in order to create the resulting simulations for redundant digital human models. Instead, it formulates an optimization problem by defining appropriate performance measures and constraints to recover the real motion of the dynamic system. In the formulation, both kinematics and kinetics parameters serve as unknowns, and equations of motion are treated as equality constraints. Procedures to choose physical performance measures and appropriate constraints based on the available information about the bio-system are presented. The proposed methodology is illustrated and studied by first predicting the swinging motion of a single pendulum with externally applied torque. The pendulum can represent the motion of upper and lower extremities. This simple problem has analytical solutions and is used to gain insights for the predictive dynamics approach. In addition, a complex human walking task is simulated by using the approach, and realistic results are obtained. Such motion prediction capabilities have a wide variety of applications for industries ranging from automotive to military to clinical analysis and design.

Kayur Patel Computer Science and Engineering University of Washington Seattle, WA kayur@cs.washington.edu ... Jeremy N. Bailenson Communication Stanford University Stanford, CA baileson@stanford.edu ... Sang Hack-Jung, Rosen Diankov, Ruzena ...

With the continuous improvements in videoanalysis techniques, automatic low-cost video surveillance gradually emerges for consumer applications. Video surveillance can contribute to the safety of people in the home and ease control of home-entrance and equipmentusage functions. In this paper, we study a flexible framework for semantic analysis of human behavior from a monocular surveillance video, captured by a consumer camera. Successful trajectory estimation and human-body modeling facilitate the semantic analysis of human activities and events in video sequences. An additional contribution is the introduction of a 3-D reconstruction scheme for scene understanding, so that the actions of persons can be analyzed from different views. The total framework consists of four processing levels: (1) a preprocessing level including background modeling and multiple-person detection, (2) an object-based level performing trajectory estimation and posture classification, (3) an event-based level for semantic analysis, and (4) a visualization level including camera calibration and 3-D scene reconstruction. Our proposed framework was evaluated and has shown its good quality (86% accuracy of posture classification and 90% for events) and effectiveness, as it achieves a near real-time performance (6-8 frames/second). 1

The human motion analysis, for gait or for most of other activities, relies mostly on the use of multibody formulations applied as kinematic or dynamic tools. In many biomechanical applications to gait analysis the choice between using direct or inverse dynamics to obtain the solution of the problem, even in pure kinematics, only depends on the personal preference of the user and not in any particular form of the data available or structure of the equations to be solved. In this work the structure of the equations of a multibody system are reviewed for direct and inverse dynamic analysis. It is shown that if the time dependencies of all degrees-of-freedom of the system are known the inverse dynamics is equivalent to a direct dynamics problem. This equivalence is particularly useful when the problem of the biomechanical analysis consists in finding the muscle forces in an over-actuated biomechanical model that leads to a prescribed motion, which is obtained by using video data acquisition or simply by designing such motion. The problem can then be solved by using optimization procedures in which the objective functions are physiological criteria and, eventually, a measure of matching the prescribed motion. If not used as part of the objective function the prescribed motion is introduced in the optimization problem as nonlinear constraints. The variables of the optimization problem are, for all type of analysis, the muscle forces, directly, or their corresponding muscle activations. It is shown that the natural choice for design variables of the optimal problem is the muscle activations. Two representations of the time history of the muscle actuation are tested in this work: the input sampling where the activations are found in a finite number of time instants and then linearly interpolated in between; the smooth exponential function approach where the actuation is described by a sum of exponential functions being the width and the size of the bumps of each of the functions the unknown quantities. Then the muscle forces are simply obtained by using a Hill type muscle model where the state of force-velocity and the forcelength relations are obtained directly from the kinematics of the biomechanical model. The methods presented in this work are demonstrated and discussed in the framework of two problems associated to the human locomotion apparatus.

Disorders in the autism spectrum are characterized by deficits in social and communication skills such as imitation, pragmatic language, theory of mind, and empathy. The discovery of the "mirror neuron system" (MNS) in macaque monkeys may provide a basis from which to explain some of the behavioral dysfunctions seen in individuals with autism spectrum disorders (ASD).We studied seven right-handed high-functioning male autistic and eight normal subjects (TD group) using functional magnetic resonance imaging during observation and execution of hand movements compared to a control condition (rest).

Tracking the activity of people in indoor environments has gained considerable attention in the robotics community over the last years. Most of the existing approaches are based on sensors which allow to accurately determine the locations of people but do not provide means to distinguish between different persons. In this paper we propose a novel approach to tracking moving objects and their identity using noisy, sparse information collected by id-sensors such as infrared and ultrasound badge systems. The key idea of our approach is to use particle filters to estimate the locations of people on the Voronoi graph of the environment. By restricting particles to a graph, we make use of the inherent structure of indoor environments. The approach has two key advantages. First, it is by far more efficient and robust than unconstrained particle filters. Second, the Voronoi graph provides a natural discretization of human motion, which allows us to apply unsupervised learning techniques to derive typical motion patterns of the people in the environment. Experiments using a robot to collect ground-truth data indicate the superior performance of Voronoi tracking. Furthermore, we demonstrate that EMbased learning of behavior patterns increases the tracking performance and provides valuable information for highlevel behavior recognition.

Disorders in the autism spectrum are characterized by deficits in social and communication skills such as imitation, pragmatic language, theory of mind, and empathy. The discovery of the "mirror neuron system" (MNS) in macaque monkeys may provide a basis from which to explain some of the behavioral dysfunctions seen in individuals with autism spectrum disorders (ASD).We studied seven right-handed high-functioning male autistic and eight normal subjects (TD group) using functional magnetic resonance imaging during observation and execution of hand movements compared to a control condition (rest).

As viewpoint issue is becoming a bottleneck for human motion analysis and its application, in recent years researchers have been devoted to view-invariant human motion analysis and have achieved inspiring progress. The challenge here is to find a methodology which can recognize human motion patterns, to reach increasingly sophisticated levels of human behaviour description. This paper provides a comprehensive survey of this significant research with the emphasis on viewinvariant representation and recognition of poses and actions. In order to help readers understand the integrated process of visual analysis of human motion, this review presents recent development in three major issues involved in a general human motion analysis system, namely human detection, view-invariant pose representation and estimation, behaviour understanding. Public available standard datasets are recommended. The concluding discussion assesses the progress thus far and outlines some research challenges and future directions, and solution to which is essential to achieve the goals of human motion analysis.

The extent to which errors in predicting body segment parameters (SP) influence biomechanical analysis of human motion is unclear. Therefore, the current study quantitatively evaluated the differences in SP estimates using literature predictive functions and computed the effect of SP variation on the kinetic output of walking. For a group of 15 young males, significant differences (PB 0.05) were observed between SP estimates for the leg and thigh using the literature predictive functions, with mass and moment of inertial values differing by more than 40%. Using kinematic and ground reaction force data collected for each subject while walking, inverse dynamic analysis was performed iteratively to compute hip forces and moments while simultaneously varying SP values over nine intervals within 940% of a baseline value. SP variations were found to significantly affect (P B 0.05) most of the kinetic estimates produced, particularly those taken during the swing phase. However, the magnitude of these effects was generally less than 1% of body weight. The data from the current study allows researchers to estimate the errors in kinetic measures due to SP variation. The results also indicate that the accuracy of SP prediction should be of concern in biomechanical research particularly for open chain and high acceleration activities. Further study is necessary to identify the importance of SP variation on other motion skills.

Human mobility modeling plays an essential role in accurately understanding the performance of data forwarding protocols in mobile networks and has been attracting increasing research interest in recent years. People's movement behaviors are strongly affected by their social interactions with each other, which, however, are not sufficiently considered in most existing mobility models. Recent studies in social network theory have provided many theoretical and experimental results, which are useful and powerful for modeling the realistic social dimension of human mobility. In this article we present a novel human mobility model based on heterogeneous centrality and overlapping community structure in social networks. Instead of extracting communities from artificially generated social graphs, our model manages to construct the k-clique overlapping community structure which satisfies the common statistical features observed from distinct real social networks. This approach achieves a good trade-off between complexity and reality. The simulation results of our model exhibit characteristics observed from real human motion traces, especially heterogeneous human mobility popularity, which has significant impact on data forwarding schemes but has never been considered by existing mobility models.

In this paper, we propose a novel concept of movement primitives called Stylistic Dynamic Movement Primitives (SDMPs) for motor learning and control in humanoid robotics. In the SDMPs, a diversity of styles in human motion observed through multiple demonstrations can be compactly encoded in a movement primitive, and this allows style manipulation of motion sequences generated from the movement primitive by a control variable called a style parameter. Focusing on discrete movements, a model of the SDMPs is presented as an extension of Dynamic Movement Primitives (DMPs) proposed by Ijspeert et al. . A novel learning procedure of the SDMPs from multiple demonstrations, including a diversity of motion styles, is also described. We present two practical applications of the SDMPs, i.e., stylistic table tennis swings and obstacle avoidance with an anthropomorphic manipulator.

The characterization of the human motion has a fundamental importance in scientific areas as diverse as medicine, sports, physical therapy, vehicle dynamics or general engineering applications. One of the most challenging problems consists in the evaluation of the internal forces in the human body and in their control, including muscles, ligaments or anatomic joints, without using intrusive techniques. Experimental procedures based on photogrametry and force platforms are used to collect the data required for the numerical methods, based on inverse dynamics and optimization techniques, to calculate the internal forces. A discussion of the formulations used in the context of the State-of-Art and of the applications presented to sports and health sciences cases helps appraising the collaborative work reported here.

For assisting human motion, assistive devices working as muscles would be useful. A robot suit HAL (Hybrid Assistive Limb) has been developed as an assistive device for lower limbs. Human can appropriately produce muscle contraction torque and control joint viscoelasticity by muscle effort such as co-contraction. Thus, to implement functions equivalent to human muscles using HAL, it is necessary to control viscoelasticity of HAL as well as to produce torque in accordance with operator's intention. Therefore the purpose of this study is to propose a control method of HAL using Biological and Motion Information. In this method, HAL produces torque corresponding to muscle contraction torque by referring to the myoelectricity that is biological information to control operator's muscles. In addition, the viscoelasticities of HAL are adjusted in proportion to operator's viscoelasticity that is estimated from motion information by using an on-line parameter identification method. To evaluate the effectiveness of the proposed method, the method was applied to a swinging motion of a lower leg. When this method was applied, HAL could work like operator's muscles in the swinging motion, and as a consequence, the muscle activities of the operator were reduced. As a result of this experiment, we confirmed the effectiveness of the proposed method.

Although a high proportion of the motion selective cells in primate motion area, MT, are disparity-selective, there is no convincing evidence for cells in this area specific to stereomotion-in-depth and the neural basis of stereomotion processing remains obscure. With functional magnetic resonance imaging (fMRI), we identify an occipito-temporal region activated by cyclopean stereomotion-in-depth stimulation, centered anterior to the human motion complex hMT+. This discovery suggests a reconceptualization of the architecture of the motion complex to incorporate the processing of motion in the third dimension.

Human motion estimation is an important issue in automotive, security or home automation applications. Radar systems are well suited for this because they are robust, are independent of day or night conditions and have accurate range and speed domain. The human response in a radar range -speed -time measurement behaves like an extended target where legs and arms coincide. A mutually non-coherent radar sensor network makes it possible to estimate additional information of the extended target response. To keep the system low cost the network uses commercial off-the-shelf (COTS) radar sensors without synchronisation of frequency or phase between the radar sensors. This article presents the results of human motion estimation with a mutually non-coherent radar sensor network. The calibration, radar processing, parameter estimation and classification of extended human objects are described. The swinging and rotating moving body parts give elliptical shapes in the differential range-speed responses. A model fit gives the legs and arms parameters on which classification is possible.

The human motion analysis, for gait or for most of other activities, relies mostly on the use of multibody formulations applied as kinematic or dynamic tools. In many biomechanical applications to gait analysis the choice between using direct or inverse dynamics to obtain the solution of the problem, even in pure kinematics, only depends on the personal preference of the user and not in any particular form of the data available or structure of the equations to be solved. In this work the structure of the equations of a multibody system are reviewed for direct and inverse dynamic analysis. It is shown that if the time dependencies of all degrees-of-freedom of the system are known the inverse dynamics is equivalent to a direct dynamics problem. This equivalence is particularly useful when the problem of the biomechanical analysis consists in finding the muscle forces in an over-actuated biomechanical model that leads to a prescribed motion, which is obtained by using video data acquisition or simply by designing such motion. The problem can then be solved by using optimization procedures in which the objective functions are physiological criteria and, eventually, a measure of matching the prescribed motion. If not used as part of the objective function the prescribed motion is introduced in the optimization problem as nonlinear constraints. The variables of the optimization problem are, for all type of analysis, the muscle forces, directly, or their corresponding muscle activations. It is shown that the natural choice for design variables of the optimal problem is the muscle activations. Two representations of the time history of the muscle actuation are tested in this work: the input sampling where the activations are found in a finite number of time instants and then linearly interpolated in between; the smooth exponential function approach where the actuation is described by a sum of exponential functions being the width and the size of the bumps of each of the functions the unknown quantities. Then the muscle forces are simply obtained by using a Hill type muscle model where the state of force-velocity and the forcelength relations are obtained directly from the kinematics of the biomechanical model. The methods presented in this work are demonstrated and discussed in the framework of two problems associated to the human locomotion apparatus.

In sport science, as in clinical gait analysis, optoelectronic motion capture systems based on passive markers are widely used to recover human movement. By processing the corresponding image points, as recorded by multiple cameras, the human kinematics is resolved through multistage processing involving spatial reconstruction, trajectory tracking, joint angle determination, and derivative computation. Key problems with this approach are that marker data can be indistinct, occluded or missing from certain cameras, that phantom markers may be present, and that both 3D reconstruction and tracking may fail. In this paper, we present a novel technique, based on state space filters, that directly estimates the kinematical variables of a virtual mannequin (biomechanical model) from 2D measurements, that is, without requiring 3D reconstruction and tracking. Using Kalman filters, the configuration of the model in terms of joint angles, first and second order derivatives is automatically updated in order to minimize the distances, as measured on TV-cameras, between the 2D measured markers placed on the subject and the corresponding back-projected virtual markers located on the model. The Jacobian and Hessian matrices of the nonlinear observation function are computed through a multidimensional extension of StirlingÕs interpolation formula. Extensive experiments on simulated and real data confirmed the reliability of the developed system that is robust against false matching and severe marker occlusions. In addition, we show how the proposed technique can be extended to account for skin artifacts and model inaccuracy.

A growing number of robotics labs cooperate in the "Blender for Robotics" project [4], to integrate the open source computer animation program Blender into their research. In the first place as a visualisation tool, but gradually more and more as a full-fledged component in a robot simulation and programming environment. The lab in Leuven focuses on improving Blender support for the task specification and control of complex, sensor-based robot tasks, in which also the motion of the human(s) interacting with the task must be tracked.

A three steps methodology is proposed to derive consistent sets of local features which may be easily compared between the different samples of a stereotypical human handwriting movement, allowing the statistical analysis its local variability. This technique is illustrated using the Sigma-Lognormal modeling of on-line triangular trajectory patterns obtained from a standardized neuromuscular task. The overall approach can be adapted and generalized to the analysis of the end-effector kinematics of many planar upper limb movements.

The present report reviews a series of functional magnetic resonance imaging (fMRI) activation studies conducted in parallel in awake monkeys and humans using the same motion stimuli in both species. These studies reveal that motion stimuli engage largely similar cortical regions in the two species. These common regions include MT/V5 and its satellites, of which FST contributes more to the human motion complex than is generally assumed in human imaging. These results also establish a direct link between selectivity of MT/V5 neurons for speed gradients and functional activation of human MT/V5 by three-dimensional (3D) structure from motion stimuli. On the other hand, striking functional differences also emerged: in humans V3A and several regions in the intraparietal sulcus (IPS) are much more motion sensitive than their simian counterparts.

Understanding the spatial dimensionality and temporal context of human hand actions can provide representations for programming grasping actions in robots and inspire design of new robotic and prosthetic hands. The natural representation of human hand motion has high dimensionality. For specific activities such as handling and grasping of objects, the commonly observed hand motions lie on a lower-dimensional non-linear manifold in hand posture space. Although full body human motion is well studied within Computer Vision and Biomechanics, there is very little work on the analysis of hand motion with nonlinear dimensionality reduction techniques. In this paper we use Gaussian Process Latent Variable Models (GPLVMs) to model the lower dimensional manifold of human hand motions during object grasping. We show how the technique can be used to embed high-dimensional grasping actions in a lower-dimensional space suitable for modeling, recognition and mapping.

& Retrieval of conceptual information from action pictures causes greater activation than from object pictures bilaterally in human motion areas (MT/MST) and nearby temporal regions. By contrast, retrieval of conceptual information from action words causes greater activation in left middle and superior temporal gyri, anterior and dorsal to the MT/MST. We performed two fMRI experiments to replicate and extend these findings regarding action words. In the first experiment, subjects performed conceptual judgments of action and object words under conditions that stressed visual semantic information. Under these conditions, action words again activated posterior temporal regions close to, but not identical with, the MT/MST. In the second experiment, we included conceptual judgments of manipulable object words in addition to judgments of action and animal words. Both action and manipulable object judgments caused greater activity than animal judgments in the posterior middle temporal gyrus. Both of these experiments support the hypothesis that middle temporal gyrus activation is related to accessing conceptual information about motion attributes, rather than alternative accounts on the basis of lexical or grammatical factors. Furthermore, these experiments provide additional support for the notion of a concrete to abstract gradient of motion representations with the lateral occipito-temporal cortex, extending anterior and dorsal from the MT/MST towards the peri-sylvian cortex. &

Recently the importance of autonomous monitoring has become an essential part of the society. Autonomous monitoring and control in a home environment can lead to energy savings. Also autonomous monitoring can serve as an alerting mechanism when something is out of the norm so that the controlling mechanism can rectify the problem. In this paper we present a system we developed using multiple sensors for autonomous monitoring and control with relatively low cost components. The sensors we integrated together in to one central operation unit can monitor room temperature, humidity, human motion activity etc. We present a prototype implementation of an energy efficient home environment using our proposed centralized monitoring platform. Using a sample scenario we show that the energy consumption in a smart home is about 15kWh while that of a normal home is about 35kWh.

In this paper, a walking pattern filter for shapememory-alloy (SMA)-actuated biped robots is presented. SMAs are known for their high power-to-mass ratio as well as slow response. When used as actuators, the SMA speed limitation can potentially lead to stability problems for biped robots. The presented filter adapts the human motion such that an SMA biped robot maintains a stable walking pattern. The zero moment point (ZMP) is used as the main criterion of the filter to guarantee the stability of the motion. The SMA actuators are designed based on the dynamics and kinematics of the motion. The response time of each SMA actuator is modeled in order to estimate the behavior of the actuator in realizing the given trajectory. After applying the delay times to the motion, the new trajectories are generated and evaluated by the filter for the ZMP criterion. Using simulations, it is shown that the filter can generate smooth trajectories for the SMA-actuated biped robots. The filter furthermore guarantees the stability of a robot mimicking the human walking motion.

This paper presents an approach for viewpoint invariant human action recognition, an area that has received scant attention so far, relative to the overall body of work in human action recognition. It has been established previously that there exist no invariants for 3D to 2D projection. However, there exist a wealth of techniques in 2D invariance that can be used to advantage in 3D to 2D projection. We exploit these techniques and model actions in terms of view-invariant canonical body poses and trajectories in 2D invariance space, leading to a simple and effective way to represent and recognize human actions from a general viewpoint. We first evaluate the approach theoretically and show why a straightforward application of the 2D invariance idea will not work. We describe strategies designed to overcome inherent problems in the straightforward approach and outline the recognition algorithm. We then present results on 2D projections of publicly available human motion capture data as well on manually segmented real image sequences. In addition to robustness to viewpoint change, the approach is robust enough to handle different people, minor variabilities in a given action, and the speed of aciton (and hence, frame-rate) while encoding sufficient distinction among actions.

This paper presents a new augmented-reality system designed to generate visual enhancements for TV broadcasted court-net sports. A probabilistic method based on the Expectation Maximization (EM) procedure is utilized to find the optimal feature points, thereby enabling the automatic acquisition of the camera parameters from the TV image with high accuracy. A virtual camera derived from the original camera, helps to synthesize a variety of virtual scenes, such as the scene from the viewpoint of a player, depending on the intention of the user. To preserve the visual nature of the original human motion, the player's shape and texture are extracted from the real video and texture-mapped onto the virtual video. The system was tested over a set of court-net sports videos containing tennis, badminton and volleyball and demonstrated promising results.

Recently, neuroimaging techniques have been applied to the study of human motion perception, complementing established techniques such as psychophysics, neurophysiology and neuropsychology. Because vision, particularly motion perception, has been studied relatively extensively, it provides an interesting case study to examine the contributions and limitations of neuroimaging to cognitive neuroscience. We suggest that in the domain of motion perception neuroimaging has: (1) revealed an extensive network of motion areas throughout the human brain, in addition to the well-studied motion complex (MT+); (2) veri®ed and extended ®ndings from other techniques; (3) suggested extensive top-down inuences on motion perception; and (4) allowed experimenters to examine the neural correlates of awareness. We discuss these contributions, along with limitations and future directions for the neuroimaging of motion. Ó

A learning scheme based on Random Forests is used to decode the EMG activity of 16 muscles of the human arm-hand system to a continuous representation of kinematics in reach-to-grasp movements in 3D space. Classification methods are used to discriminate between significantly different reach to grasp strategies, formulating a switching mechanism that may trigger the use of position and object-specific decoding models (task-specificity). These task-specific models can achieve better estimation results than the general models for the kinematics of different reach-to-grasp movements. The efficacy of the proposed methodology is assessed through a strict validation procedure, based on everyday life reach-to-grasp scenarios and data not previously seen during training. Finally, for demonstration purposes, the authors teleoperate an armhand model in the OpenRave simulation environment using the estimated from the EMG signals human motion.

Remote teleoperation of a robot manipulator by a human operator is often necessary in unstructured dynamic environments when human presence at the robot site is undesirable. Mechanical and other contacting interfaces used in teleoperation require unnatural human motions for object manipulation tasks or they may hinder human motion. Previous vision-based approaches have used only a few degrees of freedom for hand motion and have required hand motions that are unnatural for object manipulation tasks. This paper presents a noncontacting vision-based method of robot teleoperation that allows a human operator to communicate simultaneous six-degree-of-freedom motion tasks to a robot manipulator by having the operator perform the three-dimensional human hand-arm motion that would naturally be used to complete an object manipulation task. A vision-based human-robot interface is used for communication of human motion to the robot and for feedback of the robot motion and environment to the human operator. Teleoperation under operator position control was performed with high accuracy in object placement on a target. Semi-autonomous traded and shared control using robot-vision guidance aided in achieving a more accurate positioning and orientation of the end-effector for object gripping tasks.

Wireless self-organizing networks such as mesh networks strive hard to get rid of mobility and radio propagation effects. Links -the basic elements ensuring connectivity in wireless networks -are impacted first from them. But what happens if one could mitigate these effects by forecasting the links' future states?

Figure 1. Above we illustrate the range of motions typical of a single pose cluster, or local linear model. A mean pose, shown as the center of each progression and four orthogonal axes of variation specify this model. The range of motion along each axis is illustrated in both directions from the mean.

Virtual reality offers new tools for human motion understanding. Several applications have been widely used in teleoperation, military training, driving and ying simulators, and so forth. We propose to test if virtual reality is a valid training tool for the game of handball. We focused on the duel between a handball goalkeeper and a thrower. To this end, we de ned a pilot experiment divided into two steps: an experiment with real subjects and another one with virtual throwers. The throwers' motions were captured in order to animate their avatar in a reality center. In this paper, we focused on the evaluation of presence when a goalkeeper is confronting these avatars. To this end, we compared the goalkeeper's gestures in the real and in the virtual experiment to determine if virtual reality engendered the same movements for the same throw. Our results show that gestures did not differ between the real and virtual environment. As a consequence, we can say that the virtual environment offered enough realism to initiate natural gestures. Moreover, as in real games, we observed the goalkeeper's anticipation to allow us to use virtual reality in future work as a way to understand the goalkeeper and thrower interactions. The main originality of this work was to measure presence in a sporting application with new evaluation methods based on motion capture.

Motion-capture techniques are rarely based on orientation measurements for two main reasons: (1) optical motion-capture systems are designed for tracking object position rather than their orientation (which can be deduced from several trackers), (2) known animation techniques, like inverse kinematics or geometric algorithms, require position targets constantly, but orientation inputs only occasionally. We propose a complete human motion-capture technique based essentially on orientation measurements. The position measurement is used only for recovering the global position of the performer. This method allows fast tracking of human gestures for interactive applications as well as high rate recording. Several motion-capture optimizations, including the multijoint technique, improve the posture realism. This work is well suited for magnetic-based systems that rely more on orientation registration (in our environment) than position measurements that necessitate difficult system calibration.

This paper proposes a novel computer vision approach that processes video sequences of people walking and then recognises those people by their gait. Human motion carries different information that can be analysed in various ways. The skeleton carries motion information about human joints, and the silhouette carries information about boundary motion of the human body. Moreover, binary and gray-level images contain different information about human movements. This work proposes to recover these different kinds of information to interpret the global motion of the human body based on four different segmented image models, using a fusion model to improve classification. Our proposed method considers the set of the segmented frames of each individual as a distinct class and each frame as an object of this class. The methodology applies background extraction using the Gaussian Mixture Model (GMM), a scale reduction based on the Wavelet Transform (WT) and feature extraction by Principal Component Analysis (PCA). We propose four new schemas for motion information capture: the Silhouette-Gray-Wavelet model (SGW) captures motion based on grey level variations; the Silhouette-Binary-Wavelet model (SBW) captures motion based on binary information; the Silhouette-Edge-Binary model (SEW) captures motion based on edge information and the Silhouette Skeleton Wavelet model (SSW) captures motion based on skeleton movement. The classification rates obtained separately from these four different models are then merged using a new proposed fusion technique. The results suggest excellent performance in terms of recognising people by their gait.

Three dimensional motion capture facility is a powerful tool for quantitative and qualitative assessment of multijoint external movements. Electro-myograph (EMG) signals give the physiologic information of muscles while doing motions. In this paper, our objective is to integrate these two different bio-medical data together and to extract precise and accurate feature information for classifying the human motions. When both forms of data are integrated and analyzed together, the information achieved will be immensely useful to quantify the complex human motions for medical reasons or sport performances. These biological quantifications of biomechanical data, are useful for gait analysis and several orthopedic applications, such as joint mechanics, prosthetic designs, and sports medicines. The different dimensionality reduction approaches such Integral of Absolute value and Weighted Singular Value Decomposition are used to extract the preliminary features from EMG and motion capture data respectively. On combining these feature vectors, fuzzy clustering such as Fuzzy c-means (FCM) is performed on these vectors that are mapped as the points in multi-dimensional feature space. We get the degree of memberships with every cluster for each mapped point. This extracted information is used as the final feature vectors for classifying the human motions. 1-4244-0832-6/07/$20.00 ©2007 IEEE.

More and more sensors are embedded in human body for medical applications, for sport. The short lifetime of the batteries, available on the market, reveals a real problem of autonomy of these systems. A promising alternative is to scavenge the ambient energy such as the mechanical one. Up to now, few scavenging structures have operating frequencies compatible with ambient one. And, most of the developed structures are rigid and use vibration as mechanical source. For these reasons, we developed a scavenger that operates in a large frequency spectrum from quasistatic to dynamic range. This generator is fully flexible, light and does not hamper the human motion. Thus, we report in this paper an analytical model for dielectric generator with news electrical and mechanical characterization, and the development of an innovating application: scavenging energy from human motion. The generator is located on the knee and design to scavenge 0.1mJ per scavenging cycle at a frequency of 1Hz, enough to supply a low consumption system and with a poling voltage as low as possible to facilitate the power management. Our first prototype is a membrane with an area of 5*3cm and 31µm in thickness which scavenge 0.1mJ under 170V at constant charge Q.

In this paper, we present an analytic-iterative Inverse Kinematics (IK) method, called Sequential IK (SIK), that reconstructs 3D human full-body movements in real time. The input data for the reconstruction is the least possible (i.e., the positions of wrists, ankles, head and pelvis) in order to be usable within a low-cost human motion capture system that would track only these six features. The performance of our approach is compared to other well-known IK methods in reconstruction quality and computation time obtaining satisfactory results for both. The paper first describes how we handle the spine and the clavicles before offering a simple joint limit model for ball-and-socket joints and a method to avoid self-collisions induced by the elbow. The second part focuses on the algorithms comparison study.

When spatially segregated visual stimuli are presented in alternation, subjects may perceive a single stimulus moving between the two positions (apparent motion). By adjusting spatial and temporal parameters, an ambiguous condition can be created in which perception of back-and-forth motion alternates with the perception of two stationary blinking stimuli. We presented subjects with such ambiguous stimuli, asked them to signal periods of perceived motion and blinking, and measured brain activity with functional magnetic resonance imaging. Multiple regression analysis revealed that early visual areas responded with equal strength during both perceptual conditions, whereas hMT+(V5) (the human motion complex that includes the human homolog of MT and its satellites) was more active during the perception of apparent motion. These results indicate that neurons in hMT+ participate in the constructive process that creates a continuous motion percept from a discontinuous visual input.

Here a flexible framework for the sonification of human movement data is presented, capable of processing standard kinematic motion capture data as well as derived quantities such as force data. Force data are computed by inverse dynamics algorithms and can be used as input parameters for real time sonification. Simultaneous visualization is provided using OpenGL.

The goal of this study is developing a biped humanoid robot that can observe a human dance performance and imitate it. To achieve this goal, we propose a task model of lower body motion, which consists of task primitives (what to do) and skill parameters (how ...