A new method for representing and recognizing human bodymovements is presented. Assuming the availability of Cartesian tracking data, we develop techniques for representation of movements basedon spacecurves in subspaces of a "phase space." The phase space has axes of joint angles and torso location and attitude, and the axes of the subspaces are subsets of the axes of the phase space. Using this representation we develop a system for learning new movements from ground truth data by searching for constraints which are in effect during the movement to be learned, and not in effect during other movements. We then use the learned representation for recognizing movements in data. Prior approaches by other researchers used a small number of classification categories, which demanded less attention to representation. We train and test the system on nine fundamental movements from classical ballet performed by two dancers. The system learns and accurately recognizes the nine movements in an un...

This article deals with the problem of classification of human activities from video. Our approach uses motion features that are computed very efficiently, and subsequently projected into a lower dimensional space where matching is performed. Each action is represented as a manifold in this lower dimensional space and matching is done by comparing these manifolds. To demonstrate the effectiveness of this approach, it was used on a large data set of similar actions, each performed by many different actors. Classification results were very accurate and show that this approach is robust to challenges such as variations in performers' physical attributes, color of clothing, and style of motion. An important result of this article is that the recovery of the three-dimensional properties of a moving person, or even the two-dimensional tracking of the person's limbs need not precede action recognition.

In classic demonstrations of apparent motion, observers typically report seeing motion along the shortest possible path between 2 sequentially presented objects. However, when realistic photographs of a human body are sequentially presented at slow temporal rates, observers report paths of apparent motion that are consistent with the movement limitations of the human body even when those paths are not the shortest possible. The current set of experiments examined those aspects of the human form that lead to the perception of biomechanically consistent paths of motion. The authors ' findings suggest that the perception of apparent biological motion extends to human movements that involve inanimate objects. The authors also report that observers can perceive apparent motion of nonbiological objects in a manner similar to apparent motion of human bodies. However, a global hierarchy of orientation and position cues resembling the human form is required for the perception of these paths. Our successful interaction with the environment depends largely on our ability to interpret visual forms and motions accurately and rapidly. This ability appears to be particularly important for the visual perception of human movement. Johansson and his colleagues generated extensive interest in the perception of human movement with the demonstration that even under extremely reduced conditions, observers could readily identify human locomotion

To understand more about how animate motion is generated and perceived, we need quantitative analyses of motion trajectories from organisms interacting in various important adaptive tasks. Such data is difficult to obtain for most animals, but one species provides a ready source. We have developed software that allows human subjects to generate such motion data by interacting across a computer network in on-screen pursuit and evasion, fighting, courtship, and play. Each subject uses a mouse to control a "bug" that moves in a 2-D environment with another bug controlled by a second remote subject. We have visualized and analyzed the resulting motion data for each task in several ways: 3-D spacetime plots of the trajectories themselves, scatterplots of one bug's positions relative to the other, and statistical measures of trajectory parameters including velocity, vorticity, and energy. All of these methods distinguish between the different motion categories. Having human subjects perform ...

Automatic recognition by gait is subject to increasing interest and has the unique capability to recognize people at a distance when other biometrics are obscured. Its interest is reinforced by the longstanding computer vision interest in automated non-invasive analysis of human motion. Its recognition capability is supported by studies in other domains such as medicine (biomechanics), mathematics and psychology which continue to suggest that gait is unique. Further, examples of recognition by gait can be found in literature, with early reference by Shakespeare concerning recognition by the way people walk. Current approaches confirm the early results that suggested gait could be used for identification, and now on much larger databases. This has been especially

Abstract not found

Articulated motion is a subset of non-rigid motion in which the object of interest is composed of several rigid components connected to each other by ball and hinge joints. The human body, many animals and insects, and machinery all exhibit such motion. This dissertation addresses the problem of vision-based tracking and analysis of this type of motion. The importance of this problem can be seen in many application domains including surveillance, traffic monitoring, entertainment, user interfaces, medicine, sports, video annotation, and image compression. This dissertation deals with two important subproblems of the general problem: whole-body tracking and motion recognition. In whole-body tracking, the body is tracked as one unit without paying attention to the details of the posture and limbs. Current solutions to this problem suffer from being too sensitive to small changes in the environment. We present a novel approach which reduces these restrictions significantly. This is achieved by separating the concepts of a blob from that of a body and by tracking each independently while maintaining a many-to-many relationship between the two. The approach makes use of the Extended Kalman Filter and outputs trajectory information in world coordinates. The method was tested by tracking pedestrians in a variety of environments and achieved real-time performance and a high degree of robustness. Motion recognition is the high level problem of classifying an action taking place in a video sequence into one of several action categories. Most of the present approaches attempt to perform three-dimensional reconstruction of the articulated shape prior to recognition, which is an inherently difficult problem made even more difficult due to the nonrigidity of the articulated object. W...

To animate a character, a number of poses are displayed in quick succession in order to create the illusion of motion. For most real-time applications, such as games, the pose update rate is largely constrained by the available hardware and overall simulation complexity. To date, no analysis of the factors that affect the perceived smoothness of animated virtual characters has been presented. In the first perceptual studies aimed at identifying such factors and their interactions, we have determined some thresholds that could be used to produce acceptably smooth human animations in a variety of conditions. Some interesting results were found, e.g., that character type, clothing, scene complexity or motion synchronicity had no effect on smoothness perception in our experiments, but cycle rate, linear velocity, motion complexity and group size all had a significant effect, with slower or lower intensity movements generally requiring fewer updates. Our results should be of real practical use to character animators in various application areas, but in particular to developers of real-time applications where Simulation Levels Of Detail (SLOD) need to be employed.

To understand the visual analysis of biological motion, subjects viewed dynamic, stick figure renditions of a walker, car, or scissors through apertures. As a result of the aperture problem, the motion of each visible edge was ambiguous. Subjects readily identified the human figure but were unable to identify the car or scissors through invisible apertures. Recognition was orientation specific and robust across a range of stimulus durations, and it benefited from limb orientation cues. The results support the theory that the visual system performs spatially global analyses to interpret biological motion displays. This work was funded by NIH Grant 099310 to the first author. Some of these results were presented at the 1994 Cognitive Neuroscience Conference in San Francisco. We thank Bennett Bertenthal, Judy Kegl, and an anonymous reviewer for numerous helpful suggestions on a previous draft of this manuscript. We also thank Jean Dominique LaJoux for generously providing valuable information on the research of E. J. Marey. Correspondence should be addressed to

People frequently analyze the actions of other people for the purpose of action coordination. To understand whether such self-relative action perception differs from other-relative action perception, the authors had observers either compare their own walking speed with that of a point-light walker or compare the walking speeds of 2 point-light walkers. In Experiment 1, observers walked, bicycled, or stood while performing a gait-speed discrimination task. Walking observers demonstrated the poorest sensitivity to walking speed, suggesting that perception and performance of the same action alters visual–motion processes. Experiments 2–6 demonstrated that the processes used during self-relative and other-relative action perception differ significantly in their dependence on observers ’ previous motor experience, current motor effort, and potential for action coordination. These results suggest that the visual analysis of human motion during traditional laboratory studies can differ substantially from the visual analysis of human movement under more realistic conditions. Extensive research has been devoted to understanding the visual analysis of human movement. With the use of point-light displays, researchers have repeatedly demonstrated the visual system’s keen and robust sensitivity to human movement. In classic demonstrations of this sensitivity, human models with points of light attached to their major joints are filmed while they perform various actions. Although only a few points remain visible in the resultant displays, observers can quickly and accurately identify particular human actions, such as dancing and boxing (e.g., Dittrich, 1993; Johansson, 1973), as well as complex psychological attributes, such as deception and emotion (e.g., Brownlow, Dixon, Egbert, & Radcliffe,