Monocular Perception of Biological Motion in Johansson Displays

Computer perception of biological motion is key to developing convenient and powerful human–computer interfaces. Algorithms have been developed for tracking the body; however, initialization is done by hand. We propose a method for detecting a moving human body and for labeling its parts automatically in scenes that include extraneous motions and occlusion. We assume a Johansson display, i.e., that a number of moving features, some representing the unoccluded body joints and some belonging to the background, are supplied in the scene. Our method is based on maximizing the joint probability density function (PDF) of the position and velocity of the body parts. The PDF is estimated from training data. Dynamic programming is used for calculating efficiently the best global labeling on an approximation of the PDF. Detection is performed by hypothesis testing on the best labeling found. The computational cost is on the order of N^4 where N is the number of features detected. We explore the performance of our method with experiments carried on a variety of periodic and nonperiodic body motions viewed monocularly for a total of approximately 30,000 frames. The algorithm is demonstrated to be accurate and efficient