|
Song, Yang Monocular Perception of Biological Motion - Detection and Labeling Biological motion perception is an important cue for human social interactions. Our visual system is able to extract good 3D information from very poor data: a few bright spots reproducing the motions of the main joints can evoke a compelling impression of human body motion (JOHANNSON 73). We propose a computational model for detecting a moving human body and for labeling its parts. It is based on a global probabilistic model of the position and motion of the body parts. The most likely position and motion of the body is detected by maximizing the joint probability density function (PDF) of the position and velocity of the body parts. The PDF is approximated by a product of PDFs defined on triplet-cliques by exploiting the kinematic-chain structure of the human body. Dynamic programming is used for calculating efficiently the best global labeling on an approximation of the PDF. The computational cost is on the order of $N^4$ where N is the number of features detected. We tested our model with experiments carried on a variety of periodic and non-periodic body motions viewed monocularly for a total of approximately 30,000 frames. Point-markers were strapped to the joints of the subject for facilitating image analysis. We find an average of 1\% labeling error; the experiments also suggest a high degree of viewpoint-invariance.
|