A full body human motion capture system using particle filtering and on-the-fly edge detection

In this paper we present a full body human motion capture system based on particle filtering operating on monocular image sequences. Distinguishing our approach from others is that edge detection is not carried out globally on the whole picture in a preprocess manner, but is done locally on-the-fly during the calculation of the likelihood function. This approach is effective and flexible at the same time, allowing the use of various edge detection algorithms with only small modifications. Another special feature of our system is a highly optimized occlusion test based on a fast point-in-triangle test. Our system has been designed carefully with respect to serve as a basis for various research activities in the near future, including the incorporation of stereo vision. A general framework architecture for particle filters and an extension for appliance to edge tracking has been developed with which it is possible to test a whole range of search space decomposition variants with minimum implementation effort.