Proposition of a Human Motion Tracking Method by Temporal-Spatial Segmentation in an Image Sequence

In this paper, a ncw approach of a 3D human motion tracking mcthod in rcal time is proposed. The basis of thc nicthod is thc use of two scgrncntations, one temporal and the othcr spatial. It allows to extract the pertinent parts of the body (that are matcrialiscd by rcflective markcrs) and to track them : the speed computing is only made on moving objccts in thc sccne that are intcrcsting to track. The use of stereo vision allows to calculate 3D position of cach marker and a prediction of thc 2D positions of all of them in the next pair of imagcs is donc to optimise the tracking. The implcmcntation of thc algorithm, which is rcaliscd on a transputer network, reduces the processing tirnc.