Non-invasive 3D dynamic object analysis

Abstract

A non-invasive system is required to obtain three-dimensional subject extraction and description for recognition by gait. Of current three-dimensional systems, multi-view approaches appear to be the most suitable. To handle not only concavities, but also noise and occlusion, the volume intersection approach has been formulated as an evidence gathering process for moving object extraction. Results on synthetic imagery show that the technique does indeed process a multi-view image sequence to derive parameters of interest as such giving a suitable basis for development as a marker-less gait analysis system.