Canton-Ferrer C, Canton-Ferrer C, Casas J, Tekalp M, Pardàs M. Projective Kalman Filter: Multiocular Tracking of 3D Locations Towards Scene Understanding. In 2nd Joint Workshop on Multimodal Interaction and Related Machine Learning Algorithms. 2005.  (330.58 KB)

Abstract

This paper presents a novel approach to the problem of estimating and tracking 3D locations of multiple targets in a scene using measurements gathered from multiple calibrated cameras. Estimation and tracking is jointly achieved by a newly conceived computational process, the Projective Kalman ¯lter (PKF), allowing the problem to be treated in a single, uni¯ed framework. The projective nature of observed data and information redundancy among views is exploited by PKF in order to overcome occlusions and spatial ambiguity. To demonstrate the e®ectiveness of the proposed algorithm, the authors present tracking results of people in a SmartRoom scenario and compare these results with existing methods as well.