Multiple camera fusion based on DSmT for tracking objects on ground plane

This paper presents comparative results of a model for multiple camera fusion, which is based on Dezert-Smarandache theory of evidence. Our architecture works at the decision level to track objects on a ground plane using predefined zones, producing useful information for surveillance tasks such as behavior recognition. Decisions from cameras are generated by applying a perspective-based basic belief assignment function, which represent uncertainty derived from cameras perspective while tracking objects on ground plane. Results obtained from applying our tracking model to CGI animated simulations and real sequences are compared to the ones obtained by Bayesian fusion, and show how DSm theory of evidence overcomes Bayesian fusion for this application.