Kalman filter-based tracking of multiple similar objects from a moving camera platform

Vision-based tracking is becoming increasing attractive, with the availability of cost-efficient vision systems with a high level of computational power. One challenge in this area of control is the tracking of multiple stationary objects of similar appearance from a moving camera, without identity confusion. In this paper we propose a modified Kalman filter estimator of object location and velocity with robustness to measurement occlusion and spurious measurements. This algorithm includes a novel measurement assignment algorithm that robustly creates a mapping between unordered detected objects and Kalman estimates. We will show that our formulation successfully tracks and identifies multiple similar objects under dynamic camera movement and partial object occlusion.