Robust motion tracking of multiple objects with KL-IMMPDAF

This paper describes how the image sequences taken by a stationary video ca mera may be effectively processed to detect and track moving objects agains t a stationary background in real-time. Our approach is first to isolate th e moving objects in image sequences via a modified adaptive background esti mation method and then perform token tracking of multiple objects based oil features extracted from the processed image sequences. In feature based mu ltiple object tracking, the most prominent, tracking issues are track initi alization, data association, occlusions tills to traffic congestion, and ob ject maneuvering. While there are limited past works addressing these probl ems, most relevant tracking systems proposed in the past are independently focused to either "occlusion" or "data association" only. In this paper, we propose the KL-IMMPDA (Kanade Lucas-Interacting Multiple Model Probabilist ic Data Association) filtering approach for multiple-object tracking to col lectively address the key issues. The proposed method essentially employs o ptical flow measurements for both detection and track initialization while the KL-IMMPDA filter is used to accept or reject measurements, which belong to other objects. The data association performed by the proposed KL-IMMPDA results in an effective tracking scheme, which is robust to partial occlus ions and image clutter of object maneuvering. The simulation results show a significant performance improvement for tracking multi-objects in occlusio n and maneuvering, when compared to other conventional trackers such as Kal man filter.