NOVEL APPEARANCE MODEL FOR MULTIPLE OBJECTS TRACKING

In multiple target tracking (MTT) the moving objects must be tracked despite changes in their appearances. Motion challenges like partial or full occlusion, merge, split and reappearance of the moving objects increase the complexity of the MTT algorithm. In this paper, both appearance changes and motion challenges are considered for multiple target tracking. Tracking by detection method in Kalman filter framework is used in the proposed algorithm. Texture based appearance model using normalized gray level co-occurrence matrix (GLCM) is proposed in the work presented in this paper. The target association is achieved with the distance criteria applied on the appearance model along with the spatial distance between target and detection obtained through the temporal model. The appearance model must be robust to accommodate the target's appearance changes during motion. The phase congruency approach makes our proposed tracker invariant to image illumination and contrast variations. It is a simple but very efficient approach. The performance of the proposed normalized phase congruency of normalized GLCM tracking algorithm is evaluated on the standard benchmark TUD dataset. The experimental results have demonstrated that the proposed tracker's sensitivity in terms of Recall and multiple objects tracking precision (MOTP) outperforms the state-of-the-art tracking algorithms