Automatic Urban Road Users' Tracking System

This paper presents a Dynamic Gradient Pattern (DGP) based on Kalman filtering technique for urban road users tracking. DGP technique is proposed to enhance rigid object descriptive ability for improved verification. DGP descriptor along with weighted centroid was integrated with a Kalman filtering framework to enhance data association robustness and tracking accuracy. To handle multiple objects tracking, a DGP verification approach is addressed based on normalized Bhattacharyya distance. The proposed technique achieves a closer trajectory for rigid body movement. The DGP descriptor can discriminate the objects correctly, and it overcomes the partial occlusion and misdetection by verifying object location using the normalized Bhattacharyya distance between DGP features. Experimental evaluation is performed on urban videos that include a slow-motion temporary stop and partial occlusion. The experimental results demonstrate that the detecting and tracking accuracy are above 98.08% and 97.70% respectively.