A Dual-Stage Robust Vehicle Detection and Tracking for Real-Time Traffic Monitoring

This paper addresses the important problem of detecting and tracking vehicles in outdoor dynamic scenes as part of a real-time traffic surveillance system. The proposed solution is based on a dual-stage approach, using a pixel-level stage to extract foreground object from background scenes and a block-level stage to detect and track vehicles. The pixel-level stage combines a multi-background modelling with a dynamic thresholding, using a low-scale quasi-connected-components as a first stage for image object grouping/cleaning. The block-level performs a 8 times 8 block-region analysis defining a block energy function that is used to label the blocks belonging to different vehicles and track them over a stack of images. This approach has proven to be very helpful for occlusion reasoning. The proposed solution has the ability to overcome some of the most difficult problem that arise in outdoor scenes such as illumination variations, shadow-casts and waving movement resulting from trees and camera vibration. The performance and robustness of the proposed algorithm is shown using real highway traffic monitoring situations