Visual Tracking System for Dense Traffic Intersections

In this paper, a real-time visual tracking system for dense traffic intersections is presented. The system structure is based on background estimation and subtraction method. A novel method for background estimation based on statistical video background extraction (SVBE) is presented. A moving object detection algorithm for performing background subtraction from original image has been improved. To reduce noise, to improve the robustness of the system, and to reduce the error rate in detection and tracking process, morphological image processing method for binary images is used. A developed blob analysis technique for extracted binary image facilitates pedestrian and car detection. Processing blob´s information of relative size and location leads to distinguishing between pedestrian and car. Applying temporal analysis techniques and moving object detection methods improves system versatility to detect and recognize waiting and moving pedestrian and car. To enhance system robustness to scene changes and reduce error rate an innovative method of remembrance adaptive background update (RABU), is presented. The proposed method scans the scene changes and includes relatively unchanged field to background update. This system is implemented in MATLABreg and Simulinkreg, using video and image processing blockset, image acquisition blockset and image processing toolbox. The proposed system provides continuous visual tracking for traffic intersections with the ability to provide extensive data for traffic control systems. System is tested with both live real-time video and pre-recorded video. The obtained results demonstrate also good performance of the proposed system for pedestrian and car tacking and prove its adaptability to varying weather conditions