Real-time multi-vehicle detection and sub-feature based tracking for traffic surveillance systems

As traffic surveillance technologies continue to grow worldwide, vehicle detection, counting and tracking are becoming increasing important. This paper proposes a real-time multi-vehicle detection and tracking approach. Lane marker detection is carried out for vehicle counting on each lane. It also helps remove the foreground noise and shadow. Instead of tracking the entire vehicle blob, vehicle sub-feature based Kalman filter is used in tracking. By implementing sub-feature tracking, this system is more robust to partial occlusions, which happens a lot in congestions. This approach is scalable to most freeway surveillance video. Several freeway surveillance videos are used to evaluate the performance of the traffic surveillance system. The proposed approach is compared with the standard blob tracking. Test results demonstrated that our approach outperforms blob tracking in correct tracking rate. The performance of our method is also illustrated with different video image sampling rates.