Signal timing detection based on spatial-temporal map generated from CCTV surveillance video

Signal timing data are a critical component in the estimation of arterial traffic states. Estimating arterial travel time, intersection delay, and queue length from either detector or probe data relies on accurate input or detection of signal timing. Moreover, the emerging eco-intersection approaching and departure applications also require reliable signal timing input. With the accelerated deployment of adaptive and smart signal systems, signal timing data have become more and more readily available at arterial management centers. However, of the thousands of intersections around the country, many still do not report real-time signal timing data because of various maintenance, hardware, software, communication, cost, and other problems. Some existing studies have proposed the use of probe trajectories to infer signal timing on the basis of shock wave theory; however, because of the limited sample sizes, coverage and reliability are also limited. A computer vision-based algorithm to detect signal timing from the regular low-resolution CCTV cameras at many major arterial intersections around the United States is proposed in this paper. The algorithm detects vehicle trajectories by evaluating the movement of pixel colors at a predefined scan line on CCTV video footage. By detecting static objects on the scan line and their stalling duration, the proposed algorithm can efficiently detect the starting and ending times of red lights. The algorithm is calibrated and evaluated by using arterial traffic video collected from the intersection of Henderson Road at U.S. Hwy 1 in North Brunswick, New Jersey. The results show the promising performance of the algorithm in becoming a cost-effective signal detection solution.