LAUSR

In this paper, we propose a novel stereovision-based system for vehicle speed measurement. Our system is set in a fixed location to capture two view stereo videos for passing vehicles by using a calibrated binocular stereovision system. An important feature of this system is the use of an optimized single shot multibox detector network that can efficiently detect license plates in the captured two view stereo videos. Then, the system performs only on the detected license plate area to extract stereo matching point pairs rapidly. The world coordinates of 3D points corresponding to each stereo matching point pair are calculated. Abnormal 3D points are removed in accordance with distance distribution. The remaining 3D point closest to the license plate center is finally selected as the exact vehicle location of the current frame. Vehicle speed is measured by dividing the distance between two selected 3D points by frame intervals. Vehicle trajectory is depicted by connecting multiple selected 3D points. Our proposed system performs non-intrusive and stealth measurement in intelligent traffic surveillance and overcomes the challenges of speed measurement for multiple vehicles on multiple lanes in different motions simultaneously. Experimental results demonstrate its accuracy. Our system has a speed measurement error range of [−1.6, +1.1] km/h and a maximum error rate of 3.80%, which are within the [−3, +2] km/h error limit of several countries’ regulatory authorities and the 6% error rate limit of China national standard GB/T 21255-2007.