Automated Traffic Volume Analytics at Road Intersections Using Computer Vision Techniques

As the hardware cost decreases, more and more cameras have been deployed and used to monitor traffics. Widely deployed cameras enable a wide range of computer vision-based applications for traffic analytics. In this work, we propose a computer vision-based intelligent system which can analysis traffic at road interactions. Our system leverages existing traffic monitoring cameras and applies computer vision techniques to provide detailed traffic analysis results. To achieve these goals, we develop a deep learning object detection model based on Single Shot MultiBox Detector (SSD). Our approach is able to detect and track vehicles, pedestrians, traffic signs and other related-objects on the road. We build a robust model to analysis the detected objects, our system could estimate traffic volume and further infer traffic congestion, traffic rule violation and so on. To evaluate the proposed model, we use video data collected from multiple cameras at several intersections in real-world environments. The evaluation results show that our system is able to provide reliable and accurate traffic analysis results in real time. In addition, we also tested our system under different light conditions, and results show that our system could achieve similar accuracies under different light conditions.