Real-Time Multi-Camera Traffic Intersection Analysis on IoT Devices

The continuous growth of urban and suburban areas has increased the number of commuters and general transportation along central roadways, which has resulted in a trend of more traffic congestion in essential areas every year. This creates a challenge for civil engineers to improve the efficiency of the traffic intersection systems. Our team worked on this issue by building a system to improve traffic intersection analysis by video camera to count cars using multiple camera angles at a single intersection. By providing better real-time traffic network analysis, civil engineers can more accurately monitor the efficiency of traffic intersections and then make better predictions for timing changes in traffic light signaling. The novelty of our project comes from using state of the field machine learning vehicle counting algorithms for a single camera and then modifying them to work across an Internet of Things (IoT) network of devices for multiple camera angles across a single intersection. We achieved fast processing speeds on the Nvidia Jetson NX IoT devices by utilizing quantization of the object detection YOLO algorithms to be able to process video input in real time. Another improvement from current video-camera based solutions is using multiple camera angles in a network to reduce vehicle counting inaccuracy from camera obstruction from a single perspective. For future work, our project can be improved by working on the re-identification of cars when the tracking is lost, and at crowded intersections where exclusive regions of interest for intersection paths are hard to define