Important traffic regulation and safety in High populated and fast developing urban areas. Traditional applications of traffic laws involving enforcement of speed violations or illegal directions often leverage upon pricey infrastructure, especially Automatic Number Plate Recognition cameras [14] which capture data about the vehicle and eventually process it. In the following paper there is presenta a new approach - one that is cost efficient and scalable for detect vehicle speed and wrong direction violations using publicly available non-specialized cameras. The methodology in paper uses state of art deep learning object detection models namely YOLO based architectures and advanced computer vision techniques for accurate, real time speed estimation and direction detection of vehicles. It is demonstrated how the proposed system can flag critical traffic violations such as overspeeding and traveling in the wrong direction. The system’s modular design and dependency on general purpose cameras This approach must allow for its widespread, affordable implementation. Experimental results show us the robustness, accuracy, and real time capabilities of the proposed approach as well as insight into practical deployment in real world traffic surveillance applications.

Computer Vision Traffic Surveillance YOLO Vehicle Speed Detection Direction Detection Non-ANPR Cameras Frames Centroid Speed Estimation Euclidean Distance Bounding Box