Vehicle Direction Detection Based on YOLOv3

With the popularity of traffic monitoring systems and traffic intelligence management, a large number of fixed and mobile cameras are installed in the city. However, manual monitoring of the camera is not efficient. The target detection method can monitor the screen and analyze the direction in which the vehicle drives and parks, helping to reduce the labor burden and improve efficiency. In this paper, we add a fine classification network after YOLOv3 to detect and discriminate the vehicle's driving and parking directions. YOLOv3 detects and extracts vehicle location information, and from which the latter network is responsible for finely classifying directions. We achieved a 91% classification accuracy on a test database of 400 images. This modified system can increase the practicality of the YOLOv3 target detection in the traffic scene.