Automatic Vehicle Classification and Speed Tracking

The current traffic in large cities and urban areas has grown significantly, so a surveillance system is required to monitor traffic and avoid unnecessary delays and accidents. In this research study, computer vision-based speed estimation and object detection have been implemented for various automatic vehicles. Various image processing and deep learning-based methods and models have been used to test the proposed system. An open-source image dataset of five automobiles, car, bus, bike, truck, and local four-wheeler (CNG), has been utilized in this work. These 3,293 total images have been annotated with Roboflow framework and trained with YOLOv4, YOLOv5, YOLOv7, deep learning models, and the Haar cascade method. Average mAP scores of 0.956, 0.857 and 0.821 have been obtained for YOLOv5, YOLOv4 and YOLOv7 models, respectively, for different categories of vehicles. YOLOv4 and the Haar cascade methods have been employed to estimate the speed of the detected vehicles. The YOLOv4 technique performed best in speed assessment of various automobiles.