Speed detection plays a crucial role in ensuring the safe operation of vehicles in urban transportation systems, making it essential for maintaining traffic safety. However, existing methods for measuring vehicle speed suffer from high costs, susceptibility to external conditions, and limitations in installation areas. To address these issues, this paper proposes a low-cost and flexible vehicle recognition and speed measurement method based on video imagery. The approach utilizes deep learning techniques to construct the YOLOv4 framework and train it on the COCO dataset for vehicle identification. The recognition method is improved by extracting the pixel coordinates of the midpoint of the lower boundary of the maximum bounding rectangle encompassing the recognized vehicles. Additionally, a close-range photogrammetry method is introduced, and improvements are made to the collinearity equations to enable vehicle recognition in a single-camera setup. The displacement of vehicles is computed within a fixed time interval, and a velocity curve of vehicles within the monitoring area is plotted. Experimental validation is conducted to assess the feasibility and accuracy of the proposed method.
维普
万方数据