面向交通相机的轻量化车辆三维形态检测算法OA

Accurately acquiring three-dimensional vehicle information is essential for the environmental perception and path planning in cooperative vehicle-infrastructure systems.Roadside cameras,however,are influenced by perspective distortion,making it challenging to directly capture the three-dimensional information of vehicles.Moreover,insufficient geometric con-straints in existing methods reduce detection accuracy.Additionally,current algorithms often suffer from slow processing speeds and fail to meet real-time requirements.To address these issues,this paper proposes a lightweight vehicle 3D shape detection al-gorithm based on geometric constraints.Firstly,a spatial calibration model for roadside cameras is established within road scenes to obtain the two-dimensional-to-three-dimensional mapping matrix and corresponding scale information of perspective space.Next,the YOLOv8 model is employed for two-dimensional vehicle detection.By combining vehicle geometric features with the calibration results,a nonlinear constraint function is constructed to accurately detect the three-dimensional shape of vehicles.The algorithm's performance is evaluated using the UA-DETRAC dataset and real-world highway scenes.Experimental results demonstrate that the average three-dimensional detection accuracy reaches 91.04%,and the processing speed is 37.7%faster than that of existing roadside vehicle 3D detection methods..The algorithm significantly improves detection speed while maintain-ing high accuracy,meeting the demands of real-time applications.