<p>Accurate and rapid vehicle information detection is imperative for bridge health condition assessments. A lightweight computer vision-based vehicle information detection method is proposed in this study based on semantic segmentation and two-branch neural networks. Initially, a vehicle image segmentation dataset consisting of 4 categories is constructed, and the vehicle 2D bounding box and mask detector is obtained using an improved semantic segmentation network to identify the number and type of vehicles. Subsequently, a multi-layer aggregation network is utilized to achieve efficient image feature extraction. A two-branch neural network is established to decompose the 3D detection of vehicle objects into two subtasks: key point classification and 3D bounding box regression, and ultimately establishing the 3D bounding box of vehicles with monocular 3D vision. Finally, the coordinate transformation between the world and the pixel coordinate system is achieved based on the control points arranged on the bridge deck and Efficient Perspective-n-Point, enabling the accurate measurement of vehicle dimensional information. Furthermore, the accuracy and feasibility of the proposed method are verified with the Ji Canal Bridge. Comparative studies are conducted to examine the performance of the proposed method, and the results demonstrate that the proposed method shows better performance in accuracy and efficiency.</p>