Vehicle Dimensional Measurement with RGB-D Camera

Vehicle dimensional measurement is crucial for road over-limit detection. As existing methods can not balance price, accuracy, and time complexity, we propose a low-cost, accurate, and real-time method for measuring vehicle dimensions using an RGB-D camera priced at 499 dollars. The proposed method builds the ground background model from a series of video frames and uses ground point cloud and lane information to self-calibrate the camera's extrinsic parameters. Subsequently, using the self-calibrated results, the vehicle point cloud in the current frame is transformed into the measurement coordinate system to facilitate accurate measurement. Finally, we measure the dimensions of the entire vehicle by combining the vehicle displacement, local width, and local height across multiple frames. By evaluating three different types of vehicles, the results show that the average error is less than 3.3% for length and less than 2% for width and height, indicating high accuracy. Furthermore, our method achieves real-time performance with each frame processed in less than 30 milliseconds.