Optimization and Estimation Algorithm of Vehicle Spatial Form Based on Monocular Traffic Camera

Abstract

Accurate and rapid acquisition of vehicle spatial form is of great importance in the fields of intelligent transportation and autonomous driving. However, given the limitations of projective geometry, it is difficult to obtain the 3-D structure of vehicles using monocular cameras. The purpose of this paper is, therefore, to estimate the vehicle spatial form using monocular traffic cameras. Firstly, we establish the camera calibration model of the road scene, and jointly construct the geometric constraint model of the vehicle spatial form by vanishing points. Secondly, the contour and edge constraints of the vehicle are obtained based on Mask R-CNN. Then, based on these constraints, the error constraint function is constructed to calculate the projection error of the vehicle spatial form. Finally, a particle swarm optimization algorithm is used to iteratively optimize the parameters in the constraint space to obtain accurate vehicle spatial form information. Experiments are carried on the BrnoCompSpeed data set and the home-made data set. The experimental results show that the processing time of a single frame is less than 0.5 s and the average accuracy is higher than 94%. Moreover, the proposed algorithm has good robustness to the issue of vehicle occlusion and queuing in the scene, which outperforms existing methods.