A Method of Vehicle Peripheral Objects Detection Based on Binocular Vision

Aiming at the collision problem that may occur when the passenger opens the door after the vehicle stops, this paper proposes an obj ect detection algorithm around the vehicle based on binocular vision. Due to the high computational complexity of binocular registration, the stereo correction of the image is realized by combining the dual-target fixed parameters and Bouguet Reprojection Algorithm, so that the binocular image is corrected into coplanar line alignment, and the operational efficiency is improved. The You Only Look Once algorithm is used to realize the recognition of people and vehicles on the road. In order to reduce the probability of missing detection and wrong detection of the algorithm in the remote scene, the up-sampling fusion method of the feature pyramid is adopted to detect the target on three feature layers, so as to realize the real-time detection and tracking of the target. Then, Scale-invariant Feature Transform Algorithm is used to obtain the matching points in the target area, and the distance between the feature points in the image and the camera is measured through coordinate transformation. At last, the system is tested at National Supervision &Inspection Centre for Inside Vehicle. The accuracy of the proposed algorithm for measuring the distance of various types of targets is above 96%.