Object Detection and Localization Based on Binocular Vision for Autonomous Vehicles

Abstract

Environment perception based on vision plays an important role in autonomous driving technology. Although vision perception has achieved certain results in recent years, many methods can not solve the contradiction between speed and precision. In this paper, we propose a system for fast and accurate object detection and localization based on binocular vision. For object detection, a neural network model based on YOLOv3 is proposed. Specifically, MobileNet is employed in the backbone of YOLOv3 to improve the speed of feature extraction. Then the corresponding ORB feature points are extracted from continuous stereo images which take from the binocular cameras on the moving car. Thus, the disparity of each ORB feature point is calculated. After that, we use the result of object detection to screen the ORB feature points. Finally, the depth of the targets in the traffic scene can be estimated. Experiments on the KITTI dataset show the efficacy of our system, as well as the accuracy and robustness of our object localization relative to ground truth and prior works.