Beyond-line-of-sight Perception Enhancement via Information Interaction in Connected Autonomous Driving Environment

Abstract

On account of occlusion and limited visual range, the independent perception of the single vehicle is restricted, which cannot meet the requirements of high-level autonomous driving. In view of the characteristics of information interaction in connected environment, a vehicle-vehicle based beyond-line-of-sight fusion perception framework is proposed. Effective data fusion of multi-source heterogeneous sensor is realized based on D-S evidence theory. Precise object detection and recognition is achieved based on lightweight object detection Faster R-CNN algorithm with backbone used MobileNetV2. Finally, the beyond-line-of-sight perception enhancement method in typical scenes is verified and analyzed on Prescan. Results show that the presented method helps autonomous vehicles make full use of sensory data effectively, expand perception scope, avoid blind fields, which plays a supporting role in the safe and efficient operation of autonomous vehicles.