Design And Verification Platform Development Of Vehicle-Road Cooperative Over The Horizon Perception For Autonomous Driving Vehicles

Without the ability to perceive surrounding environment,which is mainly manifested as object detection,autonomous driving vehicles cannot operate safely and effectively.However,the high computational complexity of object detection models based on deep learning makes it impossible to deploy detection tasks directly on mobile devices with limited resources and offloading to the cloud will bring a large response delay.Furthermore,the perception layer is essential for object detection technology.But single-vehicle perception will have visual blind spots in many scenarios,which has an impact on the safe driving of autonomous driving vehicles.Based on the two technological challenges,this paper designs a real-time object detection framework based on edge computing to efficiently schedule the computational resources at the roadside,and builds an over-the-horizon perception platform with different air-port bands to expand the range of perception for autonomous driving vehicles.The main contribution of this paper are as follows:(1)Due to the requirements of high-performance and high-speed real-time object detection in the autonomous driving vehicles,we proposes an object detection framework based on mobile edge computing and deploys mobile edge computing server from the cloud to the edge side to provide real-time detection services.The one-stage object detection algorithm YOLOX is the current mainstream algorithm and meets the needs of autonomous driving,but the detection accuracy of the algorithm for small objects in high-density areas is not ideal.To solve this problem,this paper proposes an improved YOLOX model based on the positioning loss function of CIOU Loss.Through comparative experiments on KITTI datasets,the proposed model improves the detection ability of small objects and the detection accuracy is improved by 1.4%and 2.7%respectively compared with the original YOLOX model and Faster RCNN model.(2)Aiming at the problem of single-vehicle perception with blind field of view in autonomous driving scenarios,this paper proposes a vehicle-road cooperative over-the-horizon perception model based on millimeter wave band and Sub-6GHz band,which gives full play to the advantages of large bandwidth of millimeter wave band and wide coverage of Sub-6GHz band.And then this paper builds 28GHz millimeter wave hardware platform and Sub-6G USRP hardware platform to verify the proposed over-the-horizon perception link and uses video transmission services as the basis for performance analysis.Experimental results show that the architecture that combines millimeter wave high band and Sub-6GHz low band can reduce the delay by 29.8%and improve link throughput by 93.2%compared with the deployment mode that only considers the Sub-6GHz low band.