Design And Implementation Of Intelligent Warning System Based On Edge Computing In Connected Vehicles

With the complexity of road traffic system increasing,driving safety has become a social problem that cannot be ignored.In recent years,academia and industry have proposed lots of driving assistance schemes in the intelligent connected vehicle field.Among them,the key technologies such as multi-sensor data fusion,visual perception and cooperative driving are widely used in driving behavior understanding and environment perception.Meanwhile,the high-performance computing unit such as Graphics Processing Unit(GPU)provides powerful hardware support for large-scale data computing in the Internet of Vehicles(IoV)scenarios.In addition,Deep Neural Networks(DNNs),especially Convolutional Neural Networks(CNNs),apply multi-layer linear and nonlinear functions to extract image features,which play important roles in object detection and scene understanding of the vehicle's surroundings.However,DNNs have the characteristics of multi-layer network,huge structure and complex convolution computing.The model operation is accompanied by a large amount of consumption and occupation of computing and storage resources.The computing resources of the on-board terminals cannot meet the computational requirements of such computation-intensive and memory-intensive tasks,thus the accuracy and real-time performance required by intelligent driving assistance system in traffic scenario cannot be achieved.Considering the computation limitations of the intelligent driving system for single vehicle,many studies have been devoted to proposing vehicle-cloud based collaborative intelligent driving schemes for connected vehicles.However,the communication between the vehicle and the remote centralized cloud server leads to huge communication latencies,and the large-scale access of vehicles brings large computation pressure to cloud.Based on the above research background,this thesis proposes an intelligent warning system based on edge computing for connected vehicles.In this thesis,a cooperative computing platform is constructed by connecting vehicles to the edge computing server which is closer to the user side.The system realizes the cooperative processing of the video data in the vehicle driving scene via the communication of connected vehicles,and provides the drivers with three warning functions:vehicle collision warning,lane keep warning and vulnerable traffic participants warning according to the computing results.This cooperative computing scheme makes full use of the idle resources of the on-board terminal,thereby reducing the communication pressure between the vehicle and edge-server.Thus,the real-time and intelligent performance of driving assistance system based on computer vision are achieved.In this thesis,a high-precision detection model and a low-latency detection model with different detection precisions and speeds are trained on a special dataset,which can be selected according to the complexity of the driving scene and network transmission condition,in order to realize the real-time detection of cars,buses,pedestrians,bicycles and motorbikes.The vehicle-edge server based cooperative warning system proposed in this thesis has large-scale driving data transmission scenarios,and the challenges of information security and privacy protection caused by a large number of data interactions cannot be ignored.Aiming at the structural characteristics of this cooperative warning system,this thesis implements a scheme based on RSA asymmetric encryption,Advanced Encryption Standard(AES)and Elliptic Curve Digital Signature Algorithm(ECDSA),used to protect the information and privacy security of connected vehicles communication system.Then the performance of this system are evaluated.The experiment results show that the proposed system meets the real-time requirement for the driving scenarios of intelligent connected vehicles.