Research On Edge-Assisted Cooperative Perception Security Technology Of IoV For Autonomous Driving

Autonomous driving aims to free tedious driving tasks by deploying perception,planning,and control capabilities to vehicles.However,due to the limited perception resources of a single vehicle,with the help of an edge cooperative perception framework,perception data sharing between autonomous vehicles can bring higher perception accuracy,eliminating blind spots of perception.But,in the process of sharing,aggregating and calculating the original perception data,it faces the problem of privacy leakage.First,on the perception side,the vehicle’s perception data usually contains sensitive information such as speed,location,and identity,and the sharing of original data will seriously endanger the privacy and security of vehicles.Secondly,on the aggregation side,when the perception data is aggregated and calculated on the edge server,the server can steal the user’s private information through reconstruction,reverse engineering,reasoning,etc.Finally,on the calculation side,the calculation results of the server may be wrong due to server selfishness or malicious attacks.Since the calculation results also contain private information,how to verify the correctness of results with zero-knowledge is also important question.Therefore,in the cooperative perception process of autonomous driving,how to ensure the privacy and security of the vehicle,realize efficient aggregate calculation,and verify the correctness of the calculation results is of great significance to improve the quality of Autonomous driving service.This thesis will conduct research on the cooperative perception security technology of the Internet of Vehicles for autonomous driving.The specific research contents are as follows:(1)Research on a privacy protection scheme based on federated learning for cooperative perception of Internet of Vehicles.By mapping distributed perceptual data to a data model,sensitive information leakage can be avoided,and the proposed adaptive constrained asynchronous federated learning method can effectively reduce the impact of vehicle departure on learning efficiency.For the perceptual data sharing of matrix structure,a lightweight matrix transformation privacy protection method is designed.The original matrix is added to the additional matrix generated by its simplification to encrypt the matrix,which can be encrypted and decrypted efficiently.(2)Research the edge computing security aggregation scheme based on function encryption and differential privacy.The scheme designs a local adaptive differential privacy method to blind the parameters of the vehicle sharing model,and uses the improved function encryption to encrypt the blinded local model,to achieve the ciphertext aggregation of the shared model.Use blockchain to generate and distribute public parameters,vehicle keys and gradient constraints to ensure the reliability of shared data.(3)Research on the verifiable solution of the edge computing results of the Internet of Vehicles based on the blockchain.The edge server provides a commitment to the correctness of its calculation results,using the publicly verifiable algorithm with integrated pseudo-random functions as the underlying structure of the solution,and using the smart contract on the blockchain to publicly verify the calculation results.While ensuring the privacy and security of the calculation results,it also realizes the verification of the correctness of the server calculation results.