Edge-Assisted Privacy-Preserving Raw Data Sharing Framework for Connected Autonomous Vehicles
Jinbo Xiong, Renwan Bi, Mingfeng Zhao, Jingda Guo, Qing Yang
Abstract
Data sharing among connected and autonomous vehicles without any protection will cause private information leakage. Simply encrypting data introduces a heavy overhead; most importantly, when encrypted data (ciphertext) is decrypted on a vehicle, the receiver will be fully aware of the sender's data, implying potential data leakage. To tackle these issues, we propose an edge-assisted privacy-preserving raw data sharing framework. First, we leverage the additive secret sharing technique to encrypt raw data into two ciphertexts and construct two classes of secure functions. The functions are then used to implement a privacy-preserving convolutional neural network (P-CNN). Finally, two edge servers are deployed to cooperatively execute P-CNN to extract features from two ciphertexts to obtain the same object detection results as the original CNN. We adopt the VGG16 model as a case study to illustrate how to construct P-CNN and employ the KITTI dataset to verify our solution. Experiment results demonstrate that P-CNN offers exactly the same classification results as the VGG16 model with negligible error, and the communication overhead and computational cost on the edge servers are less than existing solutions without leaking private information.