Efficient and secure pedestrian detection in intelligent vehicles based on federated learning

Nowadays, various applications are designed for intelligent vehicles, especially for self-driving vehicles. The cloud server usually gathers data from multiple vehicles in order to obtain accurate model, but the data transmission is vulnerable to attack and the vehicles face a risk of privacy leakage. How to achieve secure multi-party computation in vehicular scenarios is an important issue. In this paper, we propose an efficient and secure pedestrian detection scheme based on federated learning, named PFL. After training the deep neural network with local images, each vehicle, as a party, transmits the encrypted gradients rather than the original data to the server. Homomorphic encryption guarantees that the server conducts the federated averaging without decryption. After the updated global model is distributed from server to vehicles, each vehicle decrypts and updates local model. Experiments show that PFL achieves similar pedestrian detection accuracy to the centralized method, while providing privacy protection.