R-MDDQN: A V2V-Based Secure Computation Offloading Algorithm for Video Analytics in Vehicle Edge Networks

Real-time analytics on video data requires substantial computational resources and high energy consumption, and computational offloading has emerged as a promising solution to support such resource-intensive services. However, most of the existing research fails to consider the collaboration between video quality and security optimization in wireless offloading, making it less efficient in real-world scenarios with diverse requirements. To address these challenges, we propose a novel V2V-based secure computation offloading algorithm for video analysis, called radial basis function (RBF)-based multi-objective double deep Q-network (R-MDDQN). We use Wyner's wiretap coding scheme to obtain the achievable secrecy capacity and ensure that video data cannot be decoded by eavesdroppers. To address the trade-off problem among multiple objectives, we employ an RBF weight network to dynamically adjust the weights by learning the variations of different objective values. Each DDQN agent receives reward evaluations based on different objective functions and effectively and dynamically approximates the optimal offloading strategy. Extensive experiments conducted using real-world datasets from Shenzhen demonstrate that the proposed R-MDDQN reduces latency by approximately 9.32$\%$, decreases energy consumption by around 7.3$\%$, improves video analysis accuracy by about 18.9$\%$, and enhances security capacity by roughly 14.8$\%$ compared to existing task offloading schemes.