Content-driven Joint Resource Allocation Based on Vehicle-edge Synergy in Vehicular Networks

With the rapid growth of computer vision applications, a large amount of video data in vehicular networks are used for content analysis. As a representative technology in the field of artificial intelligence, deep learning has been widely used in video content understanding. However, deep learning models are usually accompanied by huge amount of calculation, which put great pressure on traditional wireless communication resource and Mobile Edge Computing (MEC) server. To solve this problem, in this paper, we first propose a DNN model segmentation strategy, and then propose a content-driven joint resource allocation scheme based on vehicle-edge synergy to maximize the accuracy of video content understanding under the constraints of communication and computing resources. Due to the real-time nature of resource allocation and the variability of the environment in vehicular networks, we design a Multi-agent Distributed Q-Learning algorithm to solve such multi-constrained nonlinear programming problems. Finally, the simulation results show that our proposed vehicle-edge synergy based joint resource allocation scheme has better video content understanding performance and system delay.