Multi-Device Adaptive Computing Offloading for Real-Time Video Analytics in MEC-Enabled Vehicular Networks

Real-time video analytics play a vital role in intelligent driving, where achieving high reliability and low delay presents significant challenges. Mobile Edge Computing (MEC) has emerged as a promising solution, allowing video analytic tasks to be offloaded to edge servers via vehicular networks. However, in the context of video analytics in MEC-based vehicular networks, tackling the challenges posed by dynamic network and video scenarios, as well as the disorderly resource competition among multiple devices, has become an urgent and complex issue. In this paper, we present MAOVA, a multi-device adaptive computing offloading scheme for real-time video analytics in MEC-enabled vehicular networks, aiming to enhance the accuracy of real-time video analytics while ensuring acceptable task execution delay. MAOVA adopts a distributed decision-making framework and integrates a Vickrey-Clarke-Groves (VCG) based auction mechanism to facilitate efficient resource allocation among devices. Additionally, Model Predictive Control (MPC) is employed to achieve long-term offloading optimization by dynamically adjusting resolution bidding in response to fluctuating environments. Our experimental evaluations confirm that MAOVA outperforms the existing schemes, providing enhanced utility, enhanced accuracy, and reduced delay, marking a significant advancement in video analytics.