Air-Ground Integrated Mobile Edge Computing in Vehicular Visual Sensor Networks

In the vehicular visual sensor networks, multiple vehicles with on-board cameras capture visual observations of the environment while moving along their trajectories and the 3-D map is jointly estimated to provide guidance for real-time navigation. Due to high mobility of vehicles and heavy computation burden of 3-D map estimation, the cooperative vehicular visual sensing is latency-critical and computation-intensive, and thus suitable to be solved by edge computing. This article designs an air-ground integrated mobile edge computing (MEC) framework for the vehicular visual sensor networks, in which parallel computation is implemented by the vehicles locally and also partially offloaded to the edge server on the unmanned aerial vehicle (UAV), and our aim is to minimize the total latency of the network through the optimization of the UAV trajectory and the offloading ratios of vehicles. To characterize the motion, sensing, computation, and communication processes in the network, the timeline includes two periods, with the motion period for motion and sensing and the service period for computation and communication, where the service period is divided into several stages with the latency models for computation or communication in each stage. Then we propose a dynamic alternate trajectory design and partial offloading scheduling algorithm for the air-ground integrated MEC in the vehicular visual sensor networks. Numerical results verify the effectiveness of our scheme in terms of latency and also show the impacts of computation burden, computation capability, and communication condition on performance.