Edge-Assisted Low-Latency Object Detection for Networked Vehicles Using Point Cloud

In this paper, with the objective of enhancing the perception quality for autonomous vehicles based on point cloud data, we design an edge-assisted on-road perception framework. This framework contains three modules: adaptive point cloud transmission, driving environment complexity evaluation, and perception task scheduling. In specific, to suppress the transmission delay posed by fluctuating wireless links, we finely configure the quality of point cloud data to be transmitted with sampling and compression fidelity tuning. Then, taking the driving environment complexity into consideration, we propose perception score, a novel perception quality evaluation metric, according to which the edge server schedules the received tasks. Furthermore, an overall perception score maximization problem is formulated to obtain the optimal scheduling strategy at the edge. At last, a greedy-based time window constraint algorithm is designed, which can solve the problem at low computational complexity. Extensive experiments show that, compared to other benchmarks, our algorithm exhibits significant advantages in precise and low-latency object detection for autonomous vehicles.