부산대학교 도서관

Vehicle-Infrastructure cooperation is an advanced development stage of autonomous driving, which helps to upgrade the capability of vehicles by fully implementing real-time information interaction among vehicles, roads and pedestrians. However, perception, computing and communication are usually decoupled in today’s vehicle-road coordination applications, which significantly adds delay and cost to the system. In this paper, we propose and implement a platform that integrates perception, computing and communication to provide timely roadside feature maps to vehicles for vision fusion. A neural processing unit (NPU) for computing and a cellular vehicle-to-everything (C-V2X) wireless baseband IP for communication are both implemented on FPGA. We evaluate the effectiveness of the proposed platform using CoBEVT algorithm on the camera track of the OPV2V perception dataset. The experimental result show that our platform can expand the view of vehicles as well as improve information freshness in terms of end-to-end delay.