汽车驾驶的安全问题一直都是人们关注的问题,车辆搭载行人盲区检测系统可以有效地降低行人碰撞风险。现有的车载盲区行人检测方案往往功耗大且成本高,云端处理方案又难以满足延迟敏感的车端检测的需求。针对上述问题,采用一款功耗和成本较低的边缘计算平台K230,结合YOLOv8算法,通过WiderPerson行人数据集完成训练,最终通过量化部署实现了一个基于摄像头的车载盲区行人智能检测系统。实验结果表明,在道路行人较多、存在遮挡的情况下,系统也能准确识别到行人目标并发出安全预警。

The safety of vehicle driving has always been a concern of people. Vehicle-mounted pedestrian blind spot detection systems can effectively reduce the risk of pedestrian collisions. The existing in-vehicle blind spot pedestrian detection solutions often suffer from high power consumption and costs, while cloud-based processing solutions struggle to meet the latency-sensitive requirements of vehicle-side detection. To address these issues, this paper employs a low-power and low-cost edge computing platform K230, combined with the YOLOv8 algorithm, and completes training using the WiderPerson pedestrian dataset. Finally, a camera-based intelligent vehicle blind spot pedestrian detection system is implemented through quantization deployment. Experimental results demonstrate that the system can accurately identify pedestrian targets and issue safety warnings even in scenarios with dense road pedestrians and occlusions.