【摘要】 严禁电动车违规载人是新交通法规中的重要内容，针对目前缺乏有效的检测电动车违规载人算法的现状，设计了一种基于改进的YOLOv5目标检测算法与边缘设备相结合的电动车违规载人检测系统。首先构建电动车行驶数据集；其次以YOLOv5网络模型为基础，引入轻量化网络Mobilenetv3、ECA-Net注意力机制、Slim-Neck结构和SPPFCSPC空间金字塔池化结构，提升针对电动车违规载人的检测精度，并且与原算法做消融实验；最后将改进后的算法部署在边缘设备Jetson Nano上进行实时推理。通过分析实验数据，改进后算法的参数量下降为原YOLOv5n的18%，在Jetson Nano上其推理速度提升了62%，最快推理速度可以达到17 FPS。改进后的算法在Jetson Nano上可以在提升检测精度的同时大幅提高推理速度，满足在不同场景下进行边缘部署的需求。更多还原

【Abstract】 It is an important content in the new traffic laws and regulations to prohibit the illegal carrying of people by electric vehicles. In view of the current lack of effective algorithms to detect the illegal carrying of people by electric vehicles, a detection system for illegal carrying of people by electric vehicles is designed based on the combination of improved YOLOv5 target detection algorithm and edge equipment. Firstly, the data set of electric vehicle driving is constructed; Secondly, on the basis of YOLOv5 network model, the lightweight network Mobilenetv3, ECA Net attention mechanism, Slim Neck structure and SPPFCSPC spatial pyramid pool structure are introduced to improve the detection accuracy for illegal passenger carrying of electric vehicles, and ablation experiments are conducted with the original algorithm. Finally, the improved algorithm is deployed on the edge device Jetson Nano for real-time reasoning. By analyzing the experimental data, the parameters of the improved algorithm are reduced to 18% of the original YOLOv5n, and the reasoning speed on the Jetson Nano is increased by 62%, with the fastest reasoning speed reaching17FPS. The improved algorithm can greatly improve the reasoning speed while improving the detection accuracy on the Jetson Nano, and meet the needs of edge deployment in different scenarios.更多还原