摘要： 为保持行人在新型冠状病毒肺炎（COVID-19）疫情下的安全社交距离，有效控制和预防疫情传播，构建一种基于YOLOv4的安全社交距离风险评估模型。利用微调后的YOLOv4算法对行人进行目标提取，获取行人关键点，并将行人连续运动视为质点的连续运动，结合DeepSort算法实现对行人的跟踪处理。在此基础上，建立视觉坐标系，在鸟瞰视角下提出运动矢量分析算法计算和判断行人运动方向并评估行人的安全社交距离。在牛津城市中心的数据集上评估模型有效性，实验结果表明，微调后YOLOv4算法在行人检测中平均精度均值达到90.33%，行人社交距离风险评估准确率达到88.23%，性能优于Fast R-CNN、Faster R-CNN、YOLOv3和YOLOv4算法，表明所提模型能够有效提升安全社交距离的检测准确性。

关键词: YOLO网络, 安全社交距离, 新型冠状病毒肺炎, 目标检测, 运动矢量分析

Abstract: In order to maintain a safety social distance in the COVID-19 pandemic, control and prevent pandemic transmission, a risk assessment model of safety social distance based on YOLOv4 is constructed.The fine-tuned YOLOv4 algorithm extracts the pedestrian target, obtains their key points, and regards their continuous movement as a continuous movement of particles.When combined with the DeepSort algorithm, it can be used to track pedestrians.On this basis, a visual coordinate system is established, and a motion vector analysis algorithm is proposed to calculate and judge the direction of pedestrian movement and evaluate the safety social distance of pedestrians from an aerial perspective.The effectiveness of the model is evaluated on the Oxford City Center Dataset.The experimental results show that the mean Average Precision(mAP) of the fine-tuned YOLOv4 algorithm in pedestrian detection reaches 90.33%, and the accuracy of pedestrian social distance risk assessment reaches 88.23%.These values are superior to those of the Fast R-CNN, Faster R-CNN, YOLOv3, and YOLOv4 algorithms.Thus, the proposed model effectively improves the detection accuracy of safe social distance.

Key words: YOLO network, safety social distance, Corona Virus Disease 2019(COVID-19), object detection, motion vector analysis

中图分类号: 

• TP391.41