【摘要】 针对老人、儿童、残障人士群体存在的跌倒风险，提出了一种基于YOLOv5算法的视频图像人体跌倒检测方法。该方法通过CSI摄像头采集视频数据，对视频数据进行训练和验证，进而判断目标人物的动作姿态，连接树莓派的WiFi模块将动作姿态信息发送到监护人手机上，提醒监护人对跌倒人员进行及时救治，从而提高了救援效率。实验结果表明：1)YOLOv5算法对站立和跌倒动作的识别精确度均较高，而对下蹲动作的识别精确度相对较低；2)光线充足的环境中图片的置信度要高于昏暗环境中图片的置信度；3)YOLOv5算法检测人体动作的速率和精确度高于Faster R-CNN算法。更多还原

【Abstract】 Aiming at the fall risk of the elderly, children and the disabled, a video image human fall detection method based on YOLOv5 algorithm is proposed. This method collects video data through the CSI camera, trains and verifies the video data, and then judges the action posture of the target person. The YOLOv5 algorithm connects the WiFi module of Raspberry Pi to send the action posture information to the guardian’s mobile phone, and reminds the guardian to rescue the fall person timely. Thus YOLOv5 algorithm improves the rescue efficiency. The experimental results show that: 1)YOLOv5 algorithm has high recognition accuracy for standing and falling movements, but has relatively low recognition accuracy for squatting movements. 2)The confidence coefficient of pictures in sufficient light environment is higher than that of pictures in dark environment. 3)The speed and accuracy of YOLOv5 algorithm in detecting human motion are higher than that of Faster R-CNN algorithm.更多还原