摘要
泳池溺水事故频发,基于计算机视觉的泳池防溺水技术可以有效预防溺水事故的发生。当前的基于计算机视觉的泳池防溺水技术大多利用目标检测算法研究游泳者的头部、人体与水面的位置关系,人为设置溺水判定规则,主观因素太强,模型参数过多且难以确定,准确率有待提高。该文提出一种基于人体姿态估计的泳池溺水检测算法,利用人体姿态估计模型OpenPose对游泳者的图像进行人体关键点标记,构建人体关键点距离向量集合,通过比较游泳者的关键点距离向量与溺水状态的关键点距离向量的相似性判断溺水行为。实验结果表明,该文提出的基于人体姿态估计的泳池溺水检测算法的准确率为95%,精确率为85%,召回率为89.47%,可以有效地检测出溺水游泳者。
Swimming pool drowning accidents occur frequently.The swimming pool anti-drowning technology based on computer vision can effectively prevent drowning accidents.The current computer vision-based swimming pool anti-drowning technology mostly uses the object detection algorithm to study the position relationship between the swimmer's head and body and the water surface,and sets the drowning decision rules artificially.The subjective factors are too strong,the model parameters are too many and difficult to determine,and the accuracy needs to be improved.In this paper,a swimming pool drowning detection algorithm based on human pose estimation is proposed.Using the human pose estimation model OpenPose,the swimmer's image is marked with human key points,and the human key point distance vector set is constructed.The drowning behavior is judged by comparing the similarity between the swimmer's key point distance vector and the key point distance vector of the drowning state.The experimental results show that the accuracy rate of the swimming pool drowning detection algorithm based on human posture estimation proposed in this paper is 95%,the precision rate is 85%,and the recall rate is 89.47%,which can effectively detect drowning swimmers.
出处
《科技创新与应用》
2023年第23期66-70,74,共6页
Technology Innovation and Application
