OpenPose结合BP网络的行为检测方法

Behavior Detection Method Based on OpenPose and BP Network

采用OpenPose与BP网络相结合的方法对人体整体或局部行为进行分类检测,首先利用人体姿态估计算法获得人体骨架节点坐标数据,然后利用BP分类网络对节点坐标数据进行迭代训练与学习。检测不同整体行为的分类模型,训练准确率达100%,网络损失仅为0.091,实测各类准确率及总体准确率均达100%;检测局部行为的分类模型,训练准确率亦达100%,网络损失小于10-6,实测各类准确率及总体准确率均达100%;OpenPose与BP网络相结合的方法不仅可以实现不同整体行为或局部行为快速、准确的分类检测任务,同时还克服了传统行为检测方法的不足,能够实现更高效、更准确、更快速的分类检测。

OpenPose and BP network is used to classify and detect the whole or local human behavior.Firstly,the coordinate data of human skeleton nodes are obtained by human pose estimation algorithm,and then the coordinate data of nodes are iteratively trained and learned by BP classification network.The accuracy of training models for detecting different overall behaviors is 100%,the network loss is only 0.091,and the accuracy of each category and the overall accuracy are 100%.The accuracy of training models for detecting local behaviors is 100%,and the network loss is less than 10-6.The accuracy of each category and the overall accuracy are 100%.The combination of OpenPose and BP network not only achieves fast and accurate classification and detection tasks for different global or local behaviors,but also overcomes the shortcomings of traditional behavior detection methods,and achieves more efficient,more accurate and faster classification and detection.