首页|期刊导航|数码设计|基于深度学习的人体行为识别

基于深度学习的人体行为识别OA

Human Behavior Recognition Based on Deep Learning

中文摘要英文摘要

针对输入对象是视频的人体行为识别,常用的深度学习算法不能很好地解决数据时序问题和连续性的问题,提出了一种基于OpenPose的Bi-LSTM模型进行人体行为识别.首先选定实验数据集,然后采用OpenPose来处理数据集,提取出人体的18个骨骼关节点,对数据进行标注、类别标签对应,再用Bi-LSTM对人体行为进行分类.同时为了验证其有效性,将KNN、SVM与其进行对比分析.实验结果表明,该模型不仅能提升系统的识别性能,而且能有效地解决视频时序的问题,得出本文所提出的算法具有较高的识别准确率,达到0.96,并且应用性更广泛的结论.

A Bi-LSTM model based on OpenPose was proposed for human behavior recognition when the input object was a video.Common deep learning algorithms cannot effectively solve the problems of data timing and continuity.Firstly,the experimental dataset was selected,and then OpenPose was used to process the dataset,extracting 18 skeletal joint points of the human body.The data was annotated and labeled accordingly.Bi-LSTM was then used to classify human behavior.To verify its effectiveness,KNN and SVM are compared and analyzed.The experimental results showed that the model could not only improve the recognition performance of the system,but also effectively solved the problem of video timing.The algorithm proposed in this paper has a high recognition accuracy,reaching 0.96,and had a wider range of applications.

景云云

渭南师范学院 物理与电气工程学院,陕西渭南 714000

信息技术与安全科学

深度学习长短时记忆网络OpenPose人体行为识别

deep learninglong short-term memory networksOpenPosehuman action recognition

《数码设计》 2025 (10)

43-45,3

评论