基于轮廓关键点和LSTM的摔倒检测方法  

FALL DETECTION METHOD BASED ON CONTOUR KEYPOINTS AND LSTM

作　　者：卫少洁 周永霞 Wei Shaojie;Zhou Yongxia(College of Information Engineering,China Jiliang University,Hangzhou 310018,Zhejiang,China)

机构地区：[1]中国计量大学信息工程学院,浙江杭州310018

出　　处：《计算机应用与软件》2022年第4期213-217,241,共6页Computer Applications and Software

基　　金：国家自然科学基金项目(61303146);浙江省自然科学基金项目(LY19F030013)。

摘　　要：针对传统人体摔倒检测方法误检率高、无法有效结合时序特征等问题,提出一种新的人体轮廓关键点提取方法,并将该方法和LSTM网络相结合构建一种新的摔倒检测模型。该模型对视频中的人体进行轮廓检测,选取轮廓关键点坐标和质心坐标作为人体特征;使用LSTM对人体特征序列进行时序特征提取;用全连接层实现分类。在公开数据集上进行实验,结果表明该模型具有较高的准确率和良好的泛化性。Traditional human-falling-down detection methods have high error rates and cannot combine sequential features efficiently. To solve these problems, this paper proposes a new key point extraction method of human contour is proposed, and a new fall detection model is constructed by combining this method with LSTM network. This model detected the human contours in the video, selected the contour key point coordinates and centroid coordinates as human features, extracted the sequential features of the human feature sequence using LSTM, and realized the classification with the full connection layer. The experiments on public data sets show that the model has high accuracy and good generalization.

关 键 词：人体摔倒检测 轮廓关键点 LSTM 时序特征 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]