基于BlazePose和随机森林算法的异常步态检测
Abnormal Gait Detection Based on BlazePose and Random Forest
黄灶荣 1王春宝 2韦建军 1余学书 1谭啸海1
作者信息
- 1. 广西科技大学,广西 柳州 545616
- 2. 广西科技大学,广西 柳州 545616;广东铭凯医疗机器人有限公司,广东 珠海 519075
- 折叠
摘要
异常步态对行动能力产生严重影响,因此,及时、自动地检测异常步态具有至关重要的意义.本文提出了一种基于 BlazePose 和随机森林算法的人体异常步态检测方法.先利用 BlazePose 算法提取RGB视频中的人体骨骼关键点,然后通过数据处理获取 7 个关键的步态特征参数.最后采用随机森林算法作为步态分类器,用于区分正常步态与异常步态.利用 142 例异常步态数据和 257 例正常步态数据对分类器进行训练和测试评估,实验结果显示准确率和召回率分别达到 97.5%和 90%,表明该方法在异常步态检测方面具备一定的可行性和实用价值.
Abstract
Abnormal gait significantly affects mobility,making timely and automatic detection of abnormal gait critically important.This study proposes a human abnormal gait detection method based on BlazePose and the random forest algo-rithm.Firstly,the BlazePose algorithm is utilized to extract skeletal keypoints of the human body from RGB videos.Next,seven key gait feature parameters are obtained through data processing.Finally,the random forest algorithm is employed as the gait classifier to differentiate between normal and abnormal gaits.The classifier is trained and evaluated using 142 cases of abnormal gait data and 257 cases of normal gait data.Experimental results show an accuracy of 97.5%and a recall rate of 90%,indicating that the proposed method exhibits certain feasibility and practical value in the field of abnormal gait detec-tion.
关键词
BlazePose/随机森林/异常步态检测/数据处理Key words
BlazePose/random forest/abnormal gait detection/data processing引用本文复制引用
出版年
2024
国家科技期刊平台
NSTL
万方数据