基于Kinect的人体非对称步态识别及骨骼关键点位置重要性分析  

Recognition of Human Asymmetric Gait Based on Kinect and Analysis of Importance of Key Skeletal Points

作　　者：蒙新兴 熊启亮 MENG Xinxing;XIONG Qiliang

机构地区：[1]广西玉林市第二人民医院设备科,广西玉林537000 [2]南昌航空大学仪器科学与光电工程学院,南昌330063

出　　处：《科技创新与应用》2025年第6期94-97,101,共5页Technology Innovation and Application

基　　金：国家自然科学基金(32460238);江西省自然科学基金(20232BAB206134)。

摘　　要：首先利用Kinect采集10名受试者分别模拟正常步态及4种非对称步态时的全身骨骼关键点数据,依次对数据进行滤波去噪、标准化等预处理过程,然后利用循环神经网络模型对上述5种步态进行分类,并通过平均精确率指标评价输入不同数量和不同位置骨骼关键点时的识别效果。结果表明,当输入25个骨骼关键点信息时,模型的识别平均精确率为98.8%;当输入相同骨骼关键点数量(8个和4个)时,下肢骨骼关键点对于人体非对称步态的识别重要性大于上肢骨骼关键点。当仅输入2个骨骼关键点时,上肢骨骼关键点对于非对称步态的识别重要性大于下肢骨骼关键点。上述研究结果可以为基于Kinect的人体步态对称性分析,尤其是采集方案设计提供一定的参考。This study initially employed Kinect to capture full-body skeletal keypoint data from 10 subjects simulating both normal gait and four types of asymmetric gaits.The data underwent preprocessing steps such as filtering and normalization.Subsequently,a recurrent neural network(RNN)model was utilized to classify the aforementioned five types of gaits.The classification performance was evaluated using the average precision metric under varying numbers and locations of skeletal keypoints.Results indicate that with 25 skeletal keypoints as input,the model achieved an average precision of 98.8%in gait recognition.Moreover,for the same number of keypoints(8 and 4),lower limb keypoints were more crucial than upper limb keypoints in identifying human asymmetric gaits.Conversely,when using only 2 keypoints,upper limb keypoints were more significant for gait recognition than lower limb keypoints.These findings contribute insights into Kinect-based analysis of human gait symmetry,particularly in the design of data acquisition protocols.

关 键 词：KINECT 步态对称性 骨骼关键点 深度学习 数据处理 

分 类 号：R318[医药卫生—生物医学工程]