基于骨骼数据特征的人体行为识别方法综述

Survey of Human Action Recognition Methods Based on Skeleton DataFeatures

人体行为识别是人工智能领域的一个研究热点,相对于视频、运动流等数据,人体骨骼数据具有简洁性和矢量计算的高效性。从基于传统机器学习的手工特征提取方法和基于深度学习的深度特征提取方法两方面对基于骨骼数据的人体行为识别相关研究进行综述。将手工特征概括为物理属性特征和统计属性特征,将深度特征按卷积神经网络、循环神经网络、图卷积神经网络及混合神经网络等类别对基于骨骼数据的人体行为识别方法及其优缺点进行逐一阐述,并对相关的特征提取方法、技术路线、模型特点及识别率等进行分析与总结。

Human action recognition is a research hotspot in the field of artificial intelligence,compared with video,motion stream data,hu⁃man skeleton data has simplicity and high efficiency of vector calculation.The related works of human action recognition based on skeleton da⁃ta are reviewed from two aspects:manual feature extraction method based on traditional machine learning and deep feature extraction method based on deep learning.The manual features are summarized as physical attribute features and statistical attribute features,according to the depth feature,the human action recognition method based on skeleton data and its advantages and disadvantages are described one by one ac⁃cording to convolution neural network,recurrent neural network,graph convolution network and hybrid neural network.The feature extraction method,technical route,model characteristics and recognition rate are analyzed and summarized in tabular form.