基于双流骨架信息的人体动作识别方法

Human Action Recognition Method Based on Two-flow Skeleton Information

针对当前基于二维图像的人体动作识别算法鲁棒性差、识别率不高等问题,提出了一种融合卷积神经网络和图卷积神经网络的双流人体动作识别算法,从人体骨架信息提取动作的时间与空间特征进行人体动作识别。首先,构建人体骨架信息时空图,利用引入注意机制的图卷积网络提取骨架信息的时间和空间特征;其次,构建骨架信息运动图,将卷积神经网络网络提取到骨架运动信息的特征作为时空图卷积网络所提取特征的时间和空间特征的补充;最后,将双流网络进行融合,形成基于双流的、注意力机制的人体动作识别算法。算法增强了骨架信息的表征能力,有效提高了人体动作的识别精度,在NTU-RGB+D60数据集上取得了比较好的结果,Cross-Subject和Cross-View的识别率分别为86.5%和93.5%,相比其他同类算法有一定的提高。

Aiming at the problems of poor robustness and low recognition rate of current human action recognition algorithms based on two-dimensional images,a two-stream human action recognition algorithm based on convolutional neural network and graph convolutional neural network was proposed to extract the temporal and spatial features of human action recognition from human skeleton information.Firstly,the spatial and temporal graph of skeleton information is constructed,and the graph convolution network with attention mechanism is used to extract the temporal and spatial characteristics of skeleton information.Secondly,the skeleton information action graph is constructed,and the features extracted from the convolutional neural network are used as the time and space features of the features extracted from the spatio-temporal graph convolutional network.Finally,the two-stream networks are fused to form a human action recognition algorithm based on dual flow and attention mechanism.The proposed algorithm enhances the representation ability of skeleton information and effectively improves the recognition accuracy of human movements.It achieves good results on the NTU-RGB+D60 data set,and the recognition rates of Cross-Subject and Cross-View are 86.5%and 93.5%,respectively,which is a certain improvement compared with other similar algorithms.