融合三维骨架和深度图像特征的人体行为识别

Human Activity Recognition Based on Fusing 3D Skeleton and Depth Image Feature

人体行为识别是计算机视觉与模式识别领域最为活跃的研究方向之一。针对现有方法多采用单一特征研究人体行为识别导致识别率较低的问题,提出了一种融合三维骨架特征和深度图像特征的多特征人体行为识别方法。该方法首先从三维骨架中提取出基于运动姿态描述子的稀疏编码特征,同时从深度图像中提取出基于深度运动图的梯度方向直方图特征,以增强特征互补性;然后利用线性分类器分别获得这两种特征的识别结果;最后将这两种特征的识别结果利用对数意见汇集规则融合得出最终的识别结果。该方法在MSRAction3D数据集上的识别率为98.53%,不但超过了基于三维骨架特征方法的识别率和基于深度图像特征方法的识别率,而且相对于其他方法也取得了更高的识别率。

Human activity recognition has become one of the hottest topics in the field of computer vision and pattern recognition.In order to solve the problem that the recognition rate of human activity recognition is low due to the use of single feature in existing methods,we propose a multi-feature human activity recognition method that integrates 3D skeleton feature and depth image feature.Firstly,to increase the complementarities of features,the feature of sparse coding is extracted from 3D skeleton based on moving pose descriptor,and the feature of histograms of oriented gradients is extracted from depth image based on depth motion map.Then,the linear classifier is used to obtain the recognition result based on each kind of feature.Finally,the rule of logarithmic opinion pool is used to fuse the recognition result from each kind of feature.The recognition rate of this method on MSR Action3D dataset is 98.53%,which not only exceeds the recognition rate based on 3D skeleton feature method and based on depth image feature method,but also achieves higher recognition rate compared with other methods.