基于人体图像生成的姿态无关人物识别OACSTPCD

Person identification technology enables the robots to have the ability to recognize the identities of users,which effectively improves the intelligent interaction level of robots.One of the main challenges of person identification is the influence of the pose changes on person feature extraction.In order to solve this problem,a pose-independent person identification method based on humon body image generation is proposed,which aims to eliminate the influence of pose change on the person appearance features by generating the human body images with the same poses as the target persons in the dataset.Firstly,the method uses the human body seg-mentation map to separate the human body regions from the background to minimize the interference of the complex and changeable background on the human body appearance features.Then,a human body image with the same pose as the target image is generated under the guidance of the target pose.Finally,a feature fusion module is designed to fuse the identity features of the source and generated image to extract pose-independent robust identity features for person identification.In addition,to better distinguish different persons,negative samples with the same pose are generated in the training process to constrain the model to learn more fine-grained discriminative identity features.Experimental results on person identification and human body image generation demonstrate the effectiveness of the method.