Joint data augmentation and attention mechanism for occluded human pose estimation

Human pose estimation technology has important applications in many fields. Existing research mainly focuses on precise localization of human key points in unobstructed situations but neglects the common occlusion problem in daily environments. Occlusion can disrupt the correlation between key points in the human body, causing a decrease in model accuracy. In response to the above issues, this article proposes a joint data augmentation and attention mechanism for occluded human pose estimation. This article focuses on training human key points in images, generating a specific number and size of occluded areas to simulate scenarios where human key points are occluded. To further enhance the model's ability to extract feature information, this method considers the global context of feature and integrates the GC Block attention mechanism into the human pose network structure, enabling the network to focus on the human body regions in the image. This method has been experimentally validated on the OCHuman dataset, and the results show that it can effectively improve the accuracy of joint localization in human pose estimation.