Research on Human Pose Estimation Algorithms Considering Occlusion Perception
Traditional stage lighting techniques rely on manual operation,often making it difficult to achieve the high standards of precision desired by designers.In response to this challenge,this paper propose an automatic stage lighting tracking solution based on human pose estimation.Currently,in stage performances,occlusions between actors or between actors and props can adversely affect the accuracy of pose estimation model,thereby impacting the quality of the lighting tracking.To address occlusion issues with actors and enhance the accuracy of human pose estimation,this paper designs a top-down occlusion-aware human pose estimation model and method.The model explicitly learns structural information of the human body,utilizes spatial attention mechanisms to transform image features into keypoint feature vectors,and employs multi-head attention mechanisms to enhance the learning capability of human body structure.This approach improves the robustness of pose estimation under occlusion conditions.Additionally,the model integrates a module for predicting occlusion information.Experimental results demonstrate that this method achieves outstanding performance on public datasets,effectively reducing prediction errors under occlusion conditions.Moreover,it can determine whether keypoints are occluded,making it promising for widespread application in complex scenarios such as stage lighting automatic tracking systems.
万方数据
维普
