Human Dance Posture Detection Based on Improved Mayfly Algorithm

Abstract Conventional human dance posture detection methods have problems such as low motion detection accuracy and recognition rate, so a simplified and improved mayfly algorithm is proposed to optimize the human dance posture detection methods. To begin with, a high-precision Kinect sensor is employed to gather 3D data on human dance posture movements. Then, the movement categories are recognized based on the indirect segmentation principle of the sliding window design. Then, the improved mayfly algorithm optimizes the multi-threshold combination of image segmentation to determine the optimal segmentation threshold. It is proposed to use gesture-based feature description to fully represent the human action information, use human gesture to obtain the human body regions in the frame, extract 3D-SIFT and optical flow features for each region, respectively, and then compare with other intelligent algorithms, and the experimental analysis shows that the proposed method is better than the DSI method in terms of Average accuracy and Accuracy at the worst performance. Performance is higher than the DTW method, with a difference of 29.91% and 28.65%, respectively. The improved mayfly algorithm’s simulation results are more accurate and stable than other methods, which improves the recognition rate and allows for more precise detection of human dance postures.