Human Pose Prediction Based on LSTM and MediaPipe

Abstract

Motion capture technology is a crucial digital animation production method in the fields of film and television and games. Unmarked motion capture technology has received increasing attention and wide application in recent years due to its ability to significantly reduce the time for data collection and processing. However, this technology is prone to generating incorrect human posture information when dealing with occlusion issues. Therefore, this paper attempts to combine Long Short-Term Memory networks (LSTM) for human posture prediction to correct or eliminate incorrect posture data, thereby enhancing the accuracy and reliability of motion capture.