Pose Evaluation Method Based on Part-based Hierarchical Bidirectional Recurrent Neural Network

Human body posture evaluation can be applied to many important real-world fields such as sports referee assistance. One of the key challenges is to capture the dynamic changes of human bone key points from the video. In this paper, we propose a new network structure based on part-based hierarchical RNN (PHRNN) [6] to analyze the time series information on key points of bones, which can effectively to extract the temporal features of human pose from continuous video frames. We conduct experiments on our own volleyball data set. The experimental results show that the proposed method is effective against attitude evaluation.