Key points are not available for this paper at this time.
Abstract With the development of computers in recent years, human body recognition technology has been vigorously developed and is widely used in motion analysis, video surveillance and other fields. As the traditional human action recognition relies on video decomposition frame-by-frame, artificially designed motion features to achieve the role of recognition, this approach is both energy-consuming recognition efficiency is also very low. Thanks to the advent of deep learning, computers can automatically extract features from movements and then recognize and classify them. This research is based on deep learning to improve human pose estimation. Firstly, Involution's feature extraction network is proposed for lightweight human pose estimation, which is combined with existing human pose estimation models to recognize human pose. Each joint of the human body is labelled and classified, weights are added to each part, features are extracted between the joints at each moment, and the extracted features are then fed into a long and short term memory neural network for recognition. Experimental results show that the number of parameters and computational effort of the improved human pose estimation model is reduced by around 40% compared to the original model, while still providing a slight improvement in accuracy. The performance of the model under each algorithm is compared with the model proposed in this study, and the results show that the proposed model has better performance in recognizing different martial arts movements.
Sifang Zhang (Thu,) studied this question.