Multi-channel fitness exercise recognition method based on human skeleton articulation points

Abstract

The invention relates to a multi-channel fitness exercise recognition method based on human skeleton articulation points. The method comprises: a step 1, collecting human motion depth image time series data; and a step 2, establishing a human skeleton joint point position change data set for the first channel. The beneficial effects of the invention are that the method carries out the learning andtraining through the dual-channel model of the LSTM-based deep learning algorithm, carries out the recognition of the movement posture, is more accurate in bone joint point position data compared with the conventional image data, and is more accurate in positioning and detection of the movement of a human body; a ground equation is introduced, so that the spatial relationship between the human body articulation point position and the ground can be accurately judged, the spatial relationship of the human body can be more accurately recognized, and the adaptability of the recognition method todifferent environmental changes is enhanced; and skeleton data in an NTURB+D data set is firstly adopted to train a model, and then a self-built human body fitness exercise data set is used for transfer learning training, so that the accuracy of model recognition is improved.

Description

technical field [0001] The invention relates to the technical field of image recognition, in particular to a multi-channel fitness exercise recognition method based on human bone joint points. Background technique [0002] In order to accelerate the transformation from centering on disease treatment to centering on people's health, government departments mobilize the whole society to implement the policy of putting prevention first, implement the Healthy China Action, and improve the health of the whole people. With the support of government departments, the public pays more attention to fitness than ever before. When doing fitness exercises, non-standard or non-standard fitness movements often make the exercise effect counterproductive, which is not conducive to the improvement of the public's health quality, and may also cause sports injuries. Choosing a more accurate fitness exercise recognition method is of great significance to social public health. [0003] In image ...

AI Technical Summary

Problems solved by technology

[0003] In image recognition technology, although the traditional RGB CV-based recognition method has rich image information, when extracting human motion features, it is often affected by the limitations of the ambient light level and the technology itself, which affects its recognition speed and recognition accuracy. ; and in recent years, based on depth image joint point judgment, although it can extract the bone joint point for identification and judgment, it is difficult to determine the positional relationship between the player's human body position (such as the ground) and the equipment when exercising , which is also limited in the accuracy of sports movements

Images