A method and system for recognizing fitness motions

Abstract

The application discloses a body-building action recognition method, comprising the following steps: generating start marks and end marks by marking body-building actions in a body-building video shown to a user; when detecting any start mark, calling an identification model corresponding to a current body-building action to identify IMU data of the user; and when detecting an end mark, ending the identification of the IMU data of the user. The application also discloses a body-building action recognition system. The body-building action recognition method and system mark a body-building video, identify actions by IMU data of a user in a required time period, reduce the effect of computing resources, reduce noise interference, improve recognition accuracy, and improve user experience.

Claims

1. A method for recognizing fitness movements, characterized in that, include: In the fitness videos shown to users, fitness movements are labeled to generate start and end markers; when a user exercises through the fitness video, if any start marker is detected, the recognition model corresponding to the current fitness movement is called to recognize the user's IMU data. When the end marker is detected, the recognition of user IMU data ends; Calling the recognition model corresponding to the current fitness movement to recognize user IMU data includes: collecting user IMU data for a preset duration as recognition data; the preset duration corresponds to the current fitness movement; inputting the recognition data into the recognition model and receiving the output data of the recognition model; the input data of the recognition model is user IMU data for a preset duration, and the output data of the recognition model is the movement recognition result; Collecting user IMU data for a preset duration as recognition data includes: when the start marker of any fitness movement is detected, collecting user IMU data for a preset duration at preset time intervals as recognition data; calling the recognition model corresponding to the current fitness movement to recognize the user IMU data also includes: between the start and end times of a fitness movement, retaining only the recognition result output by the recognition model that is closest to the fitness movement.

2. The fitness movement recognition method according to claim 1, characterized in that, In a fitness video shown to a user, marking fitness movements with start and end markers includes: when recording the fitness video, obtaining the start and end times of each fitness movement in the fitness video; marking the start time of each fitness movement in the fitness video with a start marker and marking the end time of each fitness movement in the fitness video with an end marker.

3. The fitness movement recognition method according to claim 1, characterized in that, The recognition model is obtained by: collecting IMU data of each fitness movement of a preset duration in the fitness video as sample data; and training a machine learning model with the sample data corresponding to each fitness movement to generate a recognition model corresponding to each fitness movement.

4. A fitness movement recognition system, characterized in that, include: The tagging module is configured to tag fitness movements in fitness videos shown to users, generating start and end markers. The recognition module is configured to, when a user exercises through a fitness video, call the recognition model corresponding to the current fitness movement to recognize the user's IMU data when any start marker is detected; and stop recognizing the user's IMU data when an end marker is detected. The recognition module is also configured to: collect user IMU data for a preset duration as recognition data; the preset duration corresponds to the current fitness movement; The identification data is input into the identification model, and the output data of the identification model is received; the input data of the identification model is user IMU data of a preset duration, and the output data of the identification model is the action recognition result; The recognition module is also configured to: when the start marker of any fitness movement is detected, collect user IMU data of a preset duration at preset time intervals as recognition data; between the start and end times of a fitness movement, retain only the recognition result of the movement output by the recognition model that is closest to the fitness movement.

5. The fitness movement recognition system according to claim 4, characterized in that, The marking module is further configured to: when recording the fitness video, obtain the start time and end time of each fitness movement in the fitness video; mark a start mark at the start time of each fitness movement in the fitness video, and mark an end mark at the end time of each fitness movement in the fitness video.

6. The fitness movement recognition system according to claim 4, characterized in that, The recognition model is obtained by: collecting IMU data of each fitness movement of a preset duration in the fitness video as sample data; and training a machine learning model with the sample data corresponding to each fitness movement to generate a recognition model corresponding to each fitness movement.

Images