I want to make an application that can provide counting for push up, I have tried on the openpose and canobtained the pose 2d data, and using the output of the openpose bone head position, I can get the signal also. However what I am stuck is the ways to classify whether the image is doing push up or not.
The only thing I was thinking is to collect all possible human activities and make a model that can classify it is doing push up or other activities, but I think it is kind of weird for classifying not push up for all kind of activities, furthermore, it is almost impossible for collecting all types of possible human activities. How am I supposed to do the binary or non-binary classification for not doing push up or doing it?
https://aicurious.io/posts/2021-02-15-build-a-pushup-counter/
Btw, I have read a similar project for push up counter through bones data, however I can't find the details of the way for this project to do the classification part...I think it just briefly mentioned they have used the MobileNetV2....