I attempt to build a new system to quickly and accurately predict 2D human motion pose using a mobile/browser camera and give a similarity matching score based on the comparison with the correct teaching video. This system could be used in fitness area, like running or yoga.
This is a draft plan so far, to achieve my goal, I should learn some new things:
- Learn the theories of convolutional neural networks and other methods of computer vision, eg. the Stanford open courses: cs230, cs231n, cs231a.
- Learn the tutorials of TensorFlow, knowing how to implement the application and how to build the architecture then feed the training data to the model.
- Learn new programming languages for mobile deployment, eg. Swift/JavaScript/Python ect.
- Read more papers in this area and discover more possibilities using different methods, especially matching the real-time skeleton with the correct teaching motion sequence in the video.