This is an AI that gives real-time feedback to the user of how they're performing a body movement (like, workout, dance, etc) against a benchmark video.
You can give the benchmark video as a pre-saved file, and the user video as either a pre-saved file or with the webcam feed.
Its built using mediapipe in the backend, so we get a pretty high FPS (around 15 on MacBook Pro 16) when running on CPU only, as opposed to 2-3 FPS when the same application was built with tf-pose-estimation.
from move_comparison import compare_positions
benchmark_video = 'dance_videos/benchmark_dance.mp4'
user_video = 'dance_videos/right_dance.mp4' # replace with 0 for webcam
compare_positions(benchmark_video, user_video)
You can create your own moves with the create_moves.py file. It essentially opens the webcam, lets you do your move, and when you're done, you can press Q to save it with the supplied name: create_move('Move 1')