THE IDENTIFICATION OF KNEE DEVIATION DURING THE SQUAT MOVEMENT
share
This paper describes software to evaluate the ideal form to prevent injury and promotes muscle growth when squatting. The software spots if any excessive knee internal rotation or abduction is occurring. For model creation, we took Tensorflow's Body Pix model and performed 3- dimensional pose segmentations and used those coordinates in another model to correctly classify the user’s squatting form as either too far or too inward in terms of deviation. Specifically, we created a multi-variable regression model that connects these points in the XYZ plane that was trained on the resulting coordinates from the pre-trained pose estimation model used on a custom dataset of YouTube videos from power lifting competitions. Then, we modeled Euclidean distance, 3d angles, and depth value through equations to evaluate the bend and distance between the hip, knee, and ankle for targeting left and right deviation by specifying 3 stages: going down, middle squat, and going up.
READ FULL TEXT