The document for this project is on: Soccer Pose Processing and Brainstorming Space.
Relevant paper: Applying-Pose-Estimation-to-Predict-Amateur-Golf-Swing-Performance-using-Edge-Processing-February-2020
To ensure that YOLOv5 can detect the ball, the ".png" of contact_frames are actually the frames before the soccer is kicked.
- Data_labels.npy: Each row represents a training example. The columns represent different labels. From 0 -> 6 they are as follows:
- kick number,
- foot (left or right),
- direction (left=0, center=1, or right=2),
- height (low=0, center=1, high=2),
- type (curl=0, laces=1),
- quality (poor=1, average=2, good=3),
- spin (right, left, top, back, knuckle).
See file \output\Preprocessed_keypoints_1.npy. This is the numpy array containing the preprocessed data. It is of shape (20,2,25,2). (num_samples, frame, keypoints, x/y). Where frame[0] = contact frame, and frame[1] = plant foot frame. Plant foot frame refers to the frame when the foot-not-hitting-the-ball’s location is fixed.