Project aimed at building a computer-vision system that analyzes 8-ball billiard matches. The system extracts high-level information from match footage and visualizes it on a 2D top-view minimap.
The challenging part of the project was to develop such system relying only on traditional computer vision methods such as edge detection, Hough transform, SVM, projective geometry and homography matrix, and in particular without any type of deep neural network involved.
The current pipeline supports:
- Playing-field Segmentation
- Ball Detection
- Ball Classification
- Projection of the playing field onto a top planar view
- Trajectory Tracking of each ball
- Mini-map Top-view Visualization
