FA18 project for CS 196 focused on event detection/recognition. We decided to focus on sports events, specifically soccer.
Note: This repository was imported from the original CS196Illinois organization after the semester ended and the original repository was made private.
YOLO is a real-time detection algorithm developed for object detection. We used darknet (a neural network using YOLO) to detect yellow cards from soccer matches in an image/video. The pretrained models did not suppert this feature, so we ended up training a custom model for this.
Our trained weights are located here. This model can be used to identify yellow cards in still images or videos.
Click here for more information.
This package analyzes each frame of a video and filteres the pixels for the desired color and shape. Traditional computer vision techniques are used here, such as HSV filtering and contour/blob processing. This is an alternative to YOLO for detecting events with yellow cards.
Click here for more information.
In sporting events there are almost always scoring overlays that are updated in real-time as the match goes on. Our solution to detect when goals were scored in soccer uses this fact. We used color filtering to detect where the overlay was in a video, and then used optical character recognition to figure out what the scores are and when they change.
Click here for more information.
- Akhil Isanaka
- Himanshu Minocha
- Eric Zhang
- Omar Khan
- Sathwik Pochampally