SoccerVision: Player Detection and Team Classification

An implementation of computer vision algorithms for soccer player detection and team classification.

Overview

Technical Approach

1. Player Detection

We use a custom implementation of Canny edge detection and contour finding:

• Grayscale Conversion & Blurring: Convert to grayscale and apply Gaussian blur to reduce noise
• Edge Detection: Apply Canny edge detector to find boundaries
• Field Masking: Segment the green field using HSV color thresholding
• Contour Finding: Extract contours from non-field areas that likely represent players
• Size Filtering: Filter contours by area to identify player regions
• Non-Maximum Suppression: Remove overlapping detections

2. Team Classification

We use color-based segmentation with a custom clustering implementation:

• Color Extraction: Extract dominant colors from player regions in HSV color space
• Histogram Analysis: Create histograms of hue values to identify jersey colors
• Custom Clustering: Implement a custom binary clustering algorithm to separate teams
• Team Assignment: Assign each player to one of two teams based on jersey color

Implementation Details

Key Components

1. Player Detection Module: Implements Canny edge detection and contour-based player detection
2. Team Classification Module: Implements custom color-based team classification
3. Utilities: Helper functions for loading, processing, and visualizing images

Core Algorithms

1. Player Detection:

   • Canny edge detection to find boundaries
   • Color-based field segmentation
   • Contour extraction for player identification
   • Non-maximum suppression to clean up overlapping detections

2. Color-based Team Classification:

   • HSV color space analysis of player regions
   • Histogram-based dominant color extraction
   • Custom binary clustering algorithm (similar to k-means but without external libraries)