Enhanced Oriented Bounding Box (OBB) Detection

This repository contains a comprehensive implementation of Oriented Bounding Box detection using multiple approaches. The implementation includes various methods for detecting rotated objects in images and provides visualization tools.

Features

• Multiple detection methods:
  • Contour-based detection
  • Connected Components analysis
  • MSER (Maximally Stable Extremal Regions)
• Image preprocessing with CLAHE enhancement
• Configurable parameters for detection
• Visualization tools with angle and center point display
• Result export to JSON format
• Type hints and comprehensive documentation

Requirements

Install the required packages using:

pip install -r requirements.txt

Usage

Basic Usage

from enhanced_obb_detection import EnhancedOBBDetector

# Create detector instance
detector = EnhancedOBBDetector()

# Load image
image = cv2.imread("your_image.jpg")

# Detect objects (choose method: 'contour', 'connected_components', or 'mser')
detections = detector.detect_objects(image, method='contour')

# Visualize results
result = detector.visualize_detections(image, detections)

# Display
cv2.imshow("Detection Results", result)
cv2.waitKey(0)

# Save results
detector.save_results(detections, "detections.json")

Custom Configuration

You can customize the detector's parameters:

config = {
    'min_area': 150,
    'max_area': 3000,
    'min_aspect': 1.5,
    'max_aspect': 4.0,
    'canny_low': 30,
    'canny_high': 200,
    'clahe_clip_limit': 3.0,
    'clahe_grid_size': (8, 8),
    'morph_kernel_size': (3, 3)
}

detector = EnhancedOBBDetector(config)

Detection Methods

1. Contour-based Detection

   • Uses edge detection and contour analysis
   • Best for objects with clear boundaries
   • Configurable edge detection parameters

2. Connected Components

   • Uses region labeling and properties
   • Good for separated objects
   • Supports different connectivity options

3. MSER Detection

   • Uses Maximally Stable Extremal Regions
   • Effective for text and blob detection
   • Configurable region parameters

Output Format

The detection results are returned as a list of dictionaries, each containing:

• box: Corner points of the oriented bounding box
• center: Center coordinates
• width: Box width
• height: Box height
• angle: Rotation angle in degrees
• area: Area of the detected region
• confidence: Detection confidence score

Example

Run the example script:

python enhanced_obb_detection.py

This will:

1. Load a sample image
2. Apply all three detection methods
3. Display the results side by side
4. Save the detections to a JSON file

License

This project is licensed under the MIT License - see the LICENSE file for details.

Contributors Name: Dhruv Patel and Devanshi Rathod