Book Detection with SIFT & FLANN

This project detects multiple instances of books in scene images using SIFT feature extraction, FLANN-based feature matching, and RANSAC homography. It visualizes detected books with color-coded bounding boxes and computes internal metrics to assess detection quality.

---

Features

• Detect multiple instances of the same book in a scene.
• Color-coded bounding boxes per model.
• Iterative detection for crowded scenes.
• Internal quality metrics:
  • Number of good matches
  • Inlier ratio from RANSAC
  • Reprojection error
  • Bounding box area and aspect ratio
• Debug mode for visualizing keypoints, matches, and masks

---

Dependencies:

1. OpenCV (opencv-python)

2. NumPy

3. Matplotlib

4. Jupyter

Usage

1. Put the dataset.zip file under a new folder in your Google Drive account, named as you want. Just remember to change the path when extracting in the notebook.

2. Open the notebook:

   jupyter notebook book_detection.ipynb

3. Execute cells sequentially.

4. Provide paths for /models and /scenes folders if required.

Outputs:

Number of instances for each model and bounding boxes with drawn on scene images with different colors.

Configuration

scale_factor: scale applied during preprocessing (default 3.0)

min_matches: minimum number of good matches for RANSAC (default 30)

colors: list of RGB colors for bounding boxes

debug: True to visualize intermediate results (keypoints, matches, masks)

Example

Before detection

After detection

License

This project is licensed under the MIT License. See the LICENSE file for more details.