Computer Vision — Course Projects

Prof. Luca Lombardi Andrea, May 2026

---

Computer Vision Project Repo

This repository contains the two projects developed for the Computer Vision course final evaluation. Both are command-line tools that implement concepts from the course, ranging from foundational image processing to full 3D scene reconstruction.

computer-vision-projects/
│
├── activity-a_harris-corner-detector/
│   ├── detector.py               # Harris Corner Detector (from scratch)
│   ├── detected_corners.png      # Sample output
│   ├── README.md
│   └── images
│
└── activity-b_3d-reconstruction/
    ├── reconstruct.py            # Two-view SfM pipeline
    ├── requirements.txt          # Python dependencies
    ├── images
    ├── cloud_000_002.np
    ├── cloud_000_004.npy
    ├── cloud_000_006.npy
    ├── cloud_000_008.npy
    ├── cloud.png
    ├── computer_vision_project.pdf
    ├── test.py
    ├── merged_cloud.py
    ├── dino_Ps.mat
    ├── dino_cloud.npy
    └── README.md

---

Activity A — Foundational Algorithm: Harris Corner Detector

Goal

Implement a classic computer vision algorithm from scratch, using only standard Python and NumPy — no high-level OpenCV filtering calls.

What it does

I chose this task: given a grayscale image, the script detects corner keypoints by analyzing local intensity changes using the Harris response function. Detected corners are visualized as green dots overlaid on the original image.

Usage

python detector.py --image <path_to_image> [options]

Argument	Default	Description
--image	required	Path to input image
--k	0.04	Harris sensitivity constant
--threshold	0.1	NMS threshold as fraction of max R
--window	3	Window size for structure tensor and NMS
--output	detected_corners.png	Output image path

Examples:

# Basic
python detector.py --image images/viff.000.ppm

# Tuned sensitivity
python detector.py --image images/viff.000.ppm --k 0.06 --threshold 0.05 --output results/corners.png

Dependencies

pip install numpy opencv-python matplotlib

---

Activity B — Complex Problem: Sparse 3D Reconstruction

Goal

Solve a real computer vision problem using standard libraries (OpenCV, SciPy). Starting from a stereo image pair, estimate a sparse 3D point cloud of the scene using a full Structure from Motion (SfM) pipeline.

What it does

The script takes two images of the same scene from slightly different viewpoints and reconstructs the 3D geometry by detecting and matching features, estimating camera pose, triangulating points in 3D, and refining them with bundle adjustment.

Usage

python reconstruct.py --img1 <image1> --img2 <image2> --K <fx fy cx cy> [options]

Argument	Default	Description
--img1	required	Path to first (left) image
--img2	required	Path to second (right) image
--K	required	Camera intrinsics: fx fy cx cy in pixels
--ransac-thresh	1.0	RANSAC reprojection threshold in pixels
--outlier-thresh	2.0	Max std deviations from centroid (0 = disabled)
--save	—	Save 3D point cloud to .npy file
--plot	—	Show interactive 3D scatter plot
--save-plot	—	Save plot to .png (for headless systems)
--debug	—	Enable verbose DEBUG-level logging

Examples:

# Basic reconstruction with interactive plot
python reconstruct.py --img1 images/viff.000.ppm --img2 images/viff.006.ppm \
    --K 2360 2360 384 288 --plot

# Save point cloud and plot, no display
python reconstruct.py --img1 images/viff.000.ppm --img2 images/viff.006.ppm \
    --K 2360 2360 384 288 --save cloud.npy --save-plot cloud.png

Dependencies

pip install -r requirements.txt

matplotlib            3.10.9
numpy                 2.4.4
opencv-contrib-python 4.13.0.92
scipy                 1.17.1

Note: opencv-contrib-python is required (not plain opencv-python) because AGAST and FREAK live in OpenCV's xfeatures2d contrib module.

---

Course Info

Course	Computer Vision
Professor	Luca Lombardi
Academic Year	2025/2026
Student	Andrea Emir Sevincel 535462