16-820 Advanced Computer Vision Projects

A comprehensive implementation of advanced computer vision algorithms including image matching, object tracking, 3D reconstruction, neural networks, and surface reconstruction. Features practical applications like AR overlays, panorama creation, motion detection, and character recognition developed as part of CMU's 16-820 course.

Project Structure

• hw1/: Image matching, homography, and AR applications
• hw2/: Lucas-Kanade tracking and motion detection
• hw3/: Epipolar geometry and 3D reconstruction
• hw4/: Neural networks and object detection
• hw5/: Photometric stereo and surface reconstruction

Key Features

Image Processing & AR

• FAST corner detection and BRIEF feature matching
• Homography computation using RANSAC
• AR overlay on book covers and panorama creation

Object Tracking

• Lucas-Kanade tracking algorithm
• Template correction for drift prevention
• Motion detection and segmentation

3D Reconstruction

• Eight-point algorithm for fundamental matrix estimation
• Triangulation for 3D point reconstruction
• Bundle adjustment optimization

Neural Networks & OCR

• CNN implementation for character recognition
• ResNet50 for object detection
• Text detection and line segmentation

Surface Reconstruction

• Photometric stereo implementation
• Surface normal estimation
• Depth map integration

Dependencies

• NumPy
• OpenCV
• PyTorch
• Scikit-image
• Matplotlib