Introduction to Computer Vision

This is Georgia Tech's classic 'Introduction to Computer Vision' (Aaron Bobick, with Irfan Essa and Arpan Chakraborty) — and the single most important honesty note is that it is NOT an edX course: the free, self-paced version lives on Udacity as ud810, and the catalog's edX link currently 404s. It is the best free MOOC for CLASSICAL (pre-deep-learning) computer vision, building rigorously from images-as-functions through filtering, edges/Hough, cameras and projective geometry, stereo, features (Harris/SIFT/RANSAC), optic flow, tracking (Kalman/particle filters), and recognition. Bobick's lectures are widely praised as graduate-grade and genuinely engaging, but the course is mathematically heavy (linear algebra and vector calculus) and the programming problem sets are demanding. It deliberately predates modern deep learning, so it is a foundations course rather than a path to building CNN-based vision systems on its own.

Excellent and free for learners who specifically want a rigorous classical computer-vision foundation and have the math/programming background, but the wrong choice if you want a beginner on-ramp or modern deep-learning CV — and note it is delivered on Udacity (ud810), not edX as the catalog states.

Best for: Engineering/CS students and self-learners with solid linear algebra, vector calculus and working Python (or MATLAB) who want to deeply understand the classical theory behind computer vision (image formation, geometry, features, motion, tracking) before or alongside deep learning. Ideal as a foundations course feeding into a modern CNN course like Stanford CS231n.

Skip if: Complete beginners, people without a comfortable math background, and anyone whose goal is to ship deep-learning vision models quickly — the course is pre-deep-learning, math-intensive, and its problem sets are time-consuming. Also not ideal for anyone who needs a recognized completion certificate.