COMP 558 Fundamentals of Computer Vision @ McGill Class Notes
* Instructor: Kaleem Siddiqi
* Term: Fall 2022
* Grading: 20% Assignments + 20% Group Research Project + 20% Midterm + 40% Final
* Author: @asahahaha
All class notes written in notion.so, click on the lecture titles to open the corresponding note page.
- Bayer pattern
- RAW to JPG
- Smoothing (by local averaging), local differences
- Cross-correlation VS. convolution
- Impulse functions
- 1D VS. 2D
- Gaussians
- Prewitt and Sobel (1960)
- Marr & Hildreth (1979)
- Canny edge detection
- Least squares:
- by linear regression
- by total least squares
- Hough Transform
- RANSAC (Random Sample Consensus)
- Gaussian smoothing
-
- Iterative
-
- HEAT equation
-
- Gaussian property
- Lucas-Kanade Algorithm
- Coarse-to-fine
- Box(blob) detector
- Structure tensor
- (weighted) 2nd Moment Matrix
- Corner →
$\lambda_1$ &$\lambda_2$ both big and diffe - Harris Corner Detection:
$\lambda_1\lambda_2-k*(\lambda_1+\lambda_2)^2$ - Alternative Corner Detection:
$\lambda_1\lambda_2/(\lambda_1+\lambda_2+\epsilon)$ - inner scale & outer scale
- HoG, histogram equalization
- Keypoint detection using laplacian pyramid & gradient → SIFT vector ⭐️⭐️
- Laplacian == DoG
- 1D & 2D Image Registration
- iterative method
- Applications:
- Lucas-Kanade
- KLT Tracking
- Shi & Tomasi (a 6D problem)
- Depth map
- Camera movement
- similar triangles
- 1, 2, 3 point perspectives
- 2D & 3D rotaion, scaling
- Homogenous coordinates
- Camera calibration matrix K → intrinstics
- Extrinsics
$R[I|-C]$
- A =
$U∑V^T$
- Having a list of paired point coordinates
- Data normalization
$P==M_1^{-1}P_{normalized}M_2$
- Case 1: 1 camera, 1 image plane
- Case 2: 2 cameras, 1 image plane
- Case 3: 1 camera (rotation)
- Case 4: 2 images, 1 scene space (not same image plane)
- Essential matrix:
- Uses projection plane coordinates, not pixel coordinates
- Fundamental matrix
- Uses pixel coordinates
- Estimating the fundamental matrix F
- Disparity Estimation
- Estimate Depth
- Correspondence problem
- Occlusions
- Disparity space
- How does depth camera work
- Iterative Closest Points (ICP)
- Classification problems in computer vision