Computer vision I

Introduction.

• Overview of the course: topics, course rules.
• General overview of Computer Vision in 2021.

Review of useful maths.

• Linear algebra: matrices; kernel; rank; eigenvalues and eigenvectors.
• Linear algebra: SVD: QR; Cholesky.
• Homogeneous coordinates and projective geometry..
• 2D Euclidean transformations.
• 3D Euclidean transformations.
• Solid angles.

Geometry.

• The Camara Oscura.
• Lenses.
• Light-sensitive surfaces.
• Human vision.
• Introduction to projective geometry.
• Affine and euclidean geometries.
• Geometry stratification.
• Projective models for cameras.
• Affine models for cameras.

Camera calibration.

• Calibration from 2D-3D correspondences.
• Calibration from planar scenes.
• Calibration in practice.
• Principles of self-calibration.

Two-view geometry.

• Geometry of 2 views of planar scenes.
• Estimation of homographies.
• RANSAC.
• Epipolar geometry.
• Estimation of fundamental matrices.
• Structure from motion.
• Bundle adjustment.

Photogrammetry.

• Illumination and light intensity.
• Reflectance.

Color.

• The concept of color.
• The RGB and CIE spaces.
• Perceptually uniform spaces.

Image processing.

• Point operators.
• Contrast control operators. Equalization.
• Linear operators, convolution and correlation.
• Complexity of linear filters. Separability.
• Interpolation. Sub-sampling.
• Coarse-to-fine strategies and pyramidal representations.
• Non-linear filtering.
• The Fourier transform and its properties.
• Filtering in the frequency space.

Interest points.

• General principles. Invariance properties. The Harris detector.
• Popular detectors: SIFT, SURF, FAST, ORB.
• Popular descriptors: SIFT, SURF, BRIEF, ORB.
• Comparing descriptors and matching points.

Mid-level computer vision.

• Taxonomy of mid-level problems.
• Common mathematical/probabilistic formulations.
• Common optimization techniques.

Image restoration.

• Problem statement.
• Denoising methods: Nl_means and BM3D
• Contrast processing methods
• Stereovision.

Problem statement.

• Rectification.
• Local matching.
• Global and semi-global dense stereo-matching.

Optical flow.

• Definition and problem statement.
• Dense and sparse optical flow methods.

Image segmentation.

• Problem statement.
• MRFs and segmentation.
• Flow networks-based methods.

Convolutional Neural Networks.

• General principles for neural networks. Limits in computer vision.
• The convolutional neural networks.
• Stride, padding and other hyperparameters.
• Design of CNN architectures.
• Transfer learning.
• A few famous architectures.
• Practice with TF2.

Object recognition.

• Problem statement and applications. Imagenet and other benchmarks.
• A few older paradigms for object recognition.
• CNN-based solutions.

Object detection.

• Problem statement.
• General principles in the object detection problem.
• A few successful architectures.

Visual tracking.

• Problem statement.
• Tracking as a recursive Bayesian filtering problem.
• MOT and data association.
• Tracking-by-detection.
• Visual tracking and Deep Learning.
• The SORT/DeepSort strategies.
• Recent advances in visual tracking.

Applications: Visual SLAM.

• Problem statement and applications.
• A few competitive VSLAM systems.

Applications: Augmented reality.

• Problem statement.
• Marker-based augmented reality.
• Markerless augmented reality.

Applications: .

• Machine vision and autonomous driving.