Lecture 25: Computer Vision

Lecture 25: Computer Vision¶

AIMA Chapter 25 — 1 hour¶

Learning Objectives¶

• Understand image formation

• Extract features: edges, texture, optical flow

• Apply CNNs for classification and detection

• Recover 3D from images

Image Formation¶

• Pinhole camera: Perspective projection

• Lens: Focus, aberrations

• Light: Shading, color

Simple Features¶

• Edges: Gradients, Canny

• Texture: Filters, statistics

• Optical flow: Motion between frames

Image Classification¶

• CNN: Conv layers → FC → softmax

• AlexNet, VGG, ResNet: Architectures

• Why CNNs work: Hierarchical features

Object Detection¶

• Sliding window: Classify each region

• R-CNN: Proposals + CNN

• YOLO, SSD: Single-shot

3D Reconstruction¶

• Stereo: Two views, triangulation

• Structure from motion: Multiple views

• Depth from single image: Learned

Summary¶

• Formation: Pinhole, lenses

• Features: Edges, texture

• CNN: Classification, detection

• 3D: Stereo, SfM

References¶

• AIMA Ch. 25

• Russell & Norvig, AIMA 4e, Ch. 25

• Chapter PDF: chapters/chapter-25.pdf

• aima-python: perception4e.py

Questions?¶

Next lecture: Robotics (Chapter 26)