Computer Vision Foundations · free preview

Reading 3D Structure from 2D Edges

Classify edges by their physical cause, separate figure from ground, and convert generic-view regularities into linear constraints on the 3D world.

1. One intensity step, several possible causes

One intensity step, several possible causes

2. A catalog of edges in the blocks world

A catalog of edges in the blocks world

3. Nonaccidental properties and the generic view

Nonaccidental properties and the generic view

4. From edge orientation to linear constraints

From edge orientation to linear constraints

5. Classifying an edge by its cause

In the blocks world, a thresholded edge map shows a nearly horizontal image edge, and the vertical scanline through it enters colored object pixels above the edge and white ground pixels below it. The edge sits where a block rests on the ground. What is this edge, and what does it contribute to the reconstruction?

6. What cotermination licenses you to conclude

Two straight image lines meet exactly at a junction. Which conclusion about the 3D scene does the cotermination argument support, and what makes the competing explanation unlikely?

7. The vertical-edge constraint value

A camera views the blocks world with tilt angle θ\theta satisfying cos⁡(θ)=3/5\cos(\theta) = 3/5. The system detects a 3D vertical edge, and the constraint on the world height map along this edge is ∂Y∂y=1cos⁡(θ)\frac{\partial Y}{\partial y} = \frac{1}{\cos(\theta)}. Write the value that the finite-difference constraint Y(x,y)−Y(x,y−1)Y(x, y) - Y(x, y-1) takes along the edge.

Practice this interactively

Free account · instant grading · spaced review that schedules itself.

Start this course for free