Computer Vision Foundations · free preview

Spatiotemporal Blur and Frame Differences

The 3D Gaussian with independent spatial and temporal widths, velocity-steered blur that tracks one mover, and the frame-difference detector.

1. Blurring a cube

Blurring a cube

2. Steering the blur by velocity

Steering the blur by velocity

3. The frame difference

The frame difference

4. Derivatives obey the constancy relation

Derivatives obey the constancy relation

5. Choosing where the blur bites

A video is convolved with a spatiotemporal Gaussian having small spatial width σ=1\sigma = 1 and large temporal width σt=4\sigma_t = 4. What does the output look like?

6. What steering accomplishes

A video contains a pedestrian walking right at 1 pixel per frame and a cyclist crossing left at 4 pixels per frame. You filter with the steered Gaussian g(x−vxt,y,t)g(x - v_x t, y, t) using vx=1v_x = 1. What happens?

7. Temporal derivative from constancy

A pattern satisfies brightness constancy with vx=2v_x = 2 pixels per frame and no vertical motion. At some pixel the spatial derivative ∂ℓ/∂x\partial \ell / \partial x equals 33. Using ∂ℓ/∂t=−vx ∂ℓ/∂x\partial \ell/\partial t = -v_x\, \partial \ell/\partial x, compute the temporal derivative there.

Practice this interactively

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

Start this course for free