Revision f592ee3d5cea2100e047b684098abc0f1be9b889 (click the page title to view the current version)
Changes from f592ee3d5cea2100e047b684098abc0f1be9b889 to 880c52e836eeb7f11edba10cde17d403f6e57d3e
---
title: Tracking Features
categories: lecture
---
- aperture problem [page 78]
# Briefing
## Corners
+ What are distinctive points in the image?
+ Distinctive points can (to some extent) be matched in two different images.
[More Images](https://www.flickr.com/photos/ntnu-trondheim/collections/72157632165205007/)
## Corner Correspondence
+ Two images of the same scene
$I_1,I_2: \Omega\subset\mathbb{R}^2\to\mathbb{R}_+ ; \mathbf{x}\mapsto I_1(\mathbf{x}),I_2(\mathbf{x})$
+ Different in general
*Why are they different?*
+ Firstly - different colour in different directions
+ Lambertian assumption
+ Secondly - noise
+ let's assume this is insignificant
+ Thirdly - different image points $\mathbf{x}_1,\mathbf{x}_2$
correspond to the same 3D point $p$
## Brightness Constancy Constraint
+ Suppose we photograph empty space except for a single point $p$
- *Brightness Constancy Constraint*
$$I_1(\mathbf{x}_1) = I_2(\mathbf{x}_2) \sim \mathcal{R}(p)$$
+ Simple dislocation from $\mathbf{x}_1$ to $\mathbf{x}_2$
+ Motion: $h: \mathbf{x}_1\mapsto\mathbf{x}_2$ so that
$$\forall
\mathbf{x}_1\in\Omega\cap h^{-1}(\Omega)\subset\mathbb{R}^{2},
\;I_1(\mathbf{x}_1) = I_2(h(\mathbf{x}_1))$$
### Motion Models
+ **Translational Motion Model:**
$$h(\mathbf{x}_1) = \mathbf{x}_1 + \mathbf{\Delta x}$$
+ **Affine Motion Model:**
$$h(\mathbf{x}_1) = A\mathbf{x}_1 + \mathbf{d}$$
+ **Projective Motion Model:**
$$h(\mathbf{x}_1) = H\mathbf{x}_1$$
where
$H\in\mathbb{R}^{3\times3}$ is defined up to a scalar factor.
### Intencity Transformation
+ Need to accept changes to the intencity
$$I_1(\mathbf{x}_1) = I_2(h(\mathbf{x}_1)) + n(h(\mathbf{x}_1))$$
+ Occlusions
+ Non-Lambertian reflection
+ Taken at different time? Different ambient light?
## Feature Tracking
+ Estimator
$$\hat h = \arg\min_h\sum_{\tilde{\mathbf{x}}\in W(\mathbf{x})}
||I_1(\tilde{\mathbf{x}})-I_2(h(\tilde{\mathbf{x}}))||^2$$
+ The window, or aperture, $W(\vec{x})$
+ Choose $h$ from a family of functions, parameterised by $\alpha$
+ translational: $\alpha=\Delta\mathbf{x}$
+ affine: $\alpha=\{A,\mathbf{d}\}$
+ **Aperture problem:** cannot distinguish points on a blank wall
### Infinitesimal Model
+ Consider simple translational model
$$I_1(\textbf{x})= I_2(h(\textbf{x}))= I_2(\textbf{x}+\Delta\textbf{x})$$
+ Consider infitesimally small $\Delta\textbf x$
+ Model on a time axis
+ two images taken infinitesimally close in time
+ ... under motion
1. First write $\mathbf{\Delta x} = \mathbf{u}dt$$, and
rewrite the brightness constancy
$$I(\mathbf{x}(t),t) = I(\mathbf{x}(t)+\mathbf{u}dt,t+dt)$$
2. Apply Taylor series expansion and ignore higher-order terms
$$\nabla I(\mathbf{x}(t),t)^\mathrm{T}\mathrm{u}dt + I_t(\mathbf{x}(t),t)dt = 0$$
where
$$\nabla I(\mathbf{x},t) = \begin{bmatrix} I_x(\mathbf{x},t)\\ I_y(\mathbf{x},t) \end{bmatrix}
= \begin{bmatrix}\frac{\partial I}{\partial x}(\mathbf{x},t)\\ \frac{\partial I}{\partial y}(\mathbf{x},t) \end{bmatrix}
\in\mathbb{R}^2$$
and
$$I_t(\mathbf{x},t) = \frac{\partial I(\mathbf{x},t)}{\partial t}\in \mathbb{R}$$
3. Simplify
$$\nabla I(\mathbf{x}(t),t)^\mathrm{T}\mathrm{u} + I_t(\mathbf{x}(t),t) = 0$$
+ *Brightness Constancy Constraint* for the simplest possible continuous model
+ Two applications
- optical flow: fix a position $\mathbf x$ and consider particles passing through
- feature tracking: fix a particle $x(t)$ an track it through space
### Solving for $\textbf{u}$
+ Consider the equation
$$\nabla I^\mathrm{T}\mathrm{u} + I_t = 0$$
+ There are infititly many solutions, due to the *aperture problem*
+ We can solve for the component in the direction of the gradient though
1. Scalar projection of $\mathbf u$ onto $\nabla I$.
$$\frac{\nabla I^\mathrm{T}\mathrm{u}}{||\nabla I||} = - \frac{I_t}{||\nabla I||} $$
2. Multiplying by $\nabla I/||\nabla I||$, we get the vector projection:
$$\mathbf u_n =
\frac{\nabla I^\mathrm{T}\mathrm{u}}{||\nabla I||}\cdot\frac{\nabla I}{||\nabla I||} =
- \frac{I_t}{||\nabla I||}\cdot\frac{\nabla I}{||\nabla I||} $$
### Least squared errors estimate
+ Integrate over a window with sufficient texture
+ allows us to estimate $u$ in two dimensions
+ too many approximations for exact solution
+ Minimise the sum of squared errors:
$$E_b(\mathbf{u}) = \sum_{W(\mathbf{x}) [\nabla I^T(\tilde{\mathbf{x}}(\mathbf{u}(\mathbf{x})+I_t(\tilde{\mathbf{x}},t}]^2$$
