Computer Vision

CS-E4850, 5 study credits

Lecturer: Juho Kannala
 Lecture 7:
Optical flow and keypoint tracking
• Given two subsequent frames of a video, the optical flow
field indicates the apparent motion of each pixel

• If we have more than two frames, we can track features

from one frame to the next by following the optical flow

Acknowledgement: many slides from Svetlana Lazebnik, Derek Hoiem, Steve Seitz,
Rick Szeliski, M. Pollefeys, and others (detailed credits on individual slides)
 Reading & software
• Szeliski’s book, 1st edition: Chapter 8 or 2nd edition: Chapter 9

• Baker & Matthews: Lucas-Kanade 20 years on, a unifying framework,

2004
– https://www.ri.cmu.edu/publications/lucas-kanade-20-years-on-a-unifying-
framework/

• Shi & Tomasi: Good features to track, 1994

– http://www.ai.mit.edu/courses/6.891/handouts/shi94good.pdf

• OpenCV software:
– http://docs.opencv.org/3.1.0/d7/d8b/tutorial_py_lucas_kanade.html
Motivation: glimpse to the state of the art
Motion is a powerful perceptual cue
• Sometimes, it is the only cue
 Motion is a powerful perceptual cue
• Even “impoverished” motion data can evoke a
strong percept

G. Johansson, “Visual Perception of Biological Motion and a Model For Its Analysis",
Perception and Psychophysics 14, 201-211, 1973.
 Motion is a powerful perceptual cue
• Even “impoverished” motion data can evoke a
strong percept

G. Johansson, “Visual Perception of Biological Motion and a Model For Its Analysis",
Perception and Psychophysics 14, 201-211, 1973.
Uses of motion in computer vision
• 3D shape reconstruction
• Object segmentation
• Learning and tracking of dynamical models
• Event and activity recognition
 Preview: Structure from motion
• Given a set of corresponding points in two or more
images, compute the camera parameters and the 3D point
coordinates

? R ,t
1 1
?
R3,t3 ?
R2,t2
 Keypoint tracking

C. Tomasi and T. Kanade. Shape and motion from image streams under orthography:
A factorization method. IJCV, 9(2):137-154, November 1992.
Motion field
• The motion field is the projection of the 3D
scene motion into the image
Optical flow
• Definition: optical flow is the apparent motion
of brightness patterns in the image
• Ideally, optical flow would be the same as the
motion field
• Have to be careful: apparent motion can be
caused by lighting changes without any actual
motion
• Think of a uniform rotating sphere under fixed lighting
vs. a stationary sphere under moving illumination
 Estimating optical flow

I(x,y,t–1) I(x,y,t)
• Given two subsequent frames, estimate the apparent
motion field u(x,y) and v(x,y) between them
• Key assumptions
• Brightness constancy: projection of the same point looks the
same in every frame
• Small motion: points do not move very far
• Spatial coherence: points move like their neighbors
 The brightness constancy constraint

I(x,y,t–1) I(x,y,t)

Brightness Constancy Equation:

I ( x, y, t − 1) = I ( x + u ( x, y ), y + v( x, y ), t )
Linearizing the right side using Taylor expansion:

I ( x, y, t − 1) ≈ I ( x, y, t ) + I x u ( x, y ) + I y v( x, y )

Hence, I x u + I y v + It ≈ 0
The brightness constancy constraint
I x u + I y v + It = 0
• How many equations and unknowns per pixel?
• One equation, two unknowns

• What does this constraint mean?

∇I ⋅ (u, v) + I t = 0
• The component of the flow perpendicular to the
gradient (i.e., parallel to the edge) is unknown
The brightness constancy constraint
I x u + I y v + It = 0
• How many equations and unknowns per pixel?
• One equation, two unknowns

• What does this constraint mean?

∇I ⋅ (u, v) + I t = 0
• The component of the flow perpendicular to the
gradient (i.e., parallel to the edge) is unknown
gradient
(u,v)
If (u, v) satisfies the equation,
so does (u+u’, v+v’) if ∇I ⋅ (u ' , v' ) = 0 (u’,v’)
(u+u’,v+v’)
edge
The aperture problem

Perceived motion
The aperture problem

Actual motion
The barber pole illusion

http://en.wikipedia.org/wiki/Barberpole_illusion
The barber pole illusion

http://en.wikipedia.org/wiki/Barberpole_illusion
 Solving the aperture problem
• How to get more equations for a pixel?
• Spatial coherence constraint: pretend the pixel’s
neighbors have the same (u,v)
• E.g., if we use a 5x5 window, that gives us 25 equations per pixel

∇I (x i ) ⋅ [u , v] + I t (x i ) = 0
⎡ I x (x1 ) I y (x1 ) ⎤ ⎡ I t (x1 ) ⎤
⎢ I (x ) I (x )⎥ ⎢ I (x )⎥
⎢ x 2 y 2 ⎥ ⎡ ⎤u ⎢ t 2 ⎥
⎢ ⎥ =−
⎢ ! ! ⎥ ⎣v ⎦ ⎢ ! ⎥
⎢ ⎥ ⎢ ⎥
⎢⎣ I x (x n ) I y (x n )⎥⎦ ⎣ I t (x n )⎦
B. Lucas and T. Kanade. An iterative image registration technique with an application to
stereo vision. In Proceedings of the International Joint Conference on Artificial
Intelligence, pp. 674–679, 1981.
 Solving the aperture problem
• Least squares problem:

⎡ I x (x1 ) I y (x1 ) ⎤ ⎡ I t (x1 ) ⎤

⎢ I (x ) I (x )⎥ ⎢ ⎥
⎢ x 2 y 2 ⎥ ⎡u ⎤
⎢ I t ( x 2 ) ⎥
= −
⎢ ! ! ⎥ ⎢⎣ v ⎥⎦ ⎢ ! ⎥
⎢ ⎥ ⎢ ⎥
⎢⎣ I x (x n ) I y (x n )⎥⎦ ⎣ I t (x n )⎦
• When is this system solvable?
• What if the window contains just a single straight edge?

B. Lucas and T. Kanade. An iterative image registration technique with an application to

stereo vision. In Proceedings of the International Joint Conference on Artificial
Intelligence, pp. 674–679, 1981.
Conditions for solvability
• “Bad” case: single straight edge
Conditions for solvability
• “Good” case
 Lucas-Kanade flow
Linear least squares problem
⎡ I x (x1 ) I y (x1 ) ⎤ ⎡ I t (x1 ) ⎤
⎢ I (x ) I (x )⎥
⎢ x 2 y 2 ⎥ ⎡u ⎤
= −
⎢ I (x )⎥
⎢ t 2 ⎥ Ad=b
⎢ ! ! ⎥ ⎢⎣ v ⎥⎦ ⎢ ! ⎥ n×2 2×1 n×1
⎢ ⎥ ⎢ ⎥
I
⎢⎣ x n( x ) I y ( x )
n ⎥⎦ I (
⎣ t n ⎦ x )
T T
Solution given by ( A A)d = A b

⎡∑ I x I x ∑I I ⎤ ⎡u ⎤
x y ⎡∑ I x I t ⎤
⎢ ⎥ ⎢ ⎥ = −⎢ ⎥
⎢⎣∑ I x I y ∑I yIy ⎥
⎦ ⎣v ⎦ ⎢⎣∑ I y I t ⎥⎦

The summations are over all pixels in the window

B. Lucas and T. Kanade. An iterative image registration technique with an application to

stereo vision. In Proceedings of the International Joint Conference on Artificial
Intelligence, pp. 674–679, 1981.
Lucas-Kanade flow
⎡∑ I x I x ∑I Ix y ⎤ ⎡u ⎤ ⎡∑ I x I t ⎤
⎢ ⎥ ⎢ ⎥ = −⎢ ⎥
⎢⎣∑ I x I y ∑I Iy ⎦ ⎣v ⎦
y⎥ ⎢⎣∑ I y I t ⎥⎦

• Recall the Harris corner detector: M = ATA is

the second moment matrix
• We can figure out whether the system is
solvable by looking at the eigenvalues of the
second moment matrix
• The eigenvectors and eigenvalues of M relate to edge
direction and magnitude
• The eigenvector associated with the larger eigenvalue points
in the direction of fastest intensity change, and the other
eigenvector is orthogonal to it
Recall: second moment matrix
Classification of image points using eigenvalues
of the second moment matrix:
λ2 “Edge”
λ2 >> λ1 “Corner”
λ1 and λ2 are large,
λ1 ~ λ2

λ1 and λ2 are small “Flat” “Edge”

region λ1 >> λ2

λ1
Uniform region

– gradients have small magnitude

– small λ1, small λ2
– system is ill-conditioned
Edge

– gradients have one dominant direction

– large λ1, small λ2
– system is ill-conditioned
High-texture or corner region

– gradients have different directions, large magnitudes

– large λ1, large λ2
– system is well-conditioned
Example of optical flow estimation

* From Khurram Hassan-Shafique CAP5415 Computer Vision 2003

Example of optical flow estimation

* From Khurram Hassan-Shafique CAP5415 Computer Vision 2003

 Iterative Refinement

• Iterative Lukas-Kanade Algorithm

1. Estimate displacement at each pixel by solving
Lucas-Kanade equations
2. Warp I(t) towards I(t+1) using the estimated flow field
- Basically, just interpolation
3. Repeat until convergence

33
* From Khurram Hassan-Shafique CAP5415 Computer Vision 2003
Multi-resolution estimation

* From Khurram Hassan-Shafique CAP5415 Computer Vision 2003

Multi-resolution estimation

* From Khurram Hassan-Shafique CAP5415 Computer Vision 2003

 Coarse-to-fine optical flow estimation

run iterative L-K

warp & upsample

run iterative L-K

.
.
.

image J1 image
image I2

Gaussian pyramid of image 1 (t) Gaussian pyramid of image 2 (t+1)

Multi-resolution Lucas Kanade Algorithm
Optical Flow Results

* From Khurram Hassan-Shafique CAP5415 Computer Vision 2003

Optical Flow Results

* From Khurram Hassan-Shafique CAP5415 Computer Vision 2003

Errors in Lucas-Kanade
• The motion is large (larger than a pixel)
• Coarse-to-fine estimation
• Iterative refinement
• Exhaustive neighborhood search (feature matching)
• A point does not move like its neighbors
• Motion segmentation
• Brightness constancy does not hold
• Exhaustive neighborhood search with normalized correlation
 Large displacement optical flow
Start with something similar to Lucas-Kanade
+ gradient constancy
+ energy minimization with smoothing term
+ region matching
+ keypoint matching (long-range)

Region-based +Pixel-based +Keypoint-based

Large displacement optical flow, Brox et al., CVPR 2009
Feature tracking
• If we have more than two images, we can
track a feature from one frame to the next by
following the optical flow
• Challenges
• Finding good features to track
• Adding and deleting tracks
 Shi-Tomasi feature tracker
• Find good features using eigenvalues of second-
moment matrix
• Key idea: “good” features to track are the ones whose motion can
be estimated reliably
• From frame to frame, track with Lucas-Kanade
• This amounts to assuming a translation model for frame-to-frame
feature movement
• Check consistency of tracks by affine registration
to the first observed instance of the feature
• Affine model is more accurate for larger displacements
• Comparing to the first frame helps to minimize drift

J. Shi and C. Tomasi. Good Features to Track. CVPR 1994.

Tracking example

J. Shi and C. Tomasi. Good Features to Track. CVPR 1994.

Summary of KLT tracking
• Find a good point to track (harris corner)

• Use intensity second moment matrix and

difference across frames to find displacement

• Iterate and use coarse-to-fine search to deal with

larger movements

• When creating long tracks, check appearance of

registered patch against appearance of initial
patch to find points that have drifted
Implementation issues
• Window size
– Small window more sensitive to noise and may miss larger
motions (without pyramid)
– Large window more likely to cross an occlusion boundary
(and it’s slower)
– 15x15 to 31x31 seems typical

• Weighting the window

– Common to apply weights so that center matters more
(e.g., with Gaussian)
Why not just do local template matching?

• Slow (need to check more locations)

• Does not give subpixel alignment (or becomes

much slower)
– Even pixel alignment may not be good enough to
prevent drift

• May be useful as a step in tracking if there are

large movements
 Summary
• Major contributions from Lucas, Kanade, Shi, Tomasi
– Tracking feature points
– Optical flow

• Key ideas
– By assuming brightness constancy, truncated Taylor
expansion leads to simple and fast patch matching across
frames
– Coarse-to-fine registration