Download presentation
Presentation is loading. Please wait.
Published byLamar Watchorn Modified over 10 years ago
4
Some problems...
5
Lens distortion Uncalibrated structure and motion recovery assumes pinhole cameras Real cameras have real lenses How can we correct distortion, when original calibration is inaccessible?
6
1. Even small amounts of lens distortion can upset uncalibrated structure from motion 2. A single distortion parameter is enough for mapping and SFX accuracy 3. Including the parameter in the multiview relations changes the 8-point algorithm from 4. You can solve such “Polynomial Eigenvalue Problems” 5. This is as stable as computation of the Fundamental matrix, so you can use it all the time.
7
E ven small amounts of lens distortion can upset uncalibrated structure from motion—
8
A map-building problem Input movie – relatively low distortion Plan view: red is structure, blue is motion (a) (b)
9
Effects of Distortion Input movie – relatively low distortion Recovered plan view, uncorrected distortion (a) (c)
10
Does distortion do that? Distortion of image plane is conflated with focal length when the camera rotates [From: Tordoff & Murray, ICPR 2000]
11
Distortion correction in man-made scenes
12
Distortion correction in natural scenes In natural images, distortion introduces correlations in frequency domain Choose distortion parameters to minimize correlations in bispectrum Less effective on man- made scenes.... [Farid and Popescu, ICCV 2001]
13
Distortion correction in multiple images Multiple views, static scene Use motion and scene rigidity [Zhang, Stein, Sawhney, McLauchlan,...] Advantages: Applies to man-made or natural scenes Disadvantages: Iterative solutions|require initial estimates
14
A single distortion parameter is accurate enough for map-building and cinema post production—
15
Modelling lens distortion x: x ero x ed no x ious e x perimental artifa x p: p erfect p inhole p erspective p ure x p p x KnownUnknown
16
Single-parameter models
17
Single-parameter modelling power Single-parameter model Radial term only Assumes distortion centre is at centre of image A one-parameter model suffices
18
A direct solution for
19
Look at division model again
21
>> help polyeig POLYEIG Polynomial eigenvalue problem. [X,E] = POLYEIG(A0,A1,..,Ap) solves the polynomial eigenvalue problem of degree p: (A0 + lambda*A1 +... + lambda^p*Ap)*x = 0. The input is [etc etc...] >> A quick matlab session
22
Algorithm
23
T his is as stable as computation of the fundamental matrix, so you can use it all the time—
24
Performance: Synthetic data 00.20.40.60.81 -0.4 -0.3 -0.2 -0.1 0 Noise (pixels) Computed Stable – small errorbars Biased – not centred on true value
25
Analogy: Linear ellipse fitting True Data Fitted: 10 trials Best-fit line
26
Performance: Synthetic data
27
Performance: Real sequences
28
250 pairs Low distortion Linear estimate used to initialize nonlinear Number of inliers changes by [-25..49]
29
Conclusions
31
Environment matting In: magnifying glass moving over background Out: same magnifying glass, new background
32
Environment matting: why? Learn –light-transport properties of complex optical elements Previously –Ray tracing geometric models –Calibrated acquisition Here –Acquisition in situ
33
Image formation model Purely 2D-2D –Optical element performs weighted sum of (image of) background at each pixel –suffices for many interesting objects –separate receptive field for each output pixel –Environment matte is collection of all receptive fields—yes, it’s huge.
34
Image formation model
35
Step 1: Computing background Input: Mosaic: Clean plate: Point tracks:
36
Step 2: Computing w... Input:
37
Computing w(x,y,u,v) at a single (x,y)
38
Assume w i independent
39
Composite over new background
40
A more subtle example Input: Two images Moving camera Planar background - Need priors
41
Window example
42
Discussion Works well for non-translucent elements –need to develop for diffuse Combination assumes independence –ok for large movements: “an edge crosses the pixel” Need to develop for general backgrounds
44
A Clustering Problem Watch a movie, recover the cast list –Run face detector on every frame –Cluster faces Problems –Face detector unreliable –Large lighting changes –Changes in expression –Clustering is difficult
45
A sample sequence
46
Detected faces
47
Face positions
48
Lighting correction
49
Clustering: pairwise distances Raw distance
50
Clustering: pairwise distances Transform-invariant distance
51
Clusters: “tangent distance”
52
Clusters: Bayesian tangent distance
53
Conclusions Extend to feature selection, texton clustering etc Remove face detector
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.