Star Matching from Images

1
Star Matching from Images

• Computer Vision Final Project
• by Wei-Chao Chen

2
Problem Statement

• Goal
• Take pictures in the night sky.
• Match the pictures with star catalog.
• Condition Requirements
• Camera parameter are unknown a priori.
• Pictures are free with large bright objects
  (e.g., moon).
• Stars in each pictures are related by linear
  translation.

3
Sample Picture
4
Difficulty in the Problem

• Too many free variables..
• Camera parameters, time, location, orientation
  when taking the picture.
• Matching can be ambiguous..
• Except for relative brightness and color, stars
  have no other salient features.
• Sensitive to noise..
• Bad pixels can be regarded as stars.

5
A Solution

• Match by looking at similar triangles between two
  images.
• Each image with N stars generates C(N, 3)
  triangles.
• Each star can have up to N(N-1)/2 matches.
• The stars pair with higher vote of matches are
  considered as matches.

6
A Solution
4
2
a1
1
T1(a1, b1, c1) gt(a1/c1, b1/c1)
3
b1
c1
1
3
4
2
The aspect ratio of each triangle forms a
2D tuple and matching is done at the tuple space.
7
Results

• Implementation
• ANSI C/C
• Steps
• Star Detection-Matching-Output
• Running Speed
• Generally below 5 seconds on PII PCs (including
  Disk I/O)

8
Results - Star Detection
One Input Image
Detected Stars
9
Results - RotationTranslation
Red Image 1 Green Image 2 Yellow Match
10
RotateTrans.ScalingClipping
Red Image 1 Green Image 2 Yellow Match
Overlapped Region
11
Rigid Trans. Small Noise
Red Image 1 Green Image 2 Yellow Match
Noise
Noise
12
Algorithm Problems
Speckles on image 2 results in algorithm instabili
ty.
Red Image 1 Green Image 2 Yellow Match
13
Algorithm Problems
Radial Distortion caused the algorithm to break.
Red Image 1 Green Image 2 Yellow Match
14
References

• Valdes et al., FOCAS Automatic Catalog Matching
  Algorithms, in Publications of Astronomical
  Society of the Pacific, 1995 November.
• Griesen et al., Representation of World
  Coordinates in FITS, www.cv.nrao.edu/fits/documen
  ts/wcs/wcs.html