CS 791 E Computer Vision

Programming Assignment # 4

(Eight-Point Algorithm Implementation)

The program

• Download  Makefile ,  eightPoint.h ,  eightPoint.cpp ,  and eightPointDrive.cpp  .
• Goto Dr. Bebis's homepage to download the data file.
• Please note:  I use scp to from all those data into the current directory, that is, the data files reside in the subdirectory "eight_point". No file/dir names changed. When running the program, the corresponding data file automatically is accessed.
• Type:  ./make to produce the executalbe  eightPointDrive
• Type: ./eightPointDrive   <name> <step>    . 
• <name>:   is corridor,   jig ,  or house.  (jig---calibration_jig).
• <step>:    can be 1, 2,....or 40.  
• In the code, I use following number of points : 9, 10,.....3/4 * (total number of the points), if step = 1. If step = n, then using: 9, 9+n, 9 +2n... number of points for each step in calculating fundamental parameters / condition numbers/average condition numbers.
• For each number N of points, the program run 100 times with different (random) N points each time just as Hartley's paper proposed.
• In class EightPoint in eightPoint.cpp file, the member function epipolarG() does everything and it is easy to be read. All the other members fulful the functions just as their "names" suggest.
• I just used Opencv to load/show/save images and draw epipolar lines.
• Condition numbers and average errors are calculated following Hartley's paper and recorded in datafile and plotted via matlab.
• Some errors (print error?) in Hartley's paper in calculating point-epipolar-line distance (average error) found and corrected.
  

Results:

1. Condition Number

• For the House (corresponding to Hartley's fig.6)

• For the "calibation_jig"

• For the "corridor"

2. Average Errors (corresponding to Hartley's fig. 8)

• For the "house"

• For the "calibration_jig"

• For the "corridor"

3.  Epipolars

• For the "house" 
• left image:   173.154    -8064.71
• right image:  74.7648,   -3601.55
• For the "calibration_jig"
• left image:  1041.2  206.131
• right image:  1035.56  206.44
• For the "corridor"
• left image:  256.427  184.601
• right image:  255.68  183.693

4.  Epipolar lines (that overlay the images)

4.1 For the "house"

• left image:

• right image:

4.2.  For the "calibration_jig"

• left image

• right image

4.3. For the "corridor"

• left image:
  

• 

• right image

5. Problem:

In the case of the "house", every time after running, I got noticeable results as in the epipolar, fundamental matrix, to some degree.   The other cases are stable.