i'm trying to learn camera pose estimation and have been been through quite a lot of material. I want to calculate the pose of the camera using QR codes (quick response code) set up like this on an standard A4:
A0 A1
B0 B1
C0 C1
- Took some photos of the QR codes which i printed out on A4 paper, which have dimensions 210mm x 297 mm
This is a short code i wrote in MATLAB to get the coordinates of the world points where i assume the bottom left corner is (0,0)
n=2; %number of qr in x direction
m=3; %number of qr in y direction
x_FE=50; %distance from left edge to point1 in mm
y_FE=270; %distance from bottom to top qr code
sqr_dist=27.6; %distance between qr codes [mm]
sqr_size=47.4; %size of qr code height or width [mm]
QRnames=['A0'; 'A1';'B0'; 'B1';'C0'; 'C1'];
for i=1:n %number of squares in x directions
x1(i)=x_FE+sqr_dist*(i-1);
x2(i)=x_FE+sqr_size+sqr_dist*(i-1);
end
for j=1:m %number of squares in y directions
y1(j)=y_FE-sqr_dist*(j-1);
y2(j)=y_FE-sqr_size-sqr_dist*(j-1);
end
n=1;
for k=1:3 %qr in y direction
for j=1:2 %qr in x direction
p1=[x1(j),y1(k)]
p2=[x1(j),y2(k)]
p3=[x2(j),y2(k)]
p4=[x2(j),y1(k)]
varname1=(['p' int2str(i)]);
varname2=(['x' int2str(j)]);
varname3=(['y' int2str(j)]);
eval([QRnames(n) int2str(j-1) 'w=[p1 p2 p3 p4]' ';' ]);
n=n+1;
end
end
Ive also checked the coordinates by measuring with a ruler
Using a Python script called ZBar which is a QR code reader. i can extract the image coordinates from the QR codes which is embedded in the QR code itself (i think).
The coordinates get saved in a .txt file which is setup this way
QRcode name x1 y1 x2 y2 x3 y3 x4 y4
Here are the images: https://i.stack.imgur.com/Z5f0Q.jpg
here are some of the results coordinates:
image2
A0 452 495 454 878 836 878 834 493
B0 455 1099 453 1479 836 1477 836 1099
B1 1056 1099 1056 1477 1442 1481 1442 1096
C0 449 1702 444 2091 831 2093 833 1700
C1 1056 1701 1056 2095 1447 2102 1445 1706
A1 1057 491 1057 878 1443 873 1445 485
image3
A1 1046 375 1051 737 1417 725 1412 364
B0 479 955 482 1316 848 1310 843 950
A0 475 389 477 747 841 740 835 379
B1 1053 946 1058 1307 1426 1302 1421 938
C1 1061 1518 1066 1895 1444 1895 1432 1517
C0 481 1528 479 1901 851 1896 848 1521
image8
A0 528 484 483 738 879 728 891 476
A1 1104 472 1109 728 1512 716 1473 463
B0 452 906 389 1240 852 1237 870 899
B1 1113 896 1119 1234 1595 1226 1540 884
C0 343 1472 248 1957 809 1953 839 1465
C1 1126 1463 1139 1952 1717 1955 1633 1461
image9
A0 2040 560 1741 553 1818 812 2138 811
B0 1562 548 1242 537 1282 806 1625 809
A1 2200 977 1866 980 1969 1311 2325 1301
C0 1047 528 691 515 689 797 1073 804
B1 1665 979 1308 984 1359 1328 1751 1318
C1 1090 983 685 985 677 1359 1119 1340
image10
A1 562 823 943 814 993 517 643 531
B1 1169 805 1571 793 1574 484 1201 504
A0 393 1409 835 1410 907 1010 506 1017
C1 1819 782 2268 768 2209 449 1801 471
B0 1098 1408 1571 1410 1570 997 1146 1003
C0 1862 1415 2390 1420 2309 982 1833 991
Here is information about the camera i used (ipod touch 5.gen):
ExposureTime: 0.0167
FNumber: 2.4000
ExposureProgram: 'Normal program'
ISOSpeedRatings: 32
ExifVersion: [4x1 double]
DateTimeOriginal: '2013:11:23 11:44:52'
DateTimeDigitized: '2013:11:23 11:44:52'
ComponentsConfiguration: 'YCbCr'
ShutterSpeedValue: 5.9069
ApertureValue: 2.5261
BrightnessValue: 5.9402
MeteringMode: 'Pattern'
Flash: [1x148 char]
FocalLength: 3.3000
SubjectArea: [4x1 double]
MakerNote: [196x1 double]
SubsecTimeOriginal: '317'
SubsecTimeDigitized: '317'
FlashpixVersion: [4x1 double]
ColorSpace: 'sRGB'
CPixelXDimension: 1936
CPixelYDimension: 2592
SensingMethod: 'One-chip color area sensor'
SceneType: 'A directly photographed image'
ExposureMode: 'Auto exposure'
WhiteBalance: 'Auto white balance'
DigitalZoomRatio: 1.0021
FocalLengthIn35mmFilm: 33
SceneCaptureType: 'Standard'
UnknownTags: [3x1 struct]
and info about the first test image
info=imfinfo('testimage1002.jpg')
info =
Filename: 'D:\dropbox\Dropbox\campose\testimage_1\testimage1002.jpg'
FileModDate: '23-Nov-2013 11:45:16'
FileSize: 1188331
Format: 'jpg'
FormatVersion: ''
Width: 1936
Height: 2592
BitDepth: 24
ColorType: 'truecolor'
FormatSignature: ''
NumberOfSamples: 3
CodingMethod: 'Huffman'
CodingProcess: 'Sequential'
Comment: {}
Make: 'Apple'
Model: 'iPod touch'
Orientation: 1
XResolution: 72
YResolution: 72
ResolutionUnit: 'Inch'
Software: '7.0.4'
DateTime: '2013:11:23 11:45:16'
YCbCrPositioning: 'Centered'
DigitalCamera: [1x1 struct]
GPSInfo: [1x1 struct]
ExifThumbnail: [1x1 struct]
i read the .txt files and set them up like this
txt=fopen('testimage1002.txt');
datascan01=textscan(txt,'%s%f%f%f%f%f%f%f%f');
QRname01=[datascan01{1}];
x1=[datascan01{2}];
y1=[datascan01{3}];
% point 2
x2=[datascan01{4}];
y2=[datascan01{5}];
% point 3
x3=[datascan01{6}];
y3=[datascan01{7}];
% point 4
x4=[datascan01{8}];
y4=[datascan01{9}];
p1=0;p2=0;p3=0;p4=0;
for n=1:length(datascan01{1})
for i=1:4
for j=1:4
varname1=(['p' int2str(i)]);
varname2=(['x' int2str(i)]);
varname3=(['y' int2str(i)]);
varname4=([QRname01{n}]);
eval(['p' int2str(i) '=[x' int2str(i) '(n)' ' y' int2str(i) '(n)];'])
eval([varname4 '=[p1 p2 p3 p4]' ';' ]);
end
end
end
Now, i have image coordinates and world coordinates and want to find how the camera was standing when the image was captured.
ive looked at Jean- Yves Bouget's Camera Calibration Toolbox but couldnt figure out how to use the functions using QR codes
also looked at 'Finding optimal rotation and translation between corresponding 3D points' (http://nghiaho.com/?page_id=671)
But the resulting rotation matrix, translation vector and rmse seemed wrong. (got rmse=440?)
Calibration matrix
%Image Resolution: 2592 x 1936 pixels
ImgRes=[2592 1936];% pixels
Focallength= 3.3; % mm
%--- Calibration Results --- % Focals (pixels) -
Fx= 2329.13;
Fy= 2324.04;
% Optical center -
Cx= 1307; Cy= 950.382;
%Radial distortion (Brown's Model) K1= 0.236384;
K2= -0.680501;
K3= 0;
P1= 0.000319648;
P2= 0.000751849;
%pixel size fx=f/sx
sx=Focallength/Fx;
sy=Focallength/Fy;
distortion = [K1 K2 K3 P1 P2];
Any ideas on how i can do this (hopefully in matlab)? all help is appreciated
- Jonathan Brian
