rectify_uncalibrated_bug.py

Julien Rebetez, 2012-09-19 12:16 am

Download (2.6 kB)

 
1
# Python script triggering a bug in rectifyUncalibrated's outliers filtering
2
import cv2
3
import cv2.cv as cv
4
import numpy as np
5
import random
6




7
def wrap_rectify_uncalibrated(x1, x2, F, imsize, threshold):





8
    """Wrapper that converts x1, x2 into 1d array suitable for





9
       rectifyUncalibrated"""





10
    # x1, y1, x2, y2, x3, y3





11
    x1_1d = np.empty((2*x1.shape[1],), dtype=float)





12
    x1_1d[0::2] = x1[0,:]





13
    x1_1d[1::2] = x1[1,:]





14




15
    x2_1d = np.empty((2*x2.shape[1],), dtype=float)





16
    x2_1d[0::2] = x2[0,:]





17
    x2_1d[1::2] = x2[1,:]





18




19
    return cv2.stereoRectifyUncalibrated(x1_1d, x2_1d, F, imsize,





20
                                         threshold=threshold)





21




22




23
F = np.array([[-0.000005,  0.00007 ,  0.001381],





24
    [ 0.000138, -0.000006, -0.187487],





25
    [-0.009234,  0.160388,  1.      ]])





26




27
x1 = np.array([[119.61484528,  129.7978363 ,  120.65196991,  114.99479675,





28
                59.63290787,   53.01756287,  117.4389801 ,  159.73094177,





29
                145.67852783,   19.50421524],





30
               [61.97002411,   66.56116486,   64.01503754,   85.84388733,





31
                14.17597294,  111.90859985,   53.89838409,   79.10702515,





32
                86.399086  ,  133.72361755]])





33




34
x2 = np.array([[ 121.,  130.,  122.,  117.,  128.,   63.,  118.,  161.,  146.,





35
                21.],





36
               [  62.,   67.,   65.,   84.,   16.,  106.,   55.,   81.,   87.,





37
                122.]])





38




39
imsize = (266, 200)





40




41
threshold = 3





42
# Some hardcoded outlying points





43
outliers1 = np.array([[1, 10, 50, 100],





44
                      [1, 30, 60, 90]], dtype=float)





45
outliers2 = np.array([[200, 250, 300, 340],





46
                      [100, 110, 110, 110]], dtype=float)





47




48
# Sanity check : the outliers should be real outliers





49
lines1 = np.zeros((3, outliers1.shape[1]), dtype=float)





50
lines2 = np.zeros((3, outliers2.shape[1]), dtype=float)





51




52
cv.ComputeCorrespondEpilines(cv.fromarray(outliers1), 1, cv.fromarray(F),





53
                             cv.fromarray(lines1))





54
cv.ComputeCorrespondEpilines(cv.fromarray(outliers2), 1, cv.fromarray(F.T.copy()),





55
                             cv.fromarray(lines2))





56
for i in xrange(outliers1.shape[1]):





57
    assert abs(outliers1[0,i]*lines2[0,i] +





58
               outliers1[1,i]*lines2[1,i] + lines2[2,i]) > threshold





59
    assert abs(outliers2[0,i]*lines1[0,i] +





60
               outliers2[1,i]*lines1[1,i] + lines1[2,i]) > threshold





61




62
# 1. Without outliers





63
success, H1, H2 = wrap_rectify_uncalibrated(x1, x2, F, imsize, threshold)





64




65
# 2. With outliers





66
ox1 = np.c_[outliers1, x1]





67
ox2 = np.c_[outliers2, x2]





68




69
success, oH1, oH2 = wrap_rectify_uncalibrated(ox1, ox2, F, imsize, threshold)





70




71
assert np.allclose(oH1, H1)





72
assert np.allclose(oH2, H2)





73