0
votes

I am trying to rectified two stereo images using cv2.stereoRectify after finding a camera matrix of two cameras (K1,D1,K2,D2,R,T). After that, I got R1, R2, P1, P2, Q, roi1, roi2 and use these parameters in cv2.initUndistortRectifyMap to get left rectified-image and right-rectified image. But after using this method the rectified images are not good. I using alpha = -1. Here is my code:

R1, R2, P1, P2, Q, roi1, roi2 = cv2.stereoRectify(cal_data.camera_model.get('M1'), cal_data.camera_model.get('dist1'), cal_data.camera_model.get('M2'), cal_data.camera_model.get('dist2'),
                                                  (960, 544), cal_data.camera_model.get('R'), cal_data.camera_model.get('T'), alpha=-1)

print(R1, R2, P1, P2)

leftFrame = cv2.imread('/home/nikhil_m/Pictures/Webcam/2019-07-29-171837.jpg')
rightFrame = cv2.imread('/home/nikhil_m/Pictures/Webcam/2019-07-29-171809.jpg')

leftFrame =  cv2.resize(leftFrame,(960,544))
rightFrame = cv2.resize(rightFrame, (960, 544))

leftMapX, leftMapY = cv2.initUndistortRectifyMap(cal_data.camera_model.get('M1'), cal_data.camera_model.get('dist1'), R1, P1, (960,544), cv2.CV_32FC1)
left_rectified = cv2.remap(leftFrame, leftMapX, leftMapY, cv2.INTER_LINEAR, cv2.BORDER_CONSTANT)
rightMapX, rightMapY = cv2.initUndistortRectifyMap(cal_data.camera_model.get('M2'), cal_data.camera_model.get('dist2'), R2, P2, (960,544), cv2.CV_32FC1)
right_rectified = cv2.remap(rightFrame, rightMapX, rightMapY, cv2.INTER_LINEAR, cv2.BORDER_CONSTANT)

Is there any better way to rectified images or I am doing something wrong.Please help

Edit : Calibrated image. enter image description here

Original image: enter image description here enter image description here

Camera matrix:

Intrinsic_mtx_1 [[1.22248627e+03 0.00000000e+00 5.24929333e+02]
 [0.00000000e+00 1.32603348e+03 4.99669610e+01]
 [0.00000000e+00 0.00000000e+00 1.00000000e+00]]

dist_1 [[ 0.09850468  1.08533383 -0.10682535  0.01777223 -3.39061053]]

Intrinsic_mtx_2 [[1.07148978e+03 0.00000000e+00 4.21476300e+02]
 [0.00000000e+00 1.09912897e+03 2.61293969e+02]
 [0.00000000e+00 0.00000000e+00 1.00000000e+00]]

dist_2 [[-0.15751877 -0.12428592 -0.01325468  0.02449842  3.72130512]]

R [[ 0.89624385 -0.12740274 -0.42487116]
 [ 0.14523621  0.98934946  0.00969995]
 [ 0.41911026 -0.0704002   0.90520186]]

T [[16.81657383]
 [-5.69906211]
 [ 2.42601652]]

E [[ -2.74088083  -1.99896304  -5.18233385]
 [ -4.87369619   0.87480896 -16.25313836]
 [  7.55012482  15.91139212  -2.25824725]]

F [[ 1.77370674e-06  1.19257563e-06  3.10912454e-03]
 [ 3.07460616e-06 -5.08784373e-07  1.09461230e-02]
 [-6.78615804e-03 -1.05410214e-02  1.00000000e+00]]
1
Please post your source images and ectified images. Are your cameras parralel?Piotr Siekański
@PiotrSiekański mostly they are parrellNikhil Manali
Please also provide M1, M2, dist1, dist2, R, T to replicate your results.Piotr Siekański
@PiotrSiekański Please have a look to these parameters in my edit question.Nikhil Manali

1 Answers

1
votes

Look at your rotation matrix R. If your cameras are parallel it should be close to identity matrix. I used this code to examine the angles between your cameras and it turned out that the angles are [-4.44710871 -24.77842715 9.20475808] in degrees. Therefore I assume that your cameras are not parallel and that's why your approach does not work. See my answer in this thread to find out how to do a proper stereo rectification in your case.