Strange black ragion of stereo matching method 'StereoSGBM'

Python
,
1

I use the StereoSGBM method in OpenCV to generate the disparity map of a pair of stereo images.
However, there is a strange black zone at the left of the disparity map. Can anyone give some help?

import cv2
name_l = ('KITTI/depth/train/2011_09_26_drive_0001_sync/'
          'image_02/data/0000000000.png')
name_r = ('KITTI/depth/train/2011_09_26_drive_0001_sync/'
          'image_03/data/0000000000.png')
img_l = cv2.imread(name_l)
img_r = cv2.imread(name_r)

window_size = 3
stereo = cv2.StereoSGBM_create(minDisparity=0,
                               numDisparities=128,
                               blockSize=window_size,
                               P1=3 * 4 * window_size**2,
                               P2=3 * 32 * window_size**2,
                               disp12MaxDiff=1,
                               preFilterCap=63,
                               uniquenessRatio=10,
                               speckleWindowSize=100,
                               speckleRange=32,
                               mode=cv2.StereoSGBM_MODE_HH)
disp = stereo.compute(img_l, img_r).astype('float32') / 16.0

cv2.imshow('left', img_l)
cv2.imshow('right', img_r)
cv2.imshow('disparity', (disp - 0) / 128)
cv2.waitKey()
cv2.destroyAllWindows()

Selection_003

2

welcome.

you didn’t do anything wrong. that effect is also reproducible with the sample samples/python/stereo_match.py and its default data.

I believe the width of that area is directly proportional or even equal to (minDisparity + numDisparities), which is “maxDisparity” if such a parameter existed. perhaps that helps you in speculating, along with the papers cited in the documentation.