问题
What I have done: To create a novel view(right image) from left image(given), I used the formula for pure translation between views(from Zissermann book) as:
x' = x + K.t/Z
x'= [u' v' 1]; x = [u v 1]; K = [f 0 cx; 0 f cy;0 0 1]; t = [t1 t2 t3]^T; Z - depth of pixel in left image
The images and camera matrix were taken from Middlebury stereo 2014 dataset.
By implementing this, I have an image with holes(black regions) due to disocclusions.
What I need to do :
To fill these holes, several algorithms exist which modify the depth map of right view prior to warping.
Can you tell me how I can find depth map of my synthesized(right) view ?
results matlab Above is the result I have got till now, please help!
回答1:
My answer actually is a workaround from Priyamvadha's previous question and my answer to it. If you have extrinsic/intrinsic parameters and 3D points, consider to reverse the process.
transform (as you did with my previous answer) all the 3D points in the right camera's reference system (use extrinsic R and t, "reversing" the transformation)
now that you have all the 3D coordinates in that system, remember that the Z value is strongly linked with the disparity
for each point and its Z coordinate, the disparity should be equal to
D = (b*f)/Z
with b the baseline and f the focal length from intrinsics.
You should have obtained syntethized disparity for your syntetized image. Link each disparity value with the correspondant projected point in the syntethized image.
Yes, I could have join all the passage and give you a unique formula, but it would not have meant anything for you.
PS if you have no point, there will be no depth for black holes in the image.
来源：https://stackoverflow.com/questions/39898247/calculate-depth-map-of-right-image