First time here? Check out the FAQ!

Ask Your Question
1

Doubt with disparity values from stereo image

asked Apr 6 '18

Kafan gravatar image

updated Apr 13 '18

If I have a single flat object held parallel to my stereo camera plane. Given nature of pinhole camera, would the disparity values be same across the image (ideal case) or values shift as we move from center of the image towards either of the edge?

Preview: (hide)

1 answer

Sort by » oldest newest most voted
2

answered Apr 6 '18

LBerger gravatar image

updated Apr 16 '18

disparity gives you distance between object and optical center. a constant disparity constant means : a sphere see @Kafan answer : a plane Z=cste

image description

(u1,v1) pixel coordinate in image 1 (u2,v2) pixel coordinate in image 2

u2-u1=d=disparity (images are rectified)

b base

f focal length

mx,my pixels/mm in x or y direction

Preview: (hide)

Comments

@LBerger: From your post, it seems you have mentioned the disparity value (depth info) is obtained is basically displacement from optical center till the object in view. Is my understanding correct? Basically I need the disparity in terms of perpendicular distance with plane of camera rather than center of camera. So I am using a pixel's cosine of cast angle with the camera's optical axis (obtained using FOV parameters of camera). Is this approach correct?

Kafan gravatar imageKafan (Apr 12 '18)edit

No disparity is u2-u1 object coordinates is (u1,v1) in image1 and (u2,v2) in image2 It means that x axis for camera1 and camera2 are colinear

LBerger gravatar imageLBerger (Apr 12 '18)edit

@LBerger: Sorry, I could not make myself clear in earlier post. I have attached a image to explain better. https://imgur.com/a/48jMI There are two real points (x',y') and (x,y), lying on the same plane which is parallel to camera plane. I understand disparity is inversely proportional to actual distance from camera. Now this actual distance is displacement from the real object point to optical center or perpendicular distance to camera plane? If it is actual distance, then disparity of these two points calculated would not be same, inspite of having constant distance from camera plane (perpendicular distance, will mean disparity of (x',y') (x,y) will be found same). I believe in this case I can use cosine of the horizontal angle cast w.r.t. to camera axis to bring all values to same scale.

Kafan gravatar imageKafan (Apr 13 '18)edit
1

@LBerger: I don't think a constant disparity means a sphere. As per literature that I have gone through, it seems disparity is inversely proportional to perpendicular distance from real world point to camera plane. Thus , a constant disparity means a plane held parallely to camera's plane.

Kafan gravatar imageKafan (Apr 16 '18)edit

Right that's not a sphere : Im'm tired it is written in equation Z=cste

LBerger gravatar imageLBerger (Apr 16 '18)edit

Question Tools

1 follower

Stats

Asked: Apr 6 '18

Seen: 678 times

Last updated: Apr 16 '18