Stereo Calibration (HTC Evo 3D)

calibration

asked 2012-12-12 18:59:12 -0600

njaa
112 ●1 ●4 ●8

Hi,

I try to calibrate the stereo camera of my Evo 3D. Therefore I use the stereo_calib.cpp sample and a set of stereo images:

Left:

image description

Right:

image description

Result:

image description

But as you can see the result is weird. When I use the sample images everything works fine:

image description

Somebody can tell me what I'm doing wrong? The images are of size 270x480 while the sample images are 480x640. Could this be a problem? And of course the quality of the second image is worse (HTC Evo problem). But for me it seems there is a general problem. Please help :)

edit retag flag offensive close merge delete

Comments

Can you tell us how you are able to interface with your camera module? I have an HTC Evo 3D camera module and I would love to be able to use it in some applications that I'm developing. Are you using Android?

Der Luftmensch ( 2013-04-15 09:46:33 -0600 )edit

I have HTC Evo V 4G (similar to HTC Evo 3D). could you tell me how did you access both the cameras using OpenCV? I am not getting anywhere for a week or so..

jayrambhia ( 2013-04-15 18:42:58 -0600 )edit

Have you succeeded in camera calibration ? I am struggling with it right this moment and any help will be highly apprechiated. Best Regards.

markotruba ( 2014-07-24 10:21:48 -0600 )edit

add a comment

1

answered 2013-10-09 16:09:52 -0600

sftrytry
11 ●1

I have the same problem. To my understanding, the intrinsic parameters change each time we take pictures.

We need find a way to fix the parameters first.

edit flag offensive delete link

Comments

I encoutered the same problem today so if you have found an explanation or even better a solution to this I would be very grateful if you could share some details with me. Thank you very much.

markotruba ( 2014-07-24 10:19:10 -0600 )edit

add a comment

0

answered 2013-04-15 09:34:52 -0600

GizmoBots
11 ●1 ●1

How did you hook up your HTC Evo 3D to your Cathode ray tube?

edit flag offensive delete link

add a comment

0

answered 2014-08-26 17:39:37 -0600

curtis newton
1 ●1 ●3

anyone could answer this ?

i got exactly the same issue

result change all the time, it works nice with opencv example images but for real life tests, it is super hard to have correct matrix setup

edit flag offensive delete link

Comments

can it be , that it's automatically adjusting the focus ?

berak ( 2014-08-27 00:55:47 -0600 )edit

yes I thought about the lens having different focales but i use web cams so no auto focus :-) I dont understand why calibration is so random, I get different results with the same set of images the best result i can get is a realy smaller image with the corners completely out of whack so for a 320x200 image i get something like 160x100 working region what a waste.. i thought that croping/aligning one image regarding the other would ask less effort and do the trick but i dont know for sure, i need to calc depth/disparity

curtis newton ( 2014-08-27 02:48:33 -0600 )edit

sry @curtis newton, are you referring to bad rectification using webcams? How is your problem related to the Htc smartphone? Do you use the sample rectification program provided with opencv sources to perform disparity maps?

beco ( 2014-08-27 06:02:14 -0600 )edit

Htc ? yes I use opencv's code, and in fact it works depending on some images set it does not work with too many images, or images that are to much alike, but it is a complete random blind guess if it is going to work or not, a non working set becomes working set by just removing a few images but I can't really figure out the golden rule how to make it work, it took me two days tweaking the image sets to have a decent camera matrix with the result like the one above

curtis newton ( 2014-08-28 05:01:15 -0600 )edit

I said Htc because the first question is about calibrating Htc evo's cameras... I don't know any golden rule to obtain a good calibration with normal (no autofocus) webcams. See this answer for some common mistakes about camera calibration.

beco ( 2014-08-28 11:59:53 -0600 )edit

great link, but how can I detect multiple chessboard on one image ? didn't even know one could do that findChessboardCorners only detects one chessboard..

curtis newton ( 2014-08-29 10:34:41 -0600 )edit

@curtis newton the links are not about opencv but they're useful nonetheless. The toolboxes in the links can calibrate cameras using multiple chessboard in a single image but in opencv you can use several pictures of a chessboard in different position. The tips provided in the links are still valid...

beco ( 2014-08-30 13:00:57 -0600 )edit

i tired but opencv only detects one chess board per image

curtis newton ( 2014-09-01 06:28:21 -0600 )edit

add a comment

0

answered 2018-11-16 06:04:10 -0600

Phann
1 ●1

If the autofocus isn't the problem here, a different order of the chessboard corners in the stereo images could be the reason for a high error.

I answered that question here in more detail: link text

edit flag offensive delete link

add a comment

0

answered 2014-06-03 09:54:13 -0600

beco
1 ●4

i have the same problem and i tryed to use my personal camera app without calling the pictureCallback. I essentialy use the previewCallback and i save the arriving frame using Highgui.imwrite to store the jpeg picture. Using my camera app gave me no result and the rectified images are the same of njaa's ones. So the changing of the intrinsic parameters (if it is is the real problem) occurs continously during the preview and not only after the photo has been taken...

edit flag offensive delete link

add a comment

Stereo Calibration (HTC Evo 3D)

Comments

5 answers

Comments

Comments

Links

Question Tools

Stats

Related questions

Stereo Calibration (HTC Evo 3D) edit

Comments

5 answers

Comments

Comments

Links

Question Tools

Stats

Related questions

Stereo Calibration (HTC Evo 3D)