Stereo Calibration and Scene Reconstruction

I want to calibrate a stereo camera system, and reconstruct a 3-D scene just like in the example: http://www.mathworks.com/help/vision/examples/stereo-calibration-and-scene-reconstruction.html but for a much larger scene, not in a small room. a 50*50 meters scene. If I want it to be as accurate as the example, should I capture the checkerboard from different distances? 10,20,30 meters? or the algorithm will work with only close to camera images.

thanks

Answer 1

[Dima Lisin]

Hi Idan,

You should make a large checkerboard and try to take images at various distances. I doubt you can make a checkerboard large enough to be clearly visible at 50 meters, so go as far as you can.

Your accuracy will depend on the resolution of your cameras, and also on the baseline. To measure things this far out, you would need to move the cameras farther apart. You want the disparity to be greater than a pixel for objects at the far end of your distance range.

Keep in mind, that reconstruction accuracy will fall off with the distance.

[Muthu Annamalai]

Nicely put @Dima Lisin.

Also see: Rayleigh resolution https://en.wikipedia.org/wiki/Angular_resolution#Explanation

[Dhinu]

From the above case , the horizontal distance between my camera and nearest point of the object is 3m and i have done the stereo vision calibrotion using the app and exported the stereo parameters to work space.

I had 10 pairs of images placing the checkboard at differebt distances . Now that i have got the disparity map of all the 10 image pair 1544x4096 single in my wotk space , how can i use these datas and retrive the information from stereo paramters to calculate the depth for images and not video. I have refered https://de.mathworks.com/help/vision/ug/depth-estimation-from-stereo-video.html. But i can't get through for the next step of depth estimation for which i would need focal length(which i have in pixel), disparity and baseline which is known but how i am not sure how to use the estimatebaseline fucntion within matlab and how it should be decalred. Also how can i compute/reconstruct 3d point from these values when each one is in differnent datatype and format within and outside the stereo paramter datastore.