Sources of error to calculate X, Y Z distances using stereo when disparity map seems alright.

This issue is wrt Image scene reconstruction using Stereo Calibration. I have a good (compared to my earlier pointClouds) reconstructed image of my object (wheel). I am able to filter out noise and preserve only the points corresponding to the object under consideration.

I am calculating the X, Y and Z distances to this object by taking a mean of all points under consideration (i.e., mean of the point cloud in each of the 3 dimensions X, Y and Z to get X_mean Y_mean and Z_mean respectively. Euclidean distance is also calculated using these mean values )

For this particular case, the euclidean distance when measured with an inch tape was around 68cm. My result shows 62cm. There is a difference of 60 mm which is not tolerable.

What could the error sources be? The reconstructed image seems very good. During stereo calibration, mean re-projection error is lesser than 0.3 pixels (using a 10MP camera)

I have attached the pointCloud and other relevant files to plot it for your further reference.

Answer 1

[Dima Lisin]

Hi Meghana,

I am glad that you have been able to get a good reconstruction. Regarding the distance, keep in mind that the origin of your coordinate system is at the optical center of the imaging sensor, which is inside your camera. That makes it hard to measure the distance to the object with a tape.

A better way to test the accuracy of your reconstruction is to measure the diameter of the wheel, or the distance between any pair of points on the wheel.

[Meghana Dinesh]

Thanks for the input.