discontinuities in pointcloud after reprojectImageTo3D

asked 2015-09-21 03:31:36 -0500

eriba gravatar image

Hi guys,

I'm doing the implementation of Efficient Large Scale Multi-View Stereo for Ultra High Resolution Image Sets in order to perform dense reconstruction with DAISY features.

The point is that given two rectified images

image description

I compute the disparity map using the algorithm described in the paper

image description

Finally, using reprojectImageTo3D() with disparity map and Q matrix obtained from previous image rectification I get the following pointcloud

image description

The above pointcloud has discontinuities in the camera Z (or disparity level).

I asked around and someone told me that I could have a data type error. My disparity map and Q matrices are CV_8U and CV_32F respectively. In reprojectImageTo3D I'm using the default parameters: reprojectImageTo3D(disparity_map, points3d, Q, false, -1);

Any suggestion?

edit retag flag offensive close merge delete