pátek 12. září 2014

Reprojecting stereo images to 3D point cloud

I recently started to experiment with computer vision. There are still very few code examples out on the Internet, but I found Martin Peris's blog where he published few examples on computer vision. Notably his post on 3D reconstruction with OpenCV and Point Cloud Library (PCL) is a very good one and I used the code to do 3D reconstruction of Karlsruhe dataset by Andreas Geiger and his colleagues. The image pair I used was the first one from the sequence named "2010_03_09_drive_0019".

Here are the source images from 2010_03_09_drive_0019 sequence:

Left image:

Right image:

Disparity image obtained with LIBELAS library:

The disparity image actually is a LEFT disparity image, since the LIBELAS library generates a pair of disparity images - in this case probably computed as left-to-right disparity. I did not use the R disparity image that was generated.

Please excuse the orientation of the point cloud since I am a real beginner with PCL and its viewer is hard to control for me (behaves a bit weird :)).

The resulting point cloud images from various angles are below:

All the images can also be found in a single image gallery.

Please note that the original Karslruhe datased is licensed under Creative Commons by-nc-sa 3.0 license so all my derivative images have the same licensing terms.

1 komentář:

  1. hello sir,
    I have been also trying to create 3d point clouds using my own images but could not get good result. thus i consider to use the image as used by you. Please kindly could you tell me what parameter you used for the Q matrix in these images??