Revision history [back]

You can do this with openni_camera alone with the (undocumented) use_indices parameter. See my answer to this question for details. The advantage over pointcloud_to_laserscan type approaches would be that you cut out the unused information as early as possible in the pipeline, i.e. the driver doesn't spend CPU cycles projecting a bunch of depth values to 3D points that never get used anyway. Depending on how much you need to drop your CPU usage that might be important (it was in my application which is why I implemented that feature).