Is anyone aware of a Gazebo simulated depth camera that identifies the object each depth pixel belongs to?

A simulated depth camera in GAZEBO senses objects in the simulated scene. It should therefore be possible to to only return the depth image from a simulated depth camera, but also the segmented image based on objects that are visible. Alternatively, is there a simulated camera available that returns a segmented image by objects in the scene. In other word, what is required is an image rendered by object id.

Asked by Bernhard on 2016-09-18 04:01:07 UTC

Comments

I second this question & also need segmentation! A frien moves the camera then all relevant objects one at a time by a fixed offset to an arbitrary position outside the world which creates a constant background is extremely inefficient. Perhaps something could work via lower level rendering. Here is a similar question: http://answers.gazebosim.org/question/12215/scene-labeling-using-a-depth-camera/

Asked by Andrew Hundt on 2017-03-06 15:05:00 UTC

Answers

The functionality you are asking for does not exist in Gazebo. You could implement this feature, and submit a pull request.

There is a somewhat related sensor called LogicalCameraSensor that identifies objects within a view frustum.

Asked by nkoenig on 2017-03-07 10:57:26 UTC