This package semantically labels the points in an RGBD pointcould(typically obtained from a kinect) .
- Input: RGBD Pointloud of type pcl::PointXYZRGB from the topic /camera/rgb/points
Output: Pointcloud of type pcl::PointXYZRGBCamSL on the topic /scene_labler/labeled_cloud
roscd scene_processing rosrun openni_kinect openni_node rosrun scene_processing live_semantic_labeler
- Calibration :
- Live classifier assumes that your Kinect is at a constant height from ground and at a constant vertical orientation. If you are running it for the first time or have changed either of the 2, you need to calibrate. Do rosrun scene_processing GUITransform and rosrun dynamic_reconfigure reconfigure_gui and make the Z axis(blue) align with vertical and origin at ground level. Then select Done. This will save globalTransform.bag which contains the camera transformation and will be used by the live classifer. Note that the labeled pointcloud will have this transform applied.
More information at Scene Understanding for Personal Robots.