Finds areas of interest (horizontal planes, e.g. tables) in a semantically segmented reconstruction of the room. Those areas are saved in a database with mongodb_store as a Table.msg. The table_viewpoint.py script creates viewpoints around the edges of the plane and stores them in the database by updating the message.
- mongodb_store
- ros_numpy
- tf_bag
- pyrsistent==0.16.1
- transforms3d==0.3.1
- open3d_ros_helper==0.2.0.3
- open3d==0.9.0.0
- scikit-image==0.14.5
A Dockerfile that installs the package and necessary dependencies is available.
Start mongodb first. Make sure db_path is set correctly. The standard port for mongodb_store is 62345.
roslaunch --wait mongodb_store mongodb_store.launch db_path:=/home/v4r/mongo_db
Then you can run table_extractor_script.py. The path of the reconstruction file can be changed in the config.yaml, as well as other settings.
The colors parameter defines the colors of the classes from the semantic segmantation. Class_labels and class_names are only of those classes that are relevant, all the other ones are ignored.
table_viewpoint.py will read the message from the database, calculate the viewpoints and then update the message in the database with the viewpoints for every plane.
read_rosbag.py will take a rosbag file that contains recordings of the rgb and depth topic, check if a plane is in the view of the camera and save the rgb and depth images that show a plane in view in a corresponding folder. This can then be used for the next reconstruction of the planes with objects in view.
- Markus Leitner (markus.j.leitner@tuwien.ac.at)