Visualize object sensor's data (rviz)

[soo4826]

Hi!

I want to extract 3D BBOX data using ros-bridge.
Before that, I need to visualize object's tf data via rviz.
But I dont know how to add object's tf data from object sensor... (/carla/ego_vehicle/objects)
When I launched pcl_recorder.launch, I found that topic /carla/ego_vehicle/objects published.
There is no node which subscribes that topic.

So, If I want to visualize object via rviz, do I have to make tf listener?

[joel-mb]

Hi @soo4826,
From which objects do you want to extract 3D bounding box data? Right now, there are two ways to retrieve this information from traffic participants (i.e., pedestrians and vehicles).

• Using the marker sensor (in your case published in /carla/markers). You can directly visualize this topic on rviz.
• Using the object sensor (in your case published in /carla/ego_vehicle/objects which contain information about the object shape and dimensions (not sure if there is a plugin to visualize this information directly in rviz).

You don't need to create any tf listener.

[soo4826]

Hi @soo4826,
From which objects do you want to extract 3D bounding box data? Right now, there are two ways to retrieve this information from traffic participants (i.e., pedestrians and vehicles).

• Using the marker sensor (in your case published in /carla/markers). You can directly visualize this topic on rviz.
• Using the object sensor (in your case published in /carla/ego_vehicle/objects which contain information about the object shape and dimensions (not sure if there is a plugin to visualize this information directly in rviz).

You don't need to create any tf listener.

Thanks for your reply joel!

I want to make KITTI ground truth data of 3D object tracking, so I want to get object's 3D BBOX data.
To do that, I want to get each object's 3D BBOX data and tracking_id.

Also I already try to use object sensor. But there are some problems.

• The data(/carla/ego_vehicle/objects) have coordinates based on map coordinates, not 3D BBOX data based on ego_vehicles.
• The data from sensor is not BBOX data, they are tf data.

So, I'll try to use marker sensor to visualize objects on carla.

And Is there a solution to make GT data from carla..?

[joel-mb]

@soo4826 The id field of the Marker message also relates to the CARLA actor id, so this may help for your tracking purposes.

And Is there a solution to make GT data from carla..?

Regarding bboxes, a new feature to get bounding bounding was added in 0.9.10 (https://carla.readthedocs.io/en/0.9.11/python_api/#carla.World.get_level_bbs and https://carla.org/2020/09/25/release-0.9.10/#global-bounding-box-accessibility). Unfortunately, this functionality has not been ported to the ROS bridge yet. Probably, this will be integrated in the ROS bridge as a new pseudo sensor or service.

On the other hand, there are other sensors that provide ground truth data in CARLA (e.g, semantic segmentation camera, semantic lidar). Here you can take a look to the CARLA sensor reference: https://carla.readthedocs.io/en/latest/ref_sensors/