Plot Bounding Boxes over 6 Camera Images and Publish to ROS Topics to be Visualized in Rviz

This post was flagged by the community and is temporarily hidden.

It seems like you already found the relevant code snippets. Can you explain your problem related to nuScenes more in detail? I am not familiar with the CenterPoint or ROS codebase.

Thanks for the response! I want to visualize the predicted bounding boxes from the CenterPoint algorithm in the camera images. Now, I have the co-ordinates of the predicted bounding boxes but they are with respect to the Lidar frame. Hence, I need to first convert them to the ego vehicle body frame and then to the camera frame. Kindly suggest how can I achieve this.

Secondly, from the predicted bounding boxes co-ordinates how do I know which camera’s transformation (translation and rotation) matrices should I apply? Additionally, I was wondering why there are calibration information in the calibrated_sensor.json file for every time-frame? Are not these calibration information constant across the entire dataset?

Take a look at https://github.com/nutonomy/nuscenes-devkit/blob/master/python-sdk/nuscenes/nuscenes.py#L555. It’s lidar -> ego (lidar timestamp) -> global -> ego (camera timestamp) -> camera.
The calibration is not changing per time-frame, but rather per log (~30 minutes of driving). Every other day the calibration gets checked and updated if needed. There are also multiple cars used, which naturally have slightly different calibration.

This post was flagged by the community and is temporarily hidden.

The overlap between the cameras is very limited, so this rather simple code should work well enough: https://github.com/nutonomy/nuscenes-devkit/blob/master/python-sdk/nuscenes/nuscenes.py#L925