robokudo.utils.annotator_helper

Annotator utility functions for Robokudo.

This module provides helper functions for working with annotators and their data. It supports:

  • Coordinate frame transformations

  • Point cloud manipulation

  • Image processing

  • Bounding box visualization

  • Camera parameter handling

The module integrates with:

  • OpenCV for image operations

  • Open3D for point cloud operations

  • NumPy for numerical computations

  • CAS for data management

Functions

transform_pose_from_cam_to_world(...)

Transform pose from camera to world coordinates.

transform_pose_from_world_to_cam(...)

Transform pose from world to camera coordinates.

transform_cloud_from_cam_to_world(...)

Transform point cloud from camera to world coordinates.

transform_cloud_from_world_to_cam(...)

Transform point cloud from world to camera coordinates.

get_cam_to_world_transform_matrix(→ numpy.typing.NDArray)

Get camera-to-world transform matrix from CAS.

get_world_to_cam_transform_matrix(→ numpy.typing.NDArray)

Get world-to-camera transform matrix from CAS.

draw_bounding_boxes_from_object_hypotheses(→ None)

Draw bounding boxes for object hypotheses on the image.

scale_cam_intrinsics(→ None)

Scale camera intrinsics based on color-to-depth ratio.

get_color_image(→ numpy.typing.NDArray)

Load and potentially resize the color image from the CAS.

resize_mask(→ numpy.typing.NDArray)

Resize mask to match color image resolution.

generate_source_name(→ str)

Helper to create the standard name for the 'source' field of Annotations.

Module Contents

robokudo.utils.annotator_helper.transform_pose_from_cam_to_world(cas: robokudo.cas.CAS, pose: robokudo.types.annotation.PoseAnnotation) → robokudo.types.annotation.PoseAnnotation

Transform pose from camera to world coordinates.

Uses transform from CAS to convert pose from camera frame to world frame.

Parameters:

  • cas – CAS containing camera-to-world transform

  • pose – Pose in camera coordinates

Returns:

Transformed pose in world coordinates

Raises:

AssertionError – If camera-to-world transform not in CAS

robokudo.utils.annotator_helper.transform_pose_from_world_to_cam(cas: robokudo.cas.CAS, pose: robokudo.types.annotation.PoseAnnotation) → robokudo.types.annotation.PoseAnnotation

Transform pose from world to camera coordinates.

Uses transform from CAS to convert pose from world frame to camera frame.

Parameters:

  • cas – CAS containing camera-to-world transform

  • pose – Pose in world coordinates

Returns:

Transformed pose in camera coordinates

Raises:

AssertionError – If camera-to-world transform not in CAS

robokudo.utils.annotator_helper.transform_cloud_from_cam_to_world(cas: robokudo.cas.CAS, cloud: open3d.geometry.PointCloud, transform_inplace: bool = False) → open3d.geometry.PointCloud

Transform point cloud from camera to world coordinates.

Uses transform from CAS to convert point cloud from camera frame to world frame.

Parameters:

  • cas – CAS containing camera-to-world transform

  • cloud – Point cloud in camera coordinates

  • transform_inplace – Whether to modify input cloud directly

Returns:

Transformed point cloud in world coordinates

Raises:

KeyError – If camera-to-world transform not in CAS

robokudo.utils.annotator_helper.transform_cloud_from_world_to_cam(cas: robokudo.cas.CAS, cloud: open3d.geometry.PointCloud, transform_inplace: bool = False) → open3d.geometry.PointCloud

Transform point cloud from world to camera coordinates.

Uses transform from CAS to convert point cloud from world frame to camera frame.

Parameters:

  • cas – CAS containing camera-to-world transform

  • cloud – Point cloud in world coordinates

  • transform_inplace – This bool will control if the given ‘cloud’ is deepcopied before the transformation is done.

Set this value to True, if it is OK that the given cloud variable is changed. Set this to value to False (default), if a deepcopy should performed beforehand. :return: Transformed point cloud in camera coordinates :raises KeyError: If camera-to-world transform not in CAS

robokudo.utils.annotator_helper.get_cam_to_world_transform_matrix(cas: robokudo.cas.CAS) → numpy.typing.NDArray

Get camera-to-world transform matrix from CAS.

Parameters:

cas – CAS containing camera-to-world transform

Returns:

4x4 camera-to-world transform matrix

Raises:

KeyError – If camera-to-world transform not in CAS

robokudo.utils.annotator_helper.get_world_to_cam_transform_matrix(cas: robokudo.cas.CAS) → numpy.typing.NDArray

Get world-to-camera transform matrix from CAS.

Parameters:

cas – CAS containing camera-to-world transform

Returns:

4x4 world-to-camera transform matrix

Raises:

KeyError – If camera-to-world transform not in CAS

robokudo.utils.annotator_helper.draw_bounding_boxes_from_object_hypotheses(img: numpy.typing.NDArray, object_hypotheses: List[robokudo.types.scene.ObjectHypothesis], text_function: typing_extensions.Callable[[robokudo.types.scene.ObjectHypothesis], str]) → None

Draw bounding boxes for object hypotheses on the image.

Draws rectangles and labels for each object hypothesis ROI.

Parameters:

  • img – Input image to draw on

  • object_hypotheses – List of object hypotheses

  • text_function – Function to generate label text from hypothesis

robokudo.utils.annotator_helper.scale_cam_intrinsics(annotator: robokudo.annotators.core.BaseAnnotator) → None

Scale camera intrinsics based on color-to-depth ratio.

If the color2depth ratio is not 1,1, we will usually scale the color image to the same size the depth image has. This also requires an adjustment of the cam intrinsics.

If color2depth ratio is not 1,1, we’ll scale the cam intrinsics in annotator. You need to have a self.cam_intrinsics variable!

Parameters:

annotator – Annotator containing camera parameters

robokudo.utils.annotator_helper.get_color_image(annotator: robokudo.annotators.core.BaseAnnotator) → numpy.typing.NDArray

Load and potentially resize the color image from the CAS.

This is useful, as the resolution of the RGB image might differ from the resolution of the depth image. This function resizes the color image to the resolution of the depth image.

This is important later on when we want to create a point cloud from the depth image and the color image and crop the point cloud to the bounding box of the object hypothesis, which is defined in the color image coordinate system.

Parameters:

annotator – Annotator to get image from

Returns:

The resized color image.

robokudo.utils.annotator_helper.resize_mask(annotator: robokudo.annotators.core.BaseAnnotator, mask: numpy.typing.NDArray) → numpy.typing.NDArray

Resize mask to match color image resolution.

The mask is potentially created after the input image has been scaled down (see get_scaled_color_image_for_depth_image, we usually scale the RGB down). If that’s the case, we have to bring it back to the original resolution. This method checks some usual error cases and decides if we need to resize at all.

Parameters:

  • annotator – Annotator containing resolution info

  • mask – A binary image.

Returns:

The scaled version of mask, according to the COLOR2DEPTH_RATIO.

robokudo.utils.annotator_helper.generate_source_name(annotator: robokudo.annotators.core.BaseAnnotator) → str

Helper to create the standard name for the ‘source’ field of Annotations.

Parameters:

annotator – The annotator to generate the source name for

Returns:

The source name as string