Merged Point Cloud

The Merged Point Cloud feature allows users to see all the points over time (across all the frames of a sequence) aggregated in the same coordinate system of the current frame resulting in one big point cloud. Therefore, ego motion data is mandatory for this type of scene.

It is used on sequential tasks to annotate stationary objects in a single .

By enabling the MPC, it's easy to detect and label all the stationary objects since they appear denser and brighter. While moving objects will leave a trail of points.

---

Benefits

• Improve Efficiency: By allowing all data points to be merged in a single frame, the user no longer has to traverse the sequence to find the ideal frame where the most points of the object are captured.

  • Static objects are also much more clearly visible with the accumulated points

• Improve Accuracy: By integrating data from multiple frames, merged point clouds can enhance the accuracy of object annotation. Annotators can use the combined data to make more informed decisions about the precise location of objects within the point cloud.

---

Steps to use MPC

1

Toggle MPC

Toggle the MPC switch on the header of the annotation tool. This will aggregate and merge points across all the frames of the sequence, by default on the next 20 frames.

The Multi Sensor Fusion tool supports various mechanisms of aggregating the point clouds across the sequence:

• all

• partial merged point cloud

• using frame jump length

2

Identify stationary objects

Identify the stationary object (that is dense and bright).

3

Create annotation

Select the drawing tool and create the annotation on the object.

• If the annotation is a cuboid, paintbrush, 3D Polyline it is propagated across all the frames.

Partial Merged Point Cloud

MPC has further options for merging points across a range of frames.

• The tool provides multiple options:

  • select a predefined option from the list

  • use the customizable start/end frame and hit enter

• After triggering the merged point cloud, the timeline introduces a yellow indicator for the frames that are merged.

---

Cuboid vs Paintbrush/3D Polygon Behavior in Partial MPC

When using a partial Merged Point Cloud, cuboid annotations and paintbrush/3D polygon annotations behave differently in terms of scope.

Cuboid

Drawing a cuboid while in a partial merged point cloud view creates the annotation across the entire sequence, not just the frames included in that partial merge.

• This means annotators can draw a cuboid using only a subset of merged frames, and the tool will still place the annotation on every frame of the sequence.

Paintbrush / 3D Polygon

Segmenting points with the paintbrush or 3D Polygon tool while in a partial merged point cloud view only applies the segmentation to the frames that were included in that specific partial merge.

• Frames outside the merged range are unaffected, so annotators working on a partial merge should confirm the merge range covers every frame they intend to segment.

If the goal is to label a stationary object across the whole sequence, cuboid is the more efficient choice, since a single partial merge is enough to cover all frames. If using paintbrush or 3D polygon, make sure the merge range includes every frame that needs segmenting before starting, since the annotation won't extend beyond it.