fix(shape_estimation): unintented roi shape estimation

[emuemuJP]

Description

ROI is different than expected and comes out horizontally.

In the function “correctWithDefaultValue” of the corrector, which corrects the point cluster to the size according to the object type, a judgment is made as to which side of the ROI should be corrected. The decision is made based on the relative positional relationship between the vehicle and the point cluster. In this case, the ROI was corrected in an unintended direction because the second farthest edge was corrected instead of the farthest edge, especially when the point cluster was located right next to the vehicle.

Related links

Parent Issue:

• Link

How was this PR tested?

Test with specified rosbag.

shape_estimation_before.webm

shape_estimation_after.webm

Notes for reviewers

None.

Interface changes

None.

Effects on system behavior

None.

[github-actions[bot]]

Thank you for contributing to the Autoware project!

🚧 If your pull request is in progress, switch it to draft mode.

Please ensure:

• You've checked our contribution guidelines.
• Your PR follows our pull request guidelines.
• All required CI checks pass before marking the PR ready for review.

[emuemuJP]

I have considered these conditional branches. For the case of extending the ROI horizontally, it may be more appropriate to return false because the condition is indefinite.

[Owen-Liuyuxuan]

I have considered these conditional branches. For the case of extending the ROI horizontally, it may be more appropriate to return false because the condition is indefinite.

@emuemuJP Sorry, the edges'number does not seem to be labeled correctly, which could mislead my judgment.

But does that mean we should return False in case 3?

[emuemuJP]

The correspondence between roi edge index and the furthest edge index may be confusing.

The white dots in the video are /perception/obstacle_segmentation/pointcloud topic from occupancy_grid_based_outlier_filter node. The purple dots in the video are /perception/object_recognition/detection/pointcloud_map_filtered/pointcloud topic from voxel_based_compare_map_filter node. The yellow dots in the video are The white dots in the video are /perception/obstacle_segmentation/pointcloud topic from /perception/object_recognition/detection/clustering/roi_cluster/roi_pointcloud_fusion node.

We have concluded that it is not essential to change the shape estimation node.

I will investigate this issue in the future, as it may be that fewer yellow and purple points, which are the result of filtering out the white points, will be assigned to the wrong ROI.