object_merger

Purpose

object_merger is a package for merging detected objects from two methods by data association.

Inner-workings / Algorithms

The successive shortest path algorithm is used to solve the data association problem (the minimum-cost flow problem). The cost is calculated by the distance between two objects and gate functions are applied to reset cost, s.t. the maximum distance, the maximum area and the minimum area.

Inputs / Outputs

Input

Name	Type	Description
input/object0	autoware_auto_perception_msgs::msg::DetectedObjects	detection objects
input/object1	autoware_auto_perception_msgs::msg::DetectedObjects	detection objects

Output

Name	Type	Description
output/object	autoware_auto_perception_msgs::msg::DetectedObjects	modified Objects

Parameters

Name	Type	Description
can_assign_matrix	double	Assignment table for data association
max_dist_matrix	double	Maximum distance table for data association
max_area_matrix	double	Maximum area table for data association
min_area_matrix	double	Minimum area table for data association
max_rad_matrix	double	Maximum angle table for data association
base_link_frame_id	double	association frame
distance_threshold_list	std::vector<double>	Distance threshold for each class used in judging overlap. The class order depends on ObjectClassification.
generalized_iou_threshold	std::vector<double>	Generalized IoU threshold for each class

Tips

• False Positive Unknown object detected by clustering method sometimes raises the risk of sudden stop and interferes with Planning module. If ML based detector rarely misses objects, you can tune the parameter of object_merger and make Perception module ignore unknown objects.
  • If you want to remove unknown object close to large vehicle,
    • use HIGH distance_threshold_list
      • However, this causes high computational load
    • use LOW precision_threshold_to_judge_overlapped
    • use LOW generalized_iou_threshold
      • However, these 2 params raise the risk of overlooking object close to known object.

Assumptions / Known limits

(Optional) Error detection and handling

(Optional) Performance characterization

(Optional) References/External links

(Optional) Future extensions / Unimplemented parts

Data association algorithm was the same as that of multi_object_tracker, but the algorithm of multi_object_tracker was already updated.