ral-lee-ls/imm_ukf_pda

IMM-UKF-PDA Tracker

Autoware package based on IMM-UKF-PDA tracker.

• From a sourced terminal:

roslaunch lidar_tracker imm_ukf_pda_tracker.launch

• From Runtime Manager:

Computing Tab -> Detection/ lidar_detector -> imm_ukf_pda_tracker

Reference

A. Arya Senna Abdul Rachman, 3D-LIDAR Multi Object Tracking for Autonomous Driving. 2017. paper

M. Schreire, Bayesian environment representation, prediction, and criticality assessment for driver assistance systems. 2017. paper

Requirements

• eucledian_cluster node.
• ray_ground_filter node.
• /tf topic. Below video is from Suginami data which contais /tf topic: (autoware-20180205150908.bag). You can download it from ROSBAG STORE for free. Otherwise, you need to do localization with a map to produce /tf topic from velodyne to world.
• wayarea info from vectormap if is possible.

Parameters

Launch file available parameters for imm_ukf_pda_tracker

Parameter	Type	Description
tracker_input_topic	String	Input topic(type: autoware_msgs::CloudClusterArray). Default /cloud cluster.
tracker_output_topic	String	Output topic(type: autoware_msgs::CloudClusterArray). Default /tracking_cluster_array.
life_time_threshold	Int	The minimum frames for targets to be visualized. Default 8.
gating_threshold	Double	The value of gate threshold for measurement validation. Default 9.22.
gate_probability	Double	The probability that the gate contains the true measurement. Default 0.99.
detection_probability	Double	The probability that a target is detected. Default 0.9.
merge_distance_threshold	Double	The distance threshold for associating bounding box over frames. Default 0.5.
static_velocity_threshold	Double	The velocity threshold for classifying static/dynamic. Default 0.5.
static_num_history_threshold	Int	The amount of frames the velocity is averaged over to compare to static_velocity_threshold. Default 3.
prevent_explosion_threshold	Double	The threshold for stopping kalman filter update. Default 1000.
use_sukf	bool	Use standard kalman filter. Default false.

Launch file available parameters for visualize_detected_objects

Parameter	Type	Description
input_topic	String	Input topic(type: autoware_msgs::CloudClusterArray). Default /tracking_cluster_array.
pointcloud frame	String	Pointcloud frame. Default velodyne.

Subscribed topics

Node: imm_ukf_pda_tracker

Topic	Type	Objective
/detection/lidar_objects	autoware_msgs::DetectedObjectArray	Segmented pointcloud from a clustering algorithm like eucledian cluster.
/tf	tf	Tracking objects in world coordinate.

Node: visualize_detected_objects

Topic	Type	Objective
/detected_objects	autoware_msgs::DetectedObjectArray	Objects with tracking info.

Published topics

Node: imm_ukf_pda_tracker

Topic	Type	Objective
/detected_objects	autoware_msgs::DetectedObjectArray	Added info like velocity, yaw ,yaw_rate and static/dynamic class to DetectedObject msg.
/bounding_boxes_tracked	jsk_recognition_msgs::BoundingBoxArray	Visualze bounsing box nicely in rviz by JSK bounding box. Label contains information about static/dynamic class

Node: visualize_detected_objects

Topic	Type	Objective
/detected_objects/velocity_arrow	visualization_msgs::Marker	Visualize velocity and yaw of the targets.
/detected_objects/target_id	visualization_msgs::Marker	Visualize targets' id.

Video

Benchmark

Please notice that benchmark scripts are in another repository. You can tune parameters by using benchmark based on KITTI dataset. The repository is here.

imm_ukf_pda