This package provides a robust pipeline for detecting and tracking both static and dynamic objects using a 2D LIDAR sensor. It transforms raw laser scans into tracked objects with estimated poses, dimensions, and velocities.
Lidar Object Detection demo showing clustering, L-shape fitting, and real-time tracking.
- L-Shape Fitting: Implements Search-Based Rectangle Fitting to accurately represent vehicle-like or box-like geometries.
- Multi-Object Tracking: Uses the Hungarian Algorithm for data association and Kalman Filters to estimate object velocities and maintain IDs across frames.
- Robust Filtering: Includes range clipping and size-based filtering to eliminate sensor noise and robot self-reflections.
- RViz Visualization: Integrated support for publishing 3D bounding boxes, ID labels, and velocity vectors.
- Preprocessing: LiDAR points are clipped by range and transformed into a global or local reference frame.
- Clustering: Data is grouped using DBSCAN (Density-Based Spatial Clustering of Applications with Noise).
- L-Shape Fitting: Each cluster is fitted with an optimal rectangle side-matching algorithm [1].
- Tracking: - New detections are associated with existing tracks via the Hungarian Algorithm.
- A Kalman Filter (Constant Velocity Model) predicts and updates the state (pose and twist) of each object.
You can launch the node with custom parameters using a ROS 2 launch file. Below is an example configuration that includes all the tracking and visualization settings:
from launch import LaunchDescription
from launch_ros.actions import Node
def generate_launch_description():
lidar_object_detection = Node(
package="lidar_object_detection_ros2",
executable="lidar_object_detection.py",
parameters=[
# DBSCAN Parameters
{"dbscan_eps": 0.2},
{"dbscan_min_samples": 5},
# Lidar and Frame Parameters
{"lidar_angular_resolution": 0.5},
{"frame_id": "map"},
{"lidar_frame_id": "lidar"},
{"flip_x_axis": True},
{"flip_y_axis": True},
{"update_rate": 0.2},
# L-shape Detection Parameters
{"min_l": 0.05},
{"max_l": 1.0},
{"angle_step": 1.0},
# Lidar Clipping Parameters
{"min_range": 0.1},
{"max_range": 8.0},
# Kalman Filter Parameters
{"use_kalman_filter": True},
{"kf_q_std": 0.05},
{"kf_r_std": 0.1},
# Visualization Parameters
{"publish_markers": True},
{"color_bbox": [0.0, 1.0, 0.0, 0.5]}, # Green
{"color_text": [1.0, 1.0, 1.0, 1.0]}, # White
{"color_arrow": [1.0, 1.0, 0.0, 1.0]}, # Yellow
{"min_velocity_show": 0.01},
# Tracking/Association Parameters
{"max_disappeared": 6},
{"max_association_distance": 0.8}
],
output="screen",
)
return LaunchDescription([
lidar_object_detection,
])- dbscan_eps: (default:
0.1) Max distance between points to be considered a cluster. - dbscan_min_samples: (default:
5) Minimum points to form a cluster. - min_range / max_range: (default:
0.1/8.0) Radial distance window to consider LIDAR data (meters). - min_l / max_l: (default:
0.05/1.0) Constraints on the allowable size of the fitted rectangle sides. - angle_step: (default:
1.0) The step (in degrees) at which the L-Shape fitting algorithm searches for the correct orientation.
- use_kalman_filter: (default:
True) Enable velocity estimation and pose smoothing. - kf_q_std: (default:
0.05) Process noise (uncertainty in the motion model). - kf_r_std: (default:
0.1) Measurement noise (uncertainty in the LIDAR detections). - max_disappeared: (default:
6) Number of frames to keep a lost object before deleting its ID. - max_association_distance: (default:
0.8) Max distance (meters) to associate a detection with a track.
- frame_id: (default:
"") Target frame (e.g.,map). If empty, uses the LIDAR frame. - lidar_frame_id: (default:
"laser_link") The frame of the LIDAR sensor. - update_rate: (default:
0.1) Period (seconds) for the processing loop. - flip_x_axis / flip_y_axis: (default:
False) Flags to invert axes based on sensor mounting.
- publish_markers: (default:
True) Toggle publishing ofvisualization_msgs/MarkerArray. - color_bbox: (default:
[0.0, 1.0, 0.0, 0.5]) RGBA for bounding boxes. - color_text: (default:
[1.0, 1.0, 1.0, 1.0]) RGBA for ID labels. - color_arrow: (default:
[1.0, 1.0, 0.0, 1.0]) RGBA for velocity arrows. - min_velocity_show: (default:
0.01) Speed threshold to display velocity arrows.
scan(sensor_msgs/LaserScan)
lod_objects(lidar_object_detection_ros2/msg/ObjectsArray)lod_markers(visualization_msgs/MarkerArray)
The following custom message definitions are used to represent the detected and tracked objects.
Defines the geometric properties of a fitted L-shape.
geometry_msgs/Pose2D c1 # The primary corner point (x, y) and orientation (theta)
float32 l1 # Length of the first side
float32 l2 # Length of the second side
Contains the full state of a tracked object, including its unique ID and movement data.
int32 id # Unique persistent ID for tracking
LShape l_shape # Geometric L-shape properties
geometry_msgs/Pose2D pose # The calculated center pose of the object
geometry_msgs/Twist twist # Linear and angular velocity estimated by the Kalman Filter
The top-level message used to publish the current list of all tracked objects.
std_msgs/Header header # Timestamp and coordinate frame information
Object[] objects # Array of detected/tracked objects
[1] X. Zhang, W. Xu, C. Dong, and J. M. Dolan, “Efficient l-shape fitting for vehicle detection using laser scanners,” in 2017 IEEE Intelligent Vehicles Symposium (IV), pp. 54–59, IEEE, 2017. https://doi.org/10.1109/IVS.2017.7995698