PCL based ROS package to Detect/Cluster --> Track --> Classify static and dynamic objects in real-time from LIDAR scans implemented in C++.
- K-D tree based point cloud processing for object feature detection from point clouds
- Unsupervised euclidean cluster extraction (3D) or k-means clustering based on detected features and refinement using RANSAC (2D)
- Stable tracking (object ID & data association) with an ensemble of Kalman Filters
- Robust compared to k-means clustering with mean-flow tracking
Follow the steps below to use this (multi_object_tracking_lidar
) package:
- Create a catkin workspace (if you do not have one setup already).
- Navigate to the
src
folder in your catkin workspace:cd ~/catkin_ws/src
- Clone this repository:
git clone https://github.com/praveen-palanisamy/multiple-object-tracking-lidar.git
- Compile and build the package:
cd ~/catkin_ws && catkin_make
- Add the catkin workspace to your ROS environment:
source ~/catkin_ws/devel/setup.bash
- Run the
kf_tracker
ROS node in this package:rosrun multi_object_tracking_lidar kf_tracker
If all went well, the ROS node should be up and running! As long as you have the point clouds published on to the filtered_cloud
rostopic, you should see outputs from this node published onto the obj_id
, cluster_0
, cluster_1
, …, cluster_5
topics along with the markers on viz
topic which you can visualize using RViz.
The input point-clouds can be from:
- A real LiDAR or
- A simulated LiDAR or
- A point cloud dataset or
- Any other data source that produces point clouds
Note: This package expects valid point cloud data as input. The point clouds you publish to the "filtered_cloud
" is not expected to contain NaNs. The point cloud filtering is somewhat task and application dependent and therefore it is not done by this module.
PCL library provides pcl::removeNaNFromPointCloud (...)
method to filter out NaN points. You can refer to this example code snippet to easily filter out NaN points in your point cloud.
If you use the code or snippets from this repository in your work, please cite:
@software{praveen_palanisamy_2019_3559187,
author = {Praveen Palanisamy},
title = {{praveen-palanisamy/multiple-object-tracking-lidar:
Multiple-Object-Tracking-from-Point-Clouds_v1.0.2}},
month = dec,
year = 2019,
publisher = {Zenodo},
version = {1.0.2},
doi = {10.5281/zenodo.3559187},
url = {https://doi.org/10.5281/zenodo.3559186}
}