This project visualizes the real-time object tracking with point clouds only. It uses gnn-pmb filter to improve the stability and accuracy of the tracker.
Before you compile this project, you need to download the Nuscenes dataset. This projects used the v1.0-mini version of the Nuscenes dataset.
Besides, you also need to acquire a set of detection from relevent dataset. In this project, PointPIlliars was being used.
After downloading the dataset and the detection file, put them into the proper directory.
In this project, eg. the v1.0-mini Nuscenes dataset is put into /home/ez/project/dataset/nuscenes. In the directory /nuscenes, there are 4 folders named sweeps, maps, samples and v1.0-mini and 1 file named .v1.0-mini.txt.
In this project, eg. the PointPilliars detection file is put into /home/ez/project/dataset/pointpillars/. So there are 3 files in the /pointpillars: pointpillars_val.json, pointpillars_test.json and pointpillars_train.json.
conda env create -f environment.yml
conda activate OpenPCDet
cd /home/ez/project/tracking/
source devel/setup.bash
roslaunch tracking_module launch.launch
If you want to change the directory of the dataset and detection file to your path, you can refer the lines below.
/tracking/src/tracking_module/scripts/PubInfo.py: line 142, line 184, line 302.
/tracking/src/tracking_module/scripts/PubInfo.py: line 309.
This project was accomplished on June 1, 2024 and was first upload to the github on March 10, 2025.
Contact Email: 2110539202@qq.com