Person Re-Identification for Robot Person Following with Online Continual Learning
Prequities
- ROS, verified in melodic and noetic
- OpenCV with 3.4.12
- Ceres
- modified mmtrack
- Create a conda environment and install mmtrack
conda create -n mono_following python=3.8
conda activate mono_following
### Install mmtrack (core code for target-ReID) ###
This code will be released when the paper is accepted
...
- Install python related packages:
pip install -r requirements.txt
git clone https://github.com/eric-wieser/ros_numpy
cd ros_numpy
python setup.py install
- Install cpp related packages:
- OpenCV==3.4.12
- Eigen==3.0+
- Download bounding-box detection models: yolox-s and yolox-m, then make director
mono_tracking/scripts/AlphaPose/YOLOX/weightsand put the checkpoints to it. - Download 2d joint detection models: Google drive, then make directory
mono_tracking/scripts/AlphaPose/Modelsand put the checkpoints to it.
Run with our self-built dataset as ROSBAG:
# open go1 model
roslaunch go1_description go1_rviz.launch
roslaunch mono_tracking all_mono_tracking.launch sim:=true
# play bag
rosbag play --clock xxx.bagRun with the robot:
roslaunch mono_tracking all_mono_tracking.launch sim:=falseRun with icvs datasets as ROSBAG, and evaluate:
# If run in "corridor_corners" scene
roslaunch mono_tracking evaluate_MPF_in_icvs.launch scene:=corridor_corners@article{ye2023person,
title={Person Re-Identification for Robot Person Following with Online Continual Learning},
author={Ye, Hanjing and Zhao, Jieting and Zhan, Yu and Chen, Weinan and He, Li and Zhang, Hong},
journal={arXiv preprint arXiv:2309.11727},
year={2023}
}