This ROS package takes as input multiple SkeletonGroup messages, where each topic represents a separate detector, and assigns to each skeleton a proper ID by performing frame-by-frame tracking.
| Parameter | Description |
|---|---|
input_topics |
Topics published by all the detectors in the network |
output_topic |
Name of the topic that will be published containing the tracked skeletons |
fixed_delay |
Fixed delay to apply before tracking a detection (avoid possible source time inconsistencies when using multiple detectors) |
min_skeleton_confidence |
Minimum acceptable skeletons confidence |
min_marker_confidence |
Minimum acceptable markers confidence |
min_link_confidence |
Minimum acceptable links confidence |
min_markers |
Minimum number of markers to be detected to begin tracking |
min_links |
Minimum number of links to be detected to begin tracking |
min_linear_distance |
If the linear distance between an unassociated detection and a track is lower than this parameter, then the detection is considered as part of the track |
max_linear_distance |
Maximum acceptable linear distance between track and detection |
min_angular_distance |
If the angular distance between an unassociated detection and a track is lower than this parameter, then the detection is considered as part of the track |
max_angular_distance |
Maximum acceptable angular distance between track and detection |
max_delta_t |
Maximum acceptable time delta between track and detection |
use_positions |
Calculate distances based on the markers/links positions |
use_linear_velocities |
Calculate distances based on the markers/links linear velocities |
use_orientations |
Calculate distances based on the markers/links orientations |
use_angular_velocities |
Calculate distances based on the markers/links angular velocities |
velocity_weight |
Constant weight to apply to the distance calculated on the velocities |
weight_distances_by_confidences |
Weight the distance of each pair of markers/links w.r.t. the detection's confidence |
weight_distances_by_velocities |
Weight the distance of each pair of markers/links w.r.t. the inverse of the track's velocity |
ros2 launch hiros_skeleton_tracker default.launch.py
Please cite the following paper:
Guidolin, M., Tagliapietra, L., Menegatti, E., & Reggiani, M. (2023). Hi-ROS: Open-source multi-camera sensor fusion for real-time people tracking. Computer Vision and Image Understanding, 232, 103694.
Bib citation source:
@article{GUIDOLIN2023103694,
title = {Hi-ROS: Open-source multi-camera sensor fusion for real-time people tracking},
journal = {Computer Vision and Image Understanding},
volume = {232},
pages = {103694},
year = {2023},
issn = {1077-3142},
doi = {https://doi.org/10.1016/j.cviu.2023.103694},
url = {https://www.sciencedirect.com/science/article/pii/S1077314223000747},
author = {Mattia Guidolin and Luca Tagliapietra and Emanuele Menegatti and Monica Reggiani},
keywords = {Markerless motion capture, Multi-view body tracking, Real-time, ROS}
}