ROS 2 Human Tracking Node (RealSense + YOLO + ByteTrack)

Overview

This ROS 2 node performs real-time human detection and tracking using an RGB-D camera (Intel RealSense) and publishes 3D positions of detected people.

It combines:

• YOLO (Ultralytics) for person detection
• BoT-SORT / ByteTrack for multi-object tracking
• Intel RealSense depth data for 3D localization

The output is published as a PoseArray message containing the 3D positions of tracked humans.

---

Features

• Real-time human detection (class: person only)
• Persistent tracking IDs across frames
• 3D position estimation using depth data
• ROS 2 topic publishing (/tracked_poses)
• Live visualization with bounding boxes, IDs, and coordinates
• FPS display for performance monitoring

---

Requirements

Hardware

• Intel RealSense camera (e.g. D455)
• GPU recommended (CUDA-compatible)

Software

• ROS 2 (tested on Humble or newer)
• Python 3.8+

Python Dependencies

pip install ultralytics opencv-python numpy pyrealsense2

Node Description

Node Name

tracker_publisher

Published Topic

tracked_poses (geometry_msgs/PoseArray) Contains the 3D positions of all tracked humans.

• position.x: horizontal position (meters)
• position.y: vertical position (meters)
• position.z: depth (meters)

Orientation is not used.