This project develops a real-time system for detecting and tracking people using LIDAR data within dynamic environments. Built on the ROS2 framework, it leverages nodes, topics, and custom messages to process raw sensor inputs into meaningful insights about the environment. The system consists of two primary nodes: one for detecting moving objects and another for tracking and counting these individuals.
- Node 1: Moving Object Detection - Subscribes to
/scantopic for LIDAR data, identifies moving objects using a threshold-based comparison, and employs Euclidean clustering to locate individuals, publishing centroids to/person_location. - Node 2: Tracking and Counting - Subscribes to
/person_locationfor detected individuals' centroids, tracks these individuals across frames, and counts unique occurrences, publishing the cumulative count to/person_count.
A launch file coordinates the initialization and termination of system components for synchronized operation.
- ROS2 Humble Hawksbill
- PCL (Point Cloud Library) for Euclidean clustering
- OpenCV (optional) for additional visualization
Real-Time Processing: Essential for dynamic environments to provide timely information. Threshold-Based Movement Detection: Efficient at differentiating static and moving objects with minimal computational overhead. Euclidean Clustering for Object Segmentation: Effectively separates individuals in crowded scenes. Simple Tracking Algorithm with Constant Velocity Model: Balances accuracy and computational efficiency for real-time applications. Cumulative Counting of Unique Individuals: Simplifies tracking over time without complex identity management.
Movement Detection Threshold, Euclidean Clustering Parameters (distance tolerance, min/max cluster size), and Tracking Update Interval were empirically determined for optimal performance. Results and Expectations The system demonstrated high accuracy in detecting and tracking people in various test scenarios. Fine-tuning parameters allowed for significant improvements, especially in challenging conditions like crowded environments or at the edges of the LIDAR range.
ROS2 Community for the comprehensive documentation and forums. Contributors to the PCL and OpenCV libraries.