LiDAR-Based Human Detection and Multi-Target Tracking

ROS 2, 2D LiDAR, Sensor Processing — Perception and Tracking Pipeline

📋 Project Overview

Designed a geometry-driven 2D LiDAR perception pipeline using clustering, spatial constraints, and PCA-based linearity analysis to distinguish humans from walls and static structures.

Implemented multi-object data association to maintain persistent track IDs across frames, prioritizing robustness to occlusions and sensor dropouts. Structured the system into decoupled ROS 2 perception and tracking nodes, improving scalability, debuggability, and real-time performance. Visualized human trajectories and tracking states in RViz, validating detection quality and ID stability using rosbag-based offline evaluation.

⚡ Key Highlights

Geometry-Driven Pipeline: Clustering, spatial constraints, PCA-based linearity analysis
Human vs. Structure: Distinguishing humans from walls and static obstacles
Multi-Object Tracking: Persistent track IDs across frames
Robustness: Handles occlusions and sensor dropouts
Modular Design: Decoupled ROS 2 perception and tracking nodes
Validation: RViz visualization, rosbag-based offline evaluation

Skills Demonstrated

ROS 2 LiDAR 2D Perception Multi-Target Tracking Data Association PCA Clustering RViz rosbag

More details and images coming soon.