A ROS-Based Depth-Aware Motion Geometry Navigation Framework for Robust Ghost Artifact Detection in Hospital AMR Applications

Abstract

Autonomous Mobile Robots (AMRs) can effectively alleviate staffing shortages in healthcare facilities. However, stainless-steel fixtures used for infection control exhibit high surface reflectivity, often causing sensing errors and dynamic mirror artifacts that compromise navigation reliability. To address this challenge, this study proposes a ROS-Based Depth-Aware Motion Geometry Navigation Framework, a dual-layer ghost artifact recognition system designed to mitigate reflection-induced artifacts in hospital environments. The framework first performs depth-aware motion–based artifact suppression, followed by EKF processing to enable reliable dynamic obstacle avoidance for AMRs. It comprises three core components: (1) a normal-estimation-free motion geometry inference that detects reflection-induced artifacts by analyzing residuals between motion vectors without requiring surface normals; (2) a multi-type ghost candidate classification strategy incorporating IR intensity–based credibility weighting to ensure geometric and physical consistency; and (3) a directional and region-structure–aware IR masking method leveraging gradient orientation, elongated region analysis, and synchronized motion inference to identify reflective zones. Experimental validation, from simulation to real hospital experiments, shows that the proposed method reduces measurement errors by 41%, improves navigation success by 14.3% in controlled tests, and maintains over 90% success during extended hospital deployment.