This study proposes an efficient localization system called In-Stargazer for human workers and mobile robots in large-scale industrial environments to enhance accident prevention and emergency response. Conventional methods suffer from technical limitations, high costs, and low practicality. To address these issues, the proposed system employs a lightweight helmet-mounted camera (120° FOV, 50 mm × 40 mm × 15 mm, 112 g), which can also be deployed on robot platforms. Unlike approaches using front-facing images that raise privacy concerns and are sensitive to environmental changes, the system relies solely on ceiling images, exploiting their structural stability in industrial settings. The environment is segmented into predefined sections, and a database is constructed for efficient image matching. To handle textureless ceilings, a deep learning-based feature extractor and matcher combined with a particle filter are used to improve robustness and accuracy. Experimental results show 81.9% pose estimation accuracy within one sub-section and 91.4% within two. This cost-effective and practical solution enables privacy-conscious localization and supports safe and efficient operation in industrial environments shared by humans and robots.
Choi et al. (Thu,) studied this question.