Robust and Efficient Indoor Localization Using Sparse Semantic Information from a Spherical Camera.

Area of Science:

• Robotics and Artificial Intelligence
• Computer Vision and Machine Learning

Background:

• Accurate self-localization is crucial for autonomous systems to navigate and interact with environments.
• Existing methods often rely on dense data structures (e.g., point clouds), posing computational challenges for real-time applications.

Purpose of the Study:

• To propose a novel sparse semantic self-localization approach for robust and efficient indoor localization.
• To reduce uncertainty in real-time localization by utilizing sparse semantic information.

Main Methods:

• Developed a sensor model using sparse semantic information (e.g., doors, windows) on a 2D annotated map.
• Employed low-cost, lightweight sensors: an inertial measurement unit and a spherical camera.
• Map preparation involves marking approximate object centers on a 2D map, avoiding complex 3D constructions.

Main Results:

• Demonstrated efficient real-time localization by reducing uncertainty with minimal sparse semantic data.
• The approach showed robustness against partial view obstructions and geometrical map errors.
• Experimental validation confirmed the feasibility and robustness of the proposed method.

Conclusions:

• The sparse semantic self-localization approach offers an efficient and robust solution for indoor navigation.
• This method simplifies map creation and sensor requirements compared to traditional techniques.
• It provides a practical alternative for real-time localization using readily available sensors and maps.