Multi-Sensor-Assisted Low-Cost Indoor Non-Visual Semantic Map Construction and Localization for Modern Vehicles

Area of Science:

• Automotive Engineering
• Robotics and Autonomous Systems
• Geospatial Information Science

Background:

• The automotive industry's evolution necessitates advanced positioning for connected and autonomous vehicles.
• Existing indoor localization methods often lack cost-effectiveness or seamless integration for diverse vehicle systems.
• Intelligent vehicle systems (infotainment, Internet of Vehicles, Autopilot) require robust indoor/outdoor positioning.

Purpose of the Study:

• To propose a low-cost, versatile indoor non-visual semantic mapping and localization solution for modern vehicles.
• To develop a method for identifying non-visual semantic landmarks without visual input.
• To create an accurate localization system leveraging semantic map features and sensor data.

Main Methods:

• Sliding window-based semantic landmark detection for features like entrances and road nodes.
• Construction of an indoor semantic map integrating vehicle trajectory, landmarks, and Wi-Fi RSS fingerprints.
• Graph-optimized localization using landmark matching and exploiting landmark correlations.

Main Results:

• High accuracy of 98.1% in non-visual semantic landmark detection.
• Low localization error of 1.31 meters achieved in field experiments.
• Successful validation in diverse underground parking environments.

Conclusions:

• The proposed method offers a cost-effective and versatile solution for indoor vehicle localization.
• Non-visual semantic mapping and landmark-based localization enhance navigation accuracy in complex indoor environments.
• This approach supports the integration of advanced intelligent systems in modern vehicles.