Integrating Moving Platforms in a SLAM Agorithm for Pedestrian Navigation.

Area of Science:

• Robotics
• Computer Vision
• Geomatics Engineering

Background:

• Inertial Measurement Units (IMUs) struggle with position estimation in 3D indoor navigation due to unhandled moving platforms (MPs).
• Existing 3D-SLAM and pedestrian dead-reckoning (PDR) techniques require improved integration to address drift and MP challenges.

Purpose of the Study:

• To develop a robust 3D indoor navigation system that accurately estimates pedestrian position, even with the presence of moving platforms.
• To enhance simultaneous localization and mapping (SLAM) algorithms by incorporating MPs like escalators and elevators.

Main Methods:

• Integrated MPs into an upper 3D-SLAM algorithm cascaded with the pedestrian dead-reckoning (PDR) technique.
• Developed a novel proposal function for particle filter-based SLAM to account for MPs.
• Introduced a new weighting function for learned features (escalators, elevators) to improve localization accuracy upon revisits.

Main Results:

• Achieved a mean height error of approximately 0.1 m and a mean position error of less than 1 m for long-distance walks across floors.
• Demonstrated good performance with multiple floor level changes in a multistory environment.
• The final floor number was correctly estimated in all but one case, with slightly higher height errors (0.5 m) for short walks with frequent floor changes.

Conclusions:

• The proposed method effectively integrates moving platforms into 3D indoor navigation systems, significantly reducing position and height errors.
• The system shows robustness in complex multistory environments with varying walking distances and floor transitions.
• This approach offers a promising solution for accurate and reliable 3D pedestrian indoor localization.