Image-Based Localization Aided Indoor Pedestrian Trajectory Estimation Using Smartphones.

Area of Science:

• Computer Vision
• Robotics
• Geomatics Engineering

Background:

• Consumer smartphones are crucial for ubiquitous indoor localization services.
• Pedestrian dead reckoning (PDR) combined with map-matching offers low-cost tracking but struggles in open areas due to accumulated errors and map dependency.
• Existing methods fail in large scenes where traditional 2D-to-3D matching becomes unreliable.

Purpose of the Study:

• To develop a reliable indoor pedestrian localization method that overcomes map constraints and accumulated errors.
• To improve pedestrian trajectory estimation using image-based localization.
• To achieve accurate and robust localization without reliance on dedicated infrastructure.

Main Methods:

• An improved image-based localization method is proposed, recovering camera pose from 2D-3D correspondences.
• A structure-from-motion (SfM) pipeline reconstructs the scene model.
• A robust image registration strategy integrates 3D-to-2D search, utilizing visibility and co-visibility information for efficiency.

Main Results:

• The proposed method successfully determines initial locations and corrects accumulated PDR errors upon successful image registration.
• A novel 3D-to-2D search strategy enhances image registration in large, complex scenes.
• Experimental results show highly acceptable long-term pedestrian localization with an average error of 0.56 meters.

Conclusions:

• The developed image-based localization method provides accurate and robust indoor pedestrian tracking without map constraints.
• The approach effectively eliminates accumulated errors from PDR, offering a significant advancement in ubiquitous localization.
• This method demonstrates the potential for reliable, infrastructure-free pedestrian localization using consumer smartphones.