Global localization from monocular SLAM on a mobile phone.

Area of Science:

• Computer Vision
• Robotics
• Mobile Computing

Background:

• Simultaneous Localization and Mapping (SLAM) systems typically operate locally, leading to drift and limited global context.
• Mobile devices face challenges in accurate 6DoF pose estimation due to computational constraints and narrow camera fields of view.
• Existing global localization methods often require extensive pre-mapping or are computationally intensive for real-time mobile application.

Purpose of the Study:

• To develop a robust 6DoF tracking and mapping system for mobile devices.
• To integrate local SLAM with global localization for accurate, drift-free pose estimation.
• To overcome the limitations of mobile camera FOV and enable tracking beyond pre-mapped areas.

Main Methods:

• A keyframe-based monocular SLAM system runs on a mobile client for relative pose estimation and keyframe capture.
• A server process performs global localization of keyframes using a pre-existing, globally-registered map.
• Global registration corrections are sent to the client, and global anchor points refine client-side bundle adjustment, limiting drift.

Main Results:

• Achieved real-time, globally registered 6DoF tracking and mapping on a mobile device.
• Successfully addressed localization challenges posed by narrow field-of-view mobile phone cameras.
• Enabled continuous tracking and mapping in areas not covered by the initial offline reconstruction.

Conclusions:

• The proposed hybrid SLAM and global localization approach provides accurate and drift-free 6DoF tracking on mobile platforms.
• This system enhances the usability of mobile devices for augmented reality, robotics, and navigation applications.
• The method demonstrates a significant advancement in real-time mobile mapping and localization capabilities.