An Indoor Visual Positioning Method with 3D Coordinates Using Built-In Smartphone Sensors Based on Epipolar Geometry.

Area of Science:

• Computer Vision
• Robotics
• Geomatics

Background:

• Traditional visual positioning methods struggle with mixed errors, leading to inaccurate direction vector convergence.
• Existing techniques often map 3D vectors to 2D planes, resulting in positioning errors, including points at infinity.

Purpose of the Study:

• To propose an improved indoor visual positioning method for accurate 3D coordinate determination.
• To overcome limitations of existing methods by integrating smartphone sensor data with visual computing.

Main Methods:

• The proposed method transforms the positioning problem into calculating the distance from a point to multiple lines in space.
• It fuses accelerometer and magnetometer data with visual information for enhanced coordinate accuracy.
• Epipolar geometry principles are utilized for geometric relation construction.

Main Results:

• The method demonstrates stable localization performance across various poses, even with limited image retrieval results.
• Experimental results show 90% of positioning errors are below 0.58 meters.
• The average positioning error is less than 0.3 meters, meeting practical application requirements.

Conclusions:

• The developed indoor visual positioning system offers a low-cost, accurate, and robust solution for user localization.
• Integration of inertial sensors significantly improves the reliability and accuracy of visual positioning.
• The method provides a practical alternative for applications requiring precise indoor positioning.