A Mobile Outdoor Augmented Reality Method Combining Deep Learning Object Detection and Spatial Relationships for...

Area of Science:

• Computer Science
• Geographic Information Science
• Augmented Reality

Background:

• Markerless augmented reality (AR) systems struggle with accuracy and robustness in uncontrolled outdoor environments.
• Existing methods often rely on network connectivity, limiting usability in areas with poor signal.
• Precise registration and interaction with virtual objects in real-world outdoor settings remain a challenge.

Purpose of the Study:

• To develop a robust, fast, and markerless mobile AR method for registration, geovisualization, and interaction in uncontrolled outdoor environments.
• To create a lightweight, deep-learning-based object detection approach suitable for mobile or embedded devices.
• To enable network-independent AR experiences for enhanced reliability.

Main Methods:

• A lightweight deep-learning object detection model was developed for mobile devices.
• Sensor fusion combined vision-based detection with data from Global Positioning System (GPS), Inertial Measurement Unit (IMU), and magnetometer.
• A touch-gesture-based interaction method was implemented for virtual objects registered using geospatial information.

Main Results:

• The system achieved high detection accuracy and stable geovisualization results.
• Precise registration of virtual objects in the real world was demonstrated.
• The network-independent design ensured robustness in varied signal conditions.

Conclusions:

• The developed mobile AR method offers a robust and accurate solution for outdoor applications.
• The system effectively integrates deep learning with device sensors for seamless geovisualization and interaction.
• This approach enhances the feasibility of markerless AR in real-world, uncontrolled environments.