Vertex: A Semantic Graph-Based Indoor Navigation System with Vision-Language Landmark Verification.

Area of Science:

• Robotics
• Human-Computer Interaction
• Computer Vision

Background:

• Indoor navigation for older adults is difficult due to GPS limitations and complex environments.
• Existing systems often require specialized hardware like 3D scanners or beacons.
• Visually repetitive corridors and location failures pose significant challenges.

Purpose of the Study:

• To develop a multimodal indoor navigation assistant for older adults.
• To provide reliable step-by-step guidance in unfamiliar indoor environments.
• To enhance user confidence, safety, and independence during navigation.

Main Methods:

• A graph-based model of the environment using semantic landmarks from building videos.
• Dijkstra's algorithm for route calculation on the semantic graph.
• Restricted vision-language recognition for landmark verification and a voting mechanism for robust node transitions.
• Hierarchical redirection strategies for local and global navigation recovery.
• Implementation in handheld (AR cues, voice) and hands-free (voice, keywords) modes.

Main Results:

• Successful preliminary technical testing in the UK and user validation in Spain.
• Participants reported high usability, confidence, and safety.
• The system demonstrated increased perceived independence for older adults.
• The approach reduces reliance on specialized indoor positioning instruments.

Conclusions:

• The multimodal indoor navigation assistant effectively addresses challenges faced by older adults in unfamiliar buildings.
• The system offers a robust, user-friendly, and independent navigation solution.
• Graph-based planning combined with visual landmark verification provides a promising approach for indoor navigation.