MSc research · 2021
Blind Assistance Indoors
ARKit Indoor Navigation for the Visually Impaired
An AR-powered indoor navigation system that lets communities map retail and public spaces and gives visually impaired users independent, voice- and haptic-guided navigation to products and key locations — without GPS or external hardware.
MSc Dissertation
ARKit Indoor Navigation for the Visually Impaired
MSc Computer Science (Distinction) · Middlesex University
The problem
GPS does not work indoors. Visually impaired people often depend on staff or memory to navigate supermarkets, malls, and other indoor spaces — limiting independence and confidence when moving through unfamiliar environments.
The solution
A native iOS app with a dual-role system: admins map indoor spaces in AR using a centre-screen crosshair cursor, and users navigate those spaces with graph-based pathfinding, spoken directions, haptic feedback, and AR relocalization — built accessibility-first from the ground up, not retrofitted.
How it works
Primary focus: retail and supermarket navigation — finding products, aisles, checkout, restrooms, exits, and help desks.
- 01
Admin maps store in AR
Place waypoints via crosshair raycasting onto detected surfaces
- 02
Graph saved & published
Waypoints, connections, and product catalog stored with ARWorldMap data
- 03
User selects destination
Browse locations, search products, or pick store landmarks
- 04
Dijkstra shortest path
Optimal route computed from nearest entry point to target
- 05
Multi-modal guidance
AR navigation, voice prompts, and haptic patterns in real time
System architecture
Admin mode
Create maps, place waypoints in AR, assign products, publish maps
User mode
Browse locations, search products, navigate with AR guidance
Data model
Graph of waypoints/nodes, bidirectional connections, and product catalog
Navigation engine
Dijkstra shortest path plus nearest-node path following tolerant to AR drift
AR layer
ARWorldMap persistence, relocalization, and crosshair waypoint placement
Accessibility
VoiceOver, high contrast, large touch targets, and "I'm lost" help flow
Technical contributions
Six integrated systems — from AR mapping through graph pathfinding to multi-modal guidance — forming a research-grade prototype, not a concept mockup.
AR Cursor System for indoor mapping
Instead of beacons or pre-built floor plans, admins map spaces by placing waypoints through a centre-screen crosshair with raycasting onto detected surfaces. Includes floor validation, minimum spacing (~0.35 m), optional raycast smoothing, and product anchoring along aisle edges — enabling community-driven mapping without expensive infrastructure.
Graph-based navigation without GPS
Indoor spaces modelled as navigation graphs: nodes for entrances, aisle ends, products, and hazards; bidirectional edges with real-world distances; Dijkstra's algorithm for shortest paths. Product search automatically computes the optimal route from the user's position to the shelf.
ARWorldMap persistence & relocalization
Maps store ARWorldMap data so sessions relocalize when users return — demonstrating that indoor AR navigation can persist across visits, not only during a single walk-through.
Nearest-node path following
Rather than rigid segment-following that breaks when tracking drifts, a nearest-node approach tracks visited vs available nodes, guides toward the nearest unvisited node on the planned path, detects off-route conditions (~1.8 m with hysteresis), and supports recovery — a practical response to imperfect indoor AR tracking.
Multi-modal guidance engine
Combines voice ("Turn left in 3 metres", "You have arrived"), haptic patterns (light/medium impact and notifications), camera-space 3D guidance for left/right/forward detection, and phase-based navigation (orienting → advancing → final approach) with turn thresholds tuned for indoor use.
Accessibility-first UX
Built for blind and low-vision users: full VoiceOver support, high-contrast mode, 56–88 pt touch targets, prominent "I'm Lost or Need Help" flow, per-waypoint audio descriptions, and configurable haptic patterns.
Community impact
Visually impaired individuals
- Independent shopping — search a product and get guided to the shelf
- Reduced reliance on staff for checkout, restrooms, exits, and help desks
- Safety net with clear help and emergency options when disoriented
- Multi-sensory guidance — audio and touch, not vision-only
Retailers & public spaces
- Low-cost deployment — iPhone plus AR mapping, no beacons or custom hardware
- Scalable model: admin maps once, many users navigate
- Templates for supermarkets, malls, hospitals, offices, and warehouses
- Supports accessibility compliance and inclusive design expectations
Researchers & developers
- Reference implementation for ARKit, accessibility, and indoor navigation
- Demonstrates graph pathfinding, AR relocalization, and multi-modal feedback together
- Extensible toward cloud sync, voice commands, and offline caching
Why it earned distinction
- Full working prototype — admin and user flows, AR editors, pathfinding, and guidance
- Novel integration of AR cursor mapping, graph navigation, and accessibility
- Domain-specific depth — supermarket aisles, products, categories, store locations
- Robust navigation design — nearest-node following, off-route recovery, relocalization
- Research-aware engineering — structural constraints, product anchoring, map validation
- Clear social impact — directly addresses indoor independence for blind and low-vision people
Cloud sync, real-time collaboration, and some help-flow logic remain on the roadmap — the prototype demonstrates core research contributions with a clear path to production deployment.
Tech stack