27_A Proactive Smart-Glass Framework for Seamless Outdoor and Indoor Navigation for Visually Impaired with Context-Aware Obstacle Avoidance (2) (2) (1)

A Proactive Smart-Glass Framework for Seamless Navigation for Visually Impaired with Context-Aware Obstacle Avoidance Team Members Akisher J M K Kawvya Pravin G SSN College Of Engineering Under the Guidance of Dr. D. Venkata Vara Prasad Professor, Computer Science and Engineering.

PROBLEM STATEMENT Problem • Blind and visually impaired people lack an affordable device for safe navigation both indoors and outdoors. • Smartphone apps depend on GPS and internet, have limited indoor navigation, and do not provide complete obstacle guidance. • Robotic systems (e.g.CaBot) can navigate indoors, but they are expensive, bulky, and require special infrastructure. Why It Matters • Improves independent and safe mobility. • Combines indoor + outdoor navigation. • Combines destination guidance with real-time obstacle detection in one offline wearable. 2 v 1.

EXISTING SOLUTIONS & LIMITATIONS 35M+ Existing Solutions: • BLE Beacon Systems– NavCog, RightHear • Smartphone Navigation Apps– Lazarillo, BlindSquare, Clew (Outdoor navigation using GPS and voice guidance.) • Vision-Based Assistive Apps– Seeing AI, Envision AI (Object, text and scene recognition without destination navigation.) • Robotic Assistive Systems– CaBot(indoor navigation) Indian Smart Wearables– Smart Vision Glasses (SVG) limitations Identified: Indoor navigation often depends on BLE beacons, LiDAR mapping, or dedicated infrastructure. Most smartphone-based solutions require GPS, Internet connectivity, or cloud services. Existing smart wearables offer object recognition but lack seamless outdoor-to-indoor navigation with intelligent obstacle prioritization..

Proposed System Architecture.

MODULES Module 1 – Voice-Based User Interaction (phase1) Enables hands-free destination input through push-to-talk, offline speech recognition, and voice command processing. Module 2 – Outdoor Navigation System (phase2) Uses GPS and map information to determine the user's location, generate routes, and provide turn-by-turn outdoor guidance. Module 3 – Indoor Navigation System (phase2) Uses predefined node-edge maps to calculate the shortest indoor route and provide step-by-step navigation instructions. Module 4 – Visual Obstacle Detection (phase1) Uses camera input and AI-based object detection to identify surrounding objects and determine their relative positions. Module 5 – Obstacle Distance & Motion Awareness (phase1) Estimates obstacle distance and proximity while analyzing observations over time to identify stationary or moving obstacles..

MODULES Module 6 – User Position & Orientation Estimation (phase1) Uses IMU and movement information to estimate heading, orientation, walking movement, step count, and user position through PDR. Module 7 – Indoor–Outdoor Mode Selection (phase1) Determines whether the user is indoors or outdoors using GPS availability, environmental context, and entrance/exit verification. Module 8 – Multi-Sensor Fusion (phase2) Combines camera, distance sensor, IMU, GPS, and navigation information to produce reliable user, environmental, and navigation information. Module 9 – Obstacle Prioritization & Avoidance (phase2) Analyzes obstacle position, distance, motion, route relevance, and navigation context to identify obstacles that may affect the user's path and generate prioritized safety alerts. Module 10 – Voice Guidance System (phase2) Converts navigation instructions and prioritized safety alerts into audio using text-to-speech and provides offline voice guidance..

ALGORITHMS & IMPLEMENTED MODULES Implemented Algorithms / Processing Flow Teaching- Walk Records WALK_LOG Node–Edge Graph NetworkX Weighted Shortest Path Shortest path Object Detection YOLOv5 Relative Depth & Proximity MiDaS Trajectory / Waypoint Generation Planner Voice Output Piper TTS Currently Implemented Modules Module 4 — Indoor Navigation Node-edge map generation, weighted shortest-path calculation and step-by-step indoor instructions. Module 5 — Visual Obstacle Detection Live camera processing with YOLOv5 object detection, classification and relative position estimation. Module 6 — Obstacle Distance Awareness MiDaS relative depth estimation is used to determine obstacle closeness and proximity. Module 10 — Voice Guidance (partial) Piper offline text-to-speech converts generated navigation instructions..

Implementation Output – Indoor Navigation 1. Route Planner Current location + destination → Find Route 2. Route Summary & Instructions Shortest route, total steps, and step-by-step instructions 3. Generated Indoor Map & Voice Node-edge map visualization with voice guidance Implemented Indoor Navigation Functions • Node-edge graph construction from teaching-walk records • Weighted shortest-path calculation using step count • Automatic generation of turn-by-turn navigation instructions • Step-based route summary and route-path visualization • Automatic indoor node-edge map generation • Offline voice guidance using Piper text-to-speech Output: Shortest indoor route + navigation instructions + generated map + audio guidance.

Implementation Output – Obstacle Detection Real-Time Environmental Perception Camera captures the surrounding environment. YOLO-based model detects objects in real time. Detected objects are classified with bounding boxes. Relative position is identified as LEFT, CENTER, or RIGHT. Distance/proximity information is associated with detected objects..

PROJECT TIMELINE M0 Literature Survey & Requirement Analysis M1 System Design & Hardware Procurement M3 Hardware Assembly & Sensor Integration M5–M7 AI Software Development (Navigation, CV & Voice) M7 Sensor Fusion & Indoor/Outdoo r Navigation Integration M8 Prototype Testing, Evaluation & Optimisation M11–M12 Final Prototype Validation & Demonstration.

THANK YOU.