Partner Introduction

Microsoft's Soundscape project, in collaboration with the Wheelchair Alliance, aims to revolutionize urban navigation for people with mobility needs. This partnership brings together technological innovation and real-world accessibility requirements.

Microsoft Soundscape

A technology leader in accessibility solutions, providing technical expertise and platform integration capabilities.

Wheelchair Alliance

A community-driven organization offering invaluable insights into accessibility needs and user requirements.

Project Background

Current navigation systems lack comprehensive accessibility information, creating significant challenges for wheelchair users in urban environments.

Current Challenges

  • Limited accessibility data in existing mapping solutions
  • Manual data collection is time-consuming and often incomplete
  • Lack of automated systems for identifying accessibility features

Opportunity

  • Leverage satellite imagery for automated feature detection
  • Create a scalable solution for accessibility mapping
  • Integrate with existing navigation platforms

Project Goals

Primary Goal

Develop an AI-powered system to automatically detect and classify accessibility features from satellite imagery.

Technical Objectives

  • Implement robust feature detection algorithms
  • Ensure accurate georeferencing of detected features
  • Create a scalable and maintainable system architecture

Success Metrics

  • High accuracy in feature detection (>90%)
  • Efficient processing of large-scale imagery
  • Seamless integration with existing platforms

User Personas

Joe Goodman

Joe Goodman

Developer

"I need accurate APIs and data that can be significant"

Motivations

  • Wants to integrate satellite data into their applications
  • Looking for well-documented APIs and Accessibility Tools
  • Being able to recognize points of interest
  • Using data for meaningful insights

Pain points

  • Inconsistent results and unreliable data
  • Lack of flexibility to integrate points of interest
  • Technical and visual documentation issues
Emily Carter

Emily Carter

Charity Employee
Accessibility Advocate

"I want to direct wheelchair users to aid their social awareness"

Motivations

  • Improve accessibility for wheelchair users
  • Empower users to navigate safely
  • Making points of interest accessible
  • Helping users find suitable routes

Pain points

  • Limited access to accessibility data
  • Difficulty in finding suitable routes
  • Lack of real-time accessibility information
  • Complex user interfaces

MoSCoW Requirements

Project requirements prioritized using the MoSCoW method for development until end of December

Must Have

Provides Minimum Usable Subset of requirements

  • Image Segmentation & Georeferencing: Process regional coverage images into 20mx20m tile regions with 1m overlap, using Easting/Northings, Latitude/Longitude and What3Words tags
  • Pedestrian Crossing Recognition: Implement classification model for feature detection and bounding box regression for crosswalk location and dimensions
  • Maximum Reliability: Prioritize model reliability and safety over optimization and coverage completeness to ensure user well-being

Should Have

Very important but can ship without

  • Automated Data Lookup: Location-based aerial/satellite data download system
  • Park Bench Recognition: Classification model for bench detection with location and dimension analysis
  • Platform Integration: Compatibility with Group 14's platform and interface provision
  • Multi-format Support: Process various aerial/satellite data formats into compatible model inputs

Could Have

Nice to have features

  • Automated Annotation: OpenStreetMap integration for training data generation
  • GUI/WebApp: User-friendly interface with drag-and-drop features
  • Extended Recognition: Additional feature detection (e.g., ramps)
  • Obstruction Detection: Cloud coverage analysis and multi-dataset cross-referencing

Won't Have

Beyond current scope

  • Hazard Detection: Due to resolution limitations in current datasets
  • Live Updates: Analysis limited to historical and user inputs
  • Speech Interface: Natural language processing system (shelved for now)