banner
MIDS Capstone Project Spring 2025

MapAI

Team members
Ahmeda Cheick
Guanghua Li

Our Mission

At MapAI, we are building the first auto-mapping solution using AI to transform imagery into precise geolocations. Our vision is a world where everyone can define their own address, making every place visible and reachable.

Meet Our Team

Ahmeda Cheick - Chief Mapper

James MacLean - Geospatial Genius

Guanghua L - Waypoint Wizard

The Problem & Our Solution

The Problem

In today’s world, 4 billion people  live without formal addresses. They depend on word‑of‑mouth directions—“turn left at the mango tree”—or landmarks that can vanish overnight. This creates serious hurdles for last‑mile delivery drivers, medical responders racing against the clock, and anyone simply trying to find a friend’s home. Existing addressing solutions force users to memorize cumbersome alphanumeric codes (e.g. “AB.CD.EF”) or navigate labyrinthine menus, making them inaccessible to large segments of the population. In Summary:

  • 4 billion people globally lack formal addresses
  • Existing mapping solutions are not user-friendly
  • Unstructured urban growth leads to navigational chaos
  • Governments often fail to provide reliable addressing solutions

The MapAI Solution

MapAI addresses these challenges with a user-centric web and mobile application: simply tap any feature on a high-resolution satellite, street-view or 3D map—whether it's a building, gate or landmark—and our platform instantly captures the exact latitude/longitude, constructs a unique geofence from the associated image, and issues a shareable link. By leveraging DeepLab v3+, SAM for precise segmentation, and CLIP for embedding-based validation, MapAi eliminates complex alphanumeric codes and manual lookups, enabling anyone—from delivery drivers to emergency responders—to pinpoint places, and share locations with accuracy in seconds. Here's an overview on how it works: 
 
diagram
 
 
diagram2

Our Technology

Entrance Detection Pipeline

Our system integrates GPS location with Open Street Map data to identify entrances with precision. We use advanced models to enhance navigation and accessibility through Google Street View.

Comprehensive Dataset

We utilize Open Street Map and aerial images, providing entrance GPS coordinates, building polygons, and more. Our data coverage spans the globe, with OSM data from the top 10 cities on each continent and 2,000 satellite images per city.

AI-Powered Matching

Our model uses Clip for embeddings, comparing query images with existing data. This process ensures the closest match and confirms locations with high accuracy.

Performance Metrics

93% of entrances predicted within 20 meters of actual location

62% of predictions correctly identify the proper side of the building

Future Roadmap

UI Development

Creating a more user-friendly experience to make mapping accessible to everyone.

Model Improvement

Continuous enhancement of our AI models to achieve even higher accuracy in entrance and location detection.

Business Expansion

Strategic growth to expand our impact and bring our mapping solutions to more regions worldwide.

More Information

Project website
capstone-presentation-final-mapai.pdf
Last updated: April 18, 2025