Map Data Analysis for Lawn Service Optimization

This research investigates the fusion of multi-source geospatial data to overcome key scaling limitations in automated property assessment. The primary research problem addresses object occlusion in satellite imagery, where features can obscure ground-level surfaces and prevent accurate area measurement. We propose and evaluate a novel machine learning approach that geometrically fuses satellite imagery with ground-level street views, when applicable. A semantic segmentation model will be trained on this data representation to better compute residential lawns and driveways, with the hypothesis that this approach will showcase significantly higher accuracy than analysis from a single data source.

A secondary research objective explores the development of a trustworthy AI agent for data enrichment. We will construct a system that integrates the new geospatial measurements with historical data using a Retrieval-Augmented Generation (RAG) framework. This grounds a Large Language Model in a factual knowledge base, while rule-based guardrails further constrain its output. The efficacy of the system will be validated through quantitative metrics, including Intersection over Union (IoU) for segmentation performance and consistency evaluations for the LLM agent, contributing a methodology for both applied computer vision and controllable AI systems.

Faculty Supervisor:

Tom Cesari

Student:

Partner:

Ezi Home Services

Discipline:

Computer science

Sector:

Information and cultural industries; Professional, scientific and technical services

University:

University of Ottawa

Program: