OrtoSense

AI platform that detects map changes in aerial orthophotos and automates cartographic updates

Context

Automating territorial map updates with AI

OrtoSense is an AI-based platform built for public agencies, technical offices, and surveying firms that manage large volumes of territorial data. The system compares historical orthophotos with new acquisitions to automatically identify areas that have changed over time. Deep learning models and transformer architectures detect variations, segmentation errors, and the specific cartographic layers involved. The output is a set of precise coordinates for modified areas, complete with multi-level classifications, ready to feed into CAD update workflows. The platform handles large territorial extents in a scalable way, maintaining high accuracy and operational continuity. A dockerized, modular infrastructure makes deployment straightforward on both local servers and cloud environments, simplifying maintenance and future updates.

What we did on it...

Vision

Project Conception

The project started from a concrete operational problem: updating territorial maps from aerial orthophotos is slow, manual, and expensive. The goal was to replace human review cycles with an automated, AI-driven detection and classification pipeline.

Structure

Architecture and Model Design

The system was designed around deep learning models and transformer-based architectures for temporal image analysis and geospatial classification. A modular, dockerized infrastructure was chosen to support both on-premise and cloud deployments, keeping components independently updatable and scalable.

Interface

Application Development

The platform was built to ingest orthophoto pairs, run change detection, and return classified coordinates with multi-level cartographic tagging. Integration points were developed to feed results directly into existing CAD territorial workflows, reducing manual intervention to exception handling only.

Launch

Testing and Release

Validation covered detection accuracy across varied terrain types, segmentation consistency, and classification precision at multiple cartographic levels. Performance benchmarks confirmed the system's ability to process large territorial extents within acceptable time and resource limits before production release.