Rahul Yadav, chief technology officer, from Milestone Systems discusses how open platform unlocks trusted video data and best-in-class video analytics for smarter, safer cities.
The future of smart city technology isn't being shaped in Silicon Valley – it's taking root in Dubuque, Iowa. With a population of about 60,000, this mid-sized city has become a live testbed for AI-driven traffic management thanks to a unique public-private collaboration that demonstrates how cities can transform urban mobility and safety through responsible technology, without costly infrastructure overhauls.
Overcoming the challenges of high-potential AI, low-quality data
Despite extensive camera deployments, cities often struggle to derive real-time intelligence from video. AI models trained on synthetic or generic datasets frequently underperform in complex, real-world environments. High false-positive rates make them impractical for daily operations, especially in time-critical applications like traffic management, preventing municipalities from gaining the real-time insights needed to improve mobility, safety, and emergency response.
Public-private innovation in action
The solution at play in Dubuque – Project Hafnia, which was led by Milestone Systems – used the AI vision platform Vaidio to develop a high-performance, real-world AI model which was deployed across the city region. It’s a great example of shared values and visions, with each party contributing expertise, resources, and infrastructure to align toward a common goal.
At the heart of this is the secure, high-quality, and legally compliant data library that supports transparency and data traceability requirements in evolving AI regulations. Milestone transformed the city’s raw traffic footage into valuable AI training material. This investment paid off dramatically, with model accuracy jumping from 80% to over 95%. Below 90% accuracy, false positives are too frequent and undermine usability. Above this critical threshold, cities unlock new levels of accuracy and insight.
Mapping the results
With more than 100 traffic cameras participating in the trial, Project Hafnia validated AI performance in the field across a range of changing light, weather, and traffic patterns. The resulting platform is scalable and transferable, giving other cities a tested blueprint for AI-enhanced urban operations. AI models originally developed for traffic monitoring can now be repurposed to support public safety, emergency response, and infrastructure planning across departments.
The City of Dubuque used video analytics to support real-time traffic monitoring, counting vehicle volumes, analysing turn movements, and refining signal timing for optimised flow. Automated anomaly detection helps identify wrong-way vehicles, accidents, and other hazards, allowing for quicker response.
The results speak for themselves: smoother traffic flow has reduced travel times and cut congestion, while faster emergency response times and enhanced incident detection have contributed to a safer urban environment. A reduction in idling has helped support the city’s sustainability initiatives, while the automation of traffic management processes has freed up city personnel to focus on high-value tasks, ensuring resources are used effectively.
The results alone are exciting, but what makes this project unique is the approach to data. The public-private innovation structure gives access to real-world scenarios that simply can’t be replicated in a lab. At Milestone, we’ve invested thousands of hours in annotating video data so that Vaidio can train their models on footage that precisely matches what they’ll encounter in production environments.
Compliance and transparency at the core
As regulatory frameworks like the EU AI Act continue to evolve, data traceability and legal compliance have become essential. These challenges can be addressed by creating a transparent, documented data supply chain that future-proofs analytics solutions against evolving regulatory requirements.
This data library offers complete traceability: every frame of video used in AI training — thousands of hours of footage — is documented with its source, processing history, and usage permissions. Adhering to processes like EU GDPR, which provide foundational safeguards for processing and handling personal data, ensures compliance and establishes a foundation for ethical, responsible AI development.
A model for cities worldwide
Project Hafnia not only demonstrated that real-world AI can work, it showed that it can scale, ethically and collaboratively.
As we continue to refine and develop new features, we’re opening up possibilities we hadn’t even considered before. The power of AI to transform urban management is remarkable. It’s giving us insights and capabilities that would have been impossible with traditional methods, all while helping us make our cities safer and more efficient for residents.
This is a blueprint for how cities everywhere can harness AI to improve safety, mobility, and quality of life — all while staying in control of their data.