AI-powered road incident information and analysis

Transport for NSW (TfNSW), the University of Technology Sydney (UTS), and iMOVE CRC have developed an artificial-intelligence-powered solution to manage information about road-transport incidents in real time. The solution transforms how TfNSW manages such incidents through data standardisation, improved insights, and decision-making capability.

The Streamlining and integrating incident data project has created a production-ready artificial intelligence (AI) tool to identify, analyse and consolidate incident information more accurately and faster than current manual, multi-party approaches. The dataset covers incidents across state roads, motorways, and local roads that impact bus routes and other road transport modes.

The developed AI model can leverage incident location, time, description and dialogue text to identify related reports from multiple stakeholders referring to the same incident. It automatically assigns the best categories based on incident descriptions and dialogue text to resolve inconsistencies, fill missing categories, and update classifications.

Background

Currently, multiple stakeholders, such as police, private and non-private operators, and traffic controllers manually log key information about incidents. This data includes start and end times, incident descriptions, types and subtypes, lane, and location details. This results in:

• Duplicate reports with, at times, conflicting information about the same incident from different parties;
• Poor data quality and reliability;
• Statistical inaccuracies;
• Unreliable analytical insights;
• Barriers to better analysing incidents to generate useful intelligence; and
• Reduced data-driven decision-making.

Incidents recorded typically include trees down, car crashes, vehicle breakdowns, police operations, tow clearways, hazards, and signal faults. Some incidents may be multi staged, such as a car crash later involving a police operation, and then the vehicles being towed away.

A police report of ‘trees down’ could be logged as ‘wires down’ by an area operator in the raw data, creating confusion about whether it relates to the same incident.

In addition, the classification of incident types and subtypes can change due to operational system updates, where some categories are disparate or new ones introduced, creating inconsistencies in historical records.

Methodology

The project explored AI because of its scalable processing power for high-volume data. AI can also make intelligent classification decisions because it understands nuanced links between incident information. It can also be retrained to absorb new incident categories and patterns while still processing historical data. At the core of this project were advanced natural language processing, data analysis and machine learning techniques.

Researchers developed two complementary AI models. The first identifies and consolidates duplicate and related incidents from multiple data sources; the second consistently and accurately classifies incident types and subtypes using a hierarchical framework that mirrors human thinking. Business and operational rules were embedded throughout the model development process, with rule-based pre-processing and post-processing steps ensuring that the AI outputs aligned with TfNSW’s operational requirements and data standards. The codebase was housed in a private online repository.

Performance evaluation

The solution uses sophisticated AI models to detect incidents – identify and group duplicate and related incidents across various data sources. The models also harness intelligent classification to re-categorise incidents more accurately and prevent duplication. Key elements of the system include:

• Automated and real-time audit and compliance features;
• Built-in warning flags, error logs; and
• Visibility into automated decisions.

The models can process 1,000-plus incidents in minutes rather than hours. removing the bottlenecks and inconsistencies inherent in the manual classification systems.

Operational recommendations

This robust and scalable AI platform addresses TfNSW’s current data quality challenges.

The system is a solid foundation for future innovations and operational needs in transport network management. Even as incident data numbers increase exponentially, the model scales automatically, without the need for manual intervention. The system could also be expanded to capture complex, multi-faceted incidents, such as a tree down caused by traffic crashes or roadwork. The re-categorisation capability ensures consistency across historical and updated classification structures while accommodating emerging operational needs.

The system is also well-placed to integrate with broader transport management systems. In future applications, the AI models could create a strong foundation for predictive incident management, drawing on historical patterns and real-time conditions. So, instead of just reacting to incidents, transport managers could anticipate the likelihood of particular events and have strategic resources at the ready. For instance, the future system might predict more breakdowns during extreme weather or pinpoint infrastructure maintenance needs before failures happen.

Conclusions

The project’s tailored AI solution transforms inconsistent approachs to managing incident data into a system that’s standardised, reliable, and allows accurate statistical analysis. The system – yet to be named – improves TfNSW’s operations from decision-making, transport network management, and efficiency, positioning it as a leader in intelligent transport data management.

The deployed AI framework represents more than a technological upgrade; it constitutes a fundamental shift in how TfNSW approaches incident data management.

Expected project impacts

This collaboration with Transport for NSW exemplifies how cutting-edge AI research can be translated into impactful, real-world solutions. We leverage advanced machine learning, probabilistic modelling, and intelligent decision systems to push the boundaries of applied AI research in complex transport environments. We believe this project will set a precedent for how AI can be responsibly and effectively integrated into public sector innovation.

UTS Australian Artificial Intelligence Institute (AAII)