Automated assessment of road safety and conditions
CQUniversity researchers have been working with industry partners to develop a deep learning-based, automated system that analyses video data to assess road safety and conditions for the purpose of identifying safety issues and prioritising road safety maintenance.
Queensland Department of Transport and Main Roads (DTMR) and Australian Road Research Board (ARRB)
The project has the potential to provide industry with the information needed to boost operational efficiencies, and most importantly, improve overall road safety.
Professor Brijesh Verma
With almost one million kilometres of roads across Australia, maintaining this vital infrastructure is a major challenge for local, state and federal governments. The condition of roads influence user safety and research has shown poor road quality is a major contributing factor to motor vehicle accidents and subsequent driver injuries and fatalities.
Currently, manual systems used for road maintenance and safety, not just in Australia, but around the world, are inefficient and prone to error. The main challenge in assessing road safety is being able to accurately detect, segment and classify all road objects and attributes, while also calculating distances between objects.
Researchers from CQUniversity’s Centre for Intelligent Systems (CIS) have been working with industry partners, including the Queensland Department of Transport and Main Roads (DTMR) and the Australian Road Research Board (ARRB), to develop a deep learning-based, automated system that analyses video data to assess road safety and conditions for the purpose of identifying safety issues and prioritising road safety maintenance.
The research involves the automated extraction of road attribute information from digital video recordings (DVR) that use advanced image analysis. This information is then cross-validated with other data sources. The method has allowed researchers to gather information on safety attributes (set out in the AusRAP and iRAP star rating).
The preliminary experiments have shown that gathering information using deep learning methods is both time efficient and cost effective and provides the ability to identify new or emerging hazards that could impact road safety conditions. The research and subsequent field-testing will also inform the development of a tool to estimate an overall road rating.
Professor Brijesh Verma, Director of the Centre for Intelligent Systems, says that by identifying road attributes and assessing these against national safety compliance recommendations, the research has the potential to inform the creation and maintenance of safer roads.
"This research will develop novel tools for segmentation and classification of roadside objects/AusRAP attributes and classification of those objects and distances between them to identify overall road safety and conditions. We will also be able to develop a tool to estimate and assign overall road ratings."
"To date, the project has seen us develop and apply an automated assessment technique that uses existing data and assets to review and analyse road attributes that influence conditions and safety."
"The methodology was developed in conjunction with industry to allow for the automation of data assessment in order to save time and money and improve decision-making processes."
"Experiments have so far shown that the automation of assessment is delivering positive outcomes and has a high accuracy in identifying AusRAP attributes," says Professor Verma.
The CIS is continuing to work with ARRB and DTMR partners to refine the deep learning techniques being used for large data and further improve the detection and classification accuracy of roadside attributes, so that a fully automated system can be developed.
The projected long-term outcome of the proposed research work is an objective, automated method and software for analysing roadside objects and determining the road safety/rating.