Auckland-based principal transportation analyst Soroush Rashidi has spent the past few months buried deep in lines of computer code – developing a system with Auckland System Management that automatically recognises, counts and classifies cars, trucks and motorbikes, and estimates speed.
The system works thanks to a field of AI called ‘computer vision’ - where computers are trained to interpret the visual world and derive meaningful information from digital images and videos.
Soroush’s system has already been trialled with pre-recorded traffic footage from around Auckland. The next stage of the trial will see the system run with real-time footage alongside conventional vehicle ‘loop detectors’ on State Highway 2.
Soroush says most single loop detectors, which sit under the road, just count vehicles, can’t reliably classify vehicles and aren’t very accurate at estimating speed. The classification algorithm he has developed has been especially trained to identify different kinds of vehicles.
AI systems like these are set to revolutionise access to data for transport and traffic engineers. Traffic count and classification data is aggregated and can be viewed on online dashboards and in Excel spreadsheets.
“Replacing or supplementing loop detectors with computer vision systems means transport authorities will benefit from higher-quality, more granular data on traffic conditions. Plus, the system can quickly be installed anywhere,” he says.
“There are so many ways we can use the system, including better understanding annual average daily traffic (AADT). More informed traffic management decisions can then be made around queue and signal times, pavement design and asset management.
“Knowing more about the ‘what’, ‘where’ and ‘when’ of vehicle use is a boon for traffic safety. Being able to see crashes and near-misses with computer vision lets transport designers identify dangerous driving zones and where road safety upgrades need to happen.”
Soroush says computer vision technology can also help with understanding the impact that traffic interventions have had on how people use the road network.
A software version of the system will be up and running by the end of June. Watch this space.