Researchers in Leicester’s School of Computing and Mathematical Sciences and Department of Neuroscience, Psychology and Behaviour are collaborating on the study, which will track participants’ eye movements when faced with multiple fast-moving hazards simultaneously.
The research will be used to inform the next generation of a rail safety device being developed as part of an Innovate UK Knowledge Transfer Partnership (KTP) between Leicester experts and systems engineering specialists Synoptix.
Installed at a Network Rail-operated level crossing near Cheltenham earlier this year, the OPTIMUS prototype uses machine learning and an AI-based object detection system, hosted locally on the small edge-based device, to identify and quantify different types of traffic.
Now, the new interdisciplinary aspect will allow researchers to compare the accuracy and speed of its detection capability to a human completing the same task.
George Leete, KTP research associate within the Artificial Intelligence, Data Analytics and Modelling (AIDAM) Centre at Leicester University, is leading development of the project’s machine learning aspect under the supervision of Professor Ivan Tyukin.
More on artificial intelligence
Leete said that the study aims to lend valuable insight into the performance of the OPTIMUS system at the crossing, how good humans are at the current task and whether the system holds up to human standards.
“We believe this is the first time an AI-based system will be validated against a standard in this way, opening the door for other systems to be validated via a similar method,” Leete said.
According to Network Rail, there are around 6,000 level crossings in the UK. Figures for 2019/20 show that there were 316 near misses with pedestrians on UK level crossings, and two pedestrian fatalities.
The OPTIMUS prototype was installed at a site on the Cross Country Route in January 2022 and has already identified hundreds of thousands of movements on the crossing, including pedestrians, cyclists and other road traffic. Identification and categorisation of users occurs locally on the device, and the only data transmitted is of traffic numbers and types, so privacy of crossing users is protected in line with data protection guidelines.