The Challenge:

There are more than 3,500 unmanned and therefore high-risk level crossings across the UK. Our customer needed to improve the safety of those with the highest risk of accident and fatality. Our task was to create an artificial intelligence and machine learning system that could detect people and vehicles on or near the line. This new system would speed the time it takes to alert the control system, in turn leading to a reduction in fatalities.

The Approach:

Firstly, we looked at multiple applications of artificial intelligence – from time series forecasting to machine learning, computer vision, decision trees and robotic processing automation. The best solution was to develop a machine learning application to capture data and improve the accuracy of the existing risk model.

To achieve this we:

• Developed the use cases and defined the requirements and testing

• Coded a convolutional neural network in multiple computer languages

• Trained and tested the network to drive accuracy into the model

• Developed a bespoke user interface

• Showcased the tool in a live demo to the customer

By replacing the humans in the loop, Our customer has increased the speed of detection and reaction to risks at level crossings. We designed a user interface that is easy to use with minimum training, but provides accurate, important information. This system will be used by the Rail Safety and Standards Board and, following trials with Network rail, has the potential to be used by the British Transport Police and the Department of Transport.