Hertfordshire County Council has shown how to develop an AI solution in response to a business problem, writes Ellen Wilson, EMEA public sector specialist at AWS
Failing fast and often can be a great way to approach digital projects, according to Josh Wahnon, as it makes it possible to learn quickly, correct the mistakes and ensure a new tool or service works effectively when it is deployed.
The data scientist for Hertfordshire County Council (HCC) explained the success of the approach in developing an AI system for exploring the details of traffic surveys in a presentation to the recent UKAuthority AI & Data4Good conference.
Working with AWS, the council has built a system that is now delivering significant efficiencies and provides a standardised approach for monitoring traffic.
The project derived from the monitoring of traffic flows across Hertfordshire to inform decision making. It uses video of traffic and counts all the vehicles by type, while also working out the direction of travel.
HCC wanted to understand the traffic in more detail and to address four main points: the people counting at roadsides were at risk; they spent hours counting at the roadside; they spend hours calculating the results; and they spent hours updating and maintaining a database of the survey results going back more than 25 years. This led Wahnon to identify five objectives: to capture all the vehicles; identify them correctly, identify the directions in which they are travelling, automate the process of record keeping, and make the tool available to all the staff who need it.
Wahnon built a solution architecture involving a front end app developed on the AWS Amplify platform, which collects video footage from using roadside cameras over 12-hour slots and sends them to an S3 bucket. They then go into the AWS Fargate compute engine, where the details are identified through AWS Rekognition image and video analysis tool. The results of this are then fed into the database.
He showed a demonstration in which the system identifies a moving vehicle and places it in a box with a confidence score on the direction in which it is travelling. While there have been some problems identifying heavy goods vehicles the system is being trained to make itself more accurate. It is also able to split the 12 hours of footage into half-hour segments and process all of them at the same time, which is a big time saver.
It is accompanied by a website – in testing at the time of his presentation – enabling users to click on an uploaded video and view results on the numbers and types of vehicles going through a point over different periods.
HCC has seen a number of benefits from the system: the ability to process a video from a 12-hour survey in just two hours; in relying on cameras rather than humans for the counting. In initial trials, the application was able to capture 91% of vehicles that a human counted over 12 hours but was able to process this in just two hours.
Standardising the approach
The system will standardise the way HCC collects traffic data and reduces the risks from working on the highway network by reducing exposure time.
In addition, the system only picks up details of a vehicle and where it is going, then automatically deletes it after the count has been processed. It does not collect any personal data and therefore prevents any concerns over privacy.
The project derived from a living lab where HCC, in collaboration with other organisations, tests new technologies for data collection and automation. Summing up the project, Wahnon added: “This started up as a complete experiment. When we drew up the UML (unified modelling language) use case diagram we didn’t know if half of it was possible, and as a result asked AWS for help when we had bits and pieces working.
“For me this has all been an exercise about how I can get this application off my laptop and share it securely with my colleagues. In that sense AWS were incredibly helpful, guiding us through the architecture of the solution.
“Finally, break it down into small goals – can we identify this vehicle, can we track this vehicle, can we track many vehicles – and work in short sprints so you have an overall goal broken down into short tasks that are easier to complete.”
Start small, think big
Hertfordshire’s experience reflects the advice of AWS in approaching digital projects: start small but think big; pick a business problem and work backwards; migrate a business app to the cloud; pick a digital solution to answer an old problem; rationalise apps; and invest in staff and change the culture.
The company can provide a robust and flexible cloud platform for developments, a range of technologies through its industry partners, and has a team with great experience of working with public sector bodies to understand their business challenges and develop solutions.