IPL worked closely with UK Police forces to develop IP ANPR, a tool to analyse the huge amount of ANPR data recorded by police forces, councils, companies, and other organisations in the UK. Typically ANPR cameras capture the number plates of vehicles passing the camera and this information – the number plate, date, time and location – is stored in a central database. This often leads to databases containing thousands or millions of ANPR sightings.

Extracting accurate information from this data set – although it seems obvious at first – can be very hard.

One of the questions asked by the police was how to identify vehicles in convoy – to find a ‘suspect’ vehicle travelling with a known ‘target’ vehicle. For example, if you have a security van with a large amount of cash inside, it would be really useful to know if it was being followed by a criminal.

A simple and obvious analysis might be – every time the van goes through a camera, then record all the cars that go through the same camera within a set time, say 30 seconds. Then, check to see if there are any VRMs (Vehicle Registration Marks or number plates) that keep showing up. This is very simple to do, and many current ANPR systems provide a ‘convoy analysis’ that simply counts up the number of times each VRM has been seen and displays a list.

However, this simplistic analysis raises a lot of ‘false positives’ – results that you are not looking for – and this both obscures genuine results and destroys confidence in the system.

We analysed the data for several ‘target’ vehicles and found that with a simple analysis it was impossible to tell whether a vehicle was or was not actually being followed. There were several situations which tended to raise ‘false positives’ or confuse the issue:

• Cars DO often follow each other. Any line of traffic on a main road passing two cameras often contains the same cars in a very similar order. This can be either short-term (through two cameras in a town centre) or long-term (across many miles of motorway).
• If two drivers have regular habits, then it is easy to cross paths with another driver for a short time. For example, I often leave the house at 7:30 in the morning: Often, my neighbour also leaves for work at the same time, and we end up driving behind one another towards the main road. With a simplistic analysis, it can look like I am following him 5 times a week – despite the fact that we end up miles apart.
• If my neighbour and I both return home at 5pm, and leave the next morning at 7am, a simple time-based analysis may conclude that I have been following him for 14 hours!
• Criminals following security vans may try and ‘hide’ the results – because there are often two or three cameras on dual carriageways and motorways, one on each lane. All they have to do is move lanes, and they’re not caught by the same camera.

Taking these situations and others into account, it became obvious that any simple analysis – number of sightings, time, or distance – was not sufficient.

We developed a set of heuristic rules, optimised by a feedback loop, to calculate and evaluate for each possible suspect vehicle:

• How long (time) the suspect vehicle follows the target vehicle.
• How far (distance) the suspect vehicle follows the target vehicle.
• Automatically include sightings from related cameras – for example, all north-facing cameras on a motorway bridge.
• Automatically exclude any times where the target has stopped for several hours – eliminating vehicles ‘following’ it overnight, or when it is parked.
• Take account of the status of the target vehicle. It may be already known to the police as being associated with a convicted criminal, or may have been recently stolen.

Taken together, the simple initial request ‘is anyone following this security van’ actually adds up to relatively complex problem.

IP ANPR was developed to automatically read, understand, and solve complex problems like this and present it in a simple, easy to understand interface. The user enters the target vehicle, presses a button, and gets a list of possible following vehicles. The devil is in the details.