Catching Fly-Tippers in the Act: Fortix AI's Waste Detection on 4G Covert Cameras

A textbook UK fly-tipping incident — a small roadside litter point overwhelmed by what's clearly bulk household waste, dumped overnight. Defra recorded 1.15 million incidents like this across England last year. Photographing them after the event is the easy bit; catching the offender in the act is what changes the maths. Defra's most recent fly-tipping statistics make uncomfortable reading. English local authorities dealt with 1.15 million fly-tipping incidents in the year ending March 2024, the second-highest number on record. The total clearance cost to councils was over £14 million. £13.2 million was spent on enforcement actions, of which only a tiny minority resulted in successful prosecution. The vast majority of incidents are never traced back to the offender at all. The reason is not a shortage of cameras. It's that the cameras most councils have deployed in fly-tipping hotspots — fixed CCTV at the entrance to a layby, motion-trigger trail cameras in a forestry track, traditional ANPR at the road junction — produce evidence after the fact. By the time the bin lorry crew or the parks team finds the tip on Monday morning, the offender is long gone and the camera has, at best, a low-resolution clip of a vehicle from twenty metres away with no plate readable. This is the exact gap that Fortix AI's waste detection model — running on our 4G covert cameras and paired with the Fortix ANPR model — was built to close. Here's how the workflow runs end to end on a UK council deployment. The fly-tipping detection model Fortix's fly-tipping detection isn't a generic "object in scene" alert. It's behaviourally trained on the specific sequence of events that defines a fly-tipping incident: vehicle approaches the location, vehicle stops, occupant gets out, vehicle's tailgate or rear doors open, material is deposited, occupant returns to vehicle, vehicle leaves. The model fires the moment that pattern is recognised, not when the first pixel changes. That distinction matters more than it sounds. A motion-trigger camera on the same lane will fire on a fox, a deer, a dog walker, a contractor checking a fence, the Royal Mail van turning round and every other piece of legitimate rural traffic. Those false alarms train the enforcement team to ignore the alerts within two weeks, after which the camera might as well not be there. The behavioural model only fires on the actual fly-tipping event sequence. Paired ANPR: the bit that makes a prosecution possible The detection of the act is only half of the evidence pack. The other half is the vehicle. Fortix AI's ANPR model runs in parallel on the same camera stream, capturing the number plate plus the make, model and colour of the vehicle that arrived, deposited the material and left. Make and model together with the number plate handle the obscured-plate case (mud, deliberate covering) — even when the plate is partially unreadable, the make/model/colour combination plus partial digits is often enough to identify the vehicle in DVLA records. The output is a structured incident report: the timestamped detection event, the captured number plate, the vehicle description, a 30-60 second video clip of the act itself, and a metadata block aligned to council enforcement templates. The investigating officer doesn't have to assemble the evidence pack — Fortix produces it automatically. Why 4G covert cameras are the right hardware A fly-tipping hotspot has three characteristics that conventional CCTV hardware is badly suited to. It usually has no mains power. It usually has no broadband. And if the camera is visible, the offenders simply move 200 metres up the road to a different layby. The combination that works is a 4G mobile CCTV camera, often solar-powered for the off-grid case, deployed covertly. Our covert range hides the camera in everyday-looking enclosures — a small box on a fence post, an apparently unmarked utility cabinet, a forestry-style camo housing on a tree. The 4G SIM handles the connectivity. A modest data plan (typically 20-50 GB/month) easily covers the structured event uploads from a Fortix edge box, because video doesn't continuously stream to the cloud — only the actual detection clips do. Power options vary. For a long-deployment hotspot, solar with a battery bank gives months of unattended operation. For a shorter-term enforcement campaign, a lithium pack good for 2-6 weeks is often the right answer. For sites with mains access nearby, a covert PoE installation is straightforward. The Fortix edge box can be co-located with the camera or hidden separately depending on the site geometry. In either case, the inference runs on-site, which means the camera is operational with no broadband and the data plan stays modest. Bulkier single-load fly-tips — mattresses, broken furniture, builders' rubble — produce clearance bills that run into the hundreds per incident before any enforcement officer time is even charged. These are the events where the financial argument for AI-driven detection lands hardest. The economics for an enforcement team The financial argument for AI-driven enforcement is straightforward and worth running honestly. Per Defra's data, the average cost of a single fly-tipping clearance is somewhere between £40 and £85 for a small fly-tip, rising into hundreds for a single-load tip and into thousands for a multi-load event involving hazardous materials. A council enforcement officer's time, scaled by the number of incidents per month at a single hotspot, runs into the tens of thousands of pounds a year before the first prosecution. A successful prosecution under the Environment Protection Act 1990 attracts a fine up to £50,000 and recovers clearance costs. The Fortix-on-covert-camera deployment shifts the maths in two directions. The first-time enforcement actions go from a tiny fraction to a substantially higher conversion rate because the evidence pack is in the inspector's inbox the moment the incident happens, complete with the…