Detect the dumping and the overflowing bin, within minutes
Abandoned bulky waste gets noticed when a neighbour complains, usually days later. And the collection round runs the same whether the bin is full or empty. The same pole and the same camera fix both.
«The bin raises the alarm when someone leaves a sofa next to it. Not the following week: twenty minutes later.»
- Client
- Town councils, joint authorities, collection concessionaires
- Input
- Pole-mounted edge camera overlooking the drop-off point
- Anomalies
- Background subtraction + discarding people and vehicles
- State
- Classification: normal / full / overflowing
- Persistence
- 20 – 30 min before alerting
- Output
- Photo alert, per-bin state, statistics by point and time band
Two problems, two techniques
Abandoned bulky waste is an anomaly: something appears where there was nothing and stays. It is solved with background subtraction, discarding people and cars using the detector, and requiring the object to persist twenty or thirty minutes before alerting.
A full bin is a different thing: it is a classification of a fixed crop of the image into normal, full or overflowing. Annotating that means sorting photos into folders, hundreds an hour. It is among the cheapest things to train there is.
Why hysteresis is everything
Nothing fires on the first detection. Waste must persist for several minutes and a bin must read full across consecutive readings before an alert goes out. It is the only thing separating a product from a demo.
The number one cause of cancellation in systems like this is false alarms. We would rather miss an event than send the crew out twice for nothing, and the thresholds are tuned in that direction.
Can be extended with
- LoRaWAN ultrasonic sensor for the internal level
- Public-facing dashboard
- Hotspot report
- Collection route optimisation
Does it detect the fill level inside the bin?
No. The camera sees the overflow outside and what is dumped beside it, not the level inside. For the level we fit an ultrasonic sensor inside and combine it with the camera on the same dashboard. We say this in the first meeting, not once it is installed.
Does the alert include a photo? What about privacy?
It does, but only when there are no people in the scene. Abandoned waste is static and in no hurry: the system waits for the first clean frame. That resolves privacy without blurring anything, and face and plate blurring remains available as an option.
Does the pole need network coverage?
Not necessarily. The state fits in a few bytes, so a node can publish over LoRaWAN with no SIM and no data plan. If the alert must carry a photo, then yes, more bandwidth is needed.
Got problem drop-off points?
Starting with the worst point on the list is the fastest way to know in six weeks whether this is useful to you.
Request a pilot