Man Uses Ring Camera, Sonic Waves to Deter Pests

A man has "trained" his Ring camera system to detect and deter badgers and foxes. It's an example of machine learning having a practical purpose.

James Milward developed the comically-titled "Furbinator 3000" after being frustrated by the unwanted visitors damaging his garden. He also feared they could carry diseases that would be a risk to his two young children.

He tried several traditional solutions without success, before deciding an ultrasonic repellent was the only solution. However, this had several limitations including that they were triggered during the day by humans, leading him to switch them off and forget to turn them on again. He also found he couldn't always keep track of whether they had been fully charged by their solar panels or if they needed to be temporarily plugged into a power outlet for a top-up.

System Trains Self

Eventually he decided the solution was to connect the devices to his home network and then try to find a way to specifically detect badgers and foxes. This would not only avoid them being triggered by other creatures, but would mean they could emit the sound at a specific frequency for either a badger or a fox.

The basic concept was simple enough: Milward used Ring cameras to monitor movements and then connected the footage to TensorFlow, a machine learning system, to recognize the creatures. Using existing recognition models didn't work well as it detected the creatures as everything from a bear to an umbrella and even a sink. (Source: bbc.co.uk)

Milward then created and trained his own recognition model, which is a perfect example of machine learning. He fed in footage and marked up cases when a badger or fox was visible. The machine learning system then compared this with frames in which neither animal was present and developed its own rules for distinguishing the creatures.

Grass Patterns Cause Headaches

The initial results weren't perfect, with a detection rate of 62 percent. Milward eventually worked out the problem was false positives caused by patterns in grass and foliage. He then combined the system with a motion detector and set the system to retrain with the added insight that real foxes and badgers were much more likely to move about. (Source: medium.com)

While the time he spent on the project might only appeal to serious tech enthusiasts, Milward is happy with the results and has already seen footage of badgers being detected and deterred.

What's Your Opinion?

Is this a creative use of technology? Can you think of other cases when systems developing their own image recognition rules would be useful? Should we be more concerned about image recognition systems used on humans?