Questions · Video software
Why does motion detection give me so many false alerts?
Because classic detection does not look at what is moving, it looks at how many pixels change. A branch, a shadow that turns, rain, an insect in front of the infrared, a change of light when a street lamp comes on: all of that changes pixels and all of that triggers. The scene is almost never still, and less so at night and outdoors.
It is fixed in layers, and in this order. First, the zone: limit the detection to the part of the image that matters, instead of leaving it over the whole screen. Second, the thresholds: require a minimum object size and a minimum duration, which rules out the insect and the raindrop. Third, change method: use detection by object classification —alert only if what crosses is a person or a vehicle—, which is what many cameras and many systems bring today and what really breaks the problem.
And there is a layer that is not technical: if an alert goes off many times a day and is never anything, people stop looking at it, and from that moment the installation has its alarm switched off even though it is still sounding. A rule nobody attends to has to be removed or fixed, not left in place.
Where you can see it working.
On the IRIS Neural website, on real scenes.
All the questions.
Five hundred and forty-four, grouped by topic. Each with its own page.
Get started
See it working.
A written answer is enough to decide whether it interests you. To know whether it works at your site you have to see it on a real scene, with the image as it is and, on top of it, what the system understands from it. That is on the IRIS Neural website, which is the group's product.