Neural Network recognizes the objects: person, car, animal...
Let's honestly look in the face of the cameras! Oh, I mean, the lenses! What they see?
In most cases, they observe the natural interferences all their life. This is our reality. It is so arranged, although we do not notice it much, the environment is overloaded with various activities: the movement of branches, shadows from these branches in the sun, the flight of birds, the running dogs, crawling flies on the cameras and the glare from puddles... For people, all this is normal, so interferences have even become natural. But there is a problem in electronics, in the computers; the usual video detector is triggered on everything that moves. Any change in the picture – and we have a huge pile of useless recordings, as well as the need for the surveillance operator – to monitor it.
90% of the whole process of surveillance is a game of light and shadow outside. Inside is much smaller, usually 20%, but even one is a lot! Why do service people jump every time a branch twitches? In general, we are not naturalists, and should not monitor natural phenomena, but keep an eye on the safety of the facility.
And now, thanks to artificial intelligence, which no one has ever seen, but many believe in it, there is an opportunity to use neural network. You activate it – and all the chaos in the universe makes sense! We begin to understand... Oh, sorry, we the people already knew the difference between a twig and a human - God gave neural network to people a long time ago. But now the computer begins to understand what areas and objects are really interesting for us.
Now it is obvious to both of us what objects in the frame should be monitored. Yes, people are mostly interested in the actions of cars and humans – no matter what the interference looks like being a car or a person. And the computer can now recognize real aims in almost any form, not giving the operator a pile of rubbish, but only designated objects. Even in difficult, inappropriate conditions.
In this storm of interference, the neural network can find useful aims. The usual video detector works on everything that moves, but the neural network is able to selectively respond to certain actions of a person or a machine. It successfully distinguishes between the chaos of a branch moving , light and shadow.
Man constantly changes his parameters, moving in bright and in dark areas, in front and behind branches, changing his color scheme, the SpesLab neural network recognizes it in any form of light, from any angle of the camera, at different sizes both in the background or at the front.
And the SpesLab neural network is able to classify objects of all types, even if a person or a car is not in a good view of the camera: being in front at the side or behind, which is especially important for most surveillance systems – from above.
The neural network is able to classify where people, animals, machines or any other kind of objects are. On this basis, we can selectively assess their threat.ems – from above.
If necessary, we can protect ourselves from such interference like animals – our four-legged friends are in a separate category. And their activity will no longer disturb the peaceful sleep of the guard, and the owner's phone will not have false alarms. As you can see on the dog there is no outlined frame like with a person – no matter how it is spun.
We can also evaluate other images – for example, clothes, uniforms, hats, weapons, etc.
Unfortunately, the most neural network developers usually teach them on high-quality Internet pictures – without taking into account the specifics of life, but Speslab takes training material from numerous means of video surveillance of its customers such as "Distant witness".
Also Speslab recognizes useful objects, partially obstructed by other obstacles or similar objects. Both stationary or in motion being day or night. In general, this is not a toy, Speslab has developed and trained a professional neural network for real surveillance. But, of course you need to understand that it does not work with magic, so in any case, the object should be at least somehow visible: at least 30% and somehow illuminated. If it is almost completely blocked by something or is in complete darkness, then sorry – a miracle will not happen.
Unfortunately, Russia itself is in such a geographical area, where it is often dark rather than light, and in winter it is very dark. And not every forest has a light bulb. Therefore, neural networks do not remove from its agenda and other algorithms SpecLab is looking for meaning in complete chaos, even if a person is not able to recognize objects. Videosemantic is here to help.
But before neural network, we were always trying to look for meanings everywhere – regardless of whether they were people or cars in the frame. So now we have made a huge leap to reduce false video alarms. Moreover, we are talking about a jump to another level – depending, of course, on the object.
Here we analyze the whole pile of diverse movements of shadows from the branches of trees, but turn on the neural network – and in this blizzard of interference we can exactly see the zone of interest.
We are interested in these rectangles, which mark people and cars, as well as other objects – for specific tasks. And now we study the behavior of these contours, these objects, each of which received an individual number, so we can track them.
There are certain parameters, i.e. we regard them as separate participants in the behavioral plot. Yes, and we know at the same time what they are capable of – depending on whether it is a man, a machine or an animal.
Accordingly, you need to understand that the neural network is only part of the task, maybe even half of it, but without event logic it is meaningless. What, for example, to do, if a car was found in a frame? What then? And in the next ten thousand frames? Shout about it to the user? Yelling all the time until the car leaves? Having reported that in a zone of visibility the person came, not to analyze it anymore? Still, it is necessary to understand that the neural network is a powerful tool, but which only tells our program who or what it is dealing with – nothing more. And then takes effect behavioral logic – videosemantic. Which Speslab has been improving for a long time and which even without a neural network successfully coped with its duties. Now with the neural network – we did not think that we would ever have to say this – videosemantic has become almost perfect, because it acquires powerful protection from false interference.
By the way, neural network itself has been invented for a long time, but finally, there are capacities of the household price range that are affordable. In fact, this requires a good video card in the computer, and of course – a good program, without the intelligence of programmers nothing could be done. SpesLab for 22 years, as you know, is ahead of the surveillance industry and at present has also a fully ready-made version of its own neural network for pattern recognition. It is free for 4 cameras. Or pay for more channels. The future is already here!
And this is what we have. From the chaos of desperate movements we get a great event-based surveillance, in which there are no interferences. Well, sorry, drifted interferences will still have, but their quantity will be lower than 1%. We have brought surveillance to a qualitatively new level: instead of 90% on the streets and 20% in premises, we now have less than one percent interference everywhere. Do you have any idea how many lives we could save? But this is not a joke, the attention people spend on their monitors in the 21st century is only comparable to sleep. By the way, about sleep, now you can sleep peacefully at the monitors, because when there is any kind of human or car activity instead of interference, the computer will tell you about it.
It would be convenient for you to get a reliable alarm on your smartphone – now you will not swear that a swinging branch or a running dog distracted you from your sleep.
Now you only have useful events with only people and cars.
How to determine their usefulness -we'll talk about this next time. And videosemantic-technology will help us!