Artificial neural networks are at the core of our products
Low resolution-friendly
When you work with the video that comes from a low-resolution camera, the resolution of individual objects is very low. And if you have a high-res camera, but the objects are too far away - still, the object’s resolution is low.
Thanks to our ‘XLR-technology’ the detector can safely recognize a vehicle or a person even if it consists of just a few pixels. The technology is so powerful that sometimes the system recognizes the vehicles even better that an average human does.
Thanks to our ‘XLR-technology’ the detector can safely recognize a vehicle or a person even if it consists of just a few pixels. The technology is so powerful that sometimes the system recognizes the vehicles even better that an average human does.
Support a wide range of vantage points
We create our neural recognition modules with the idea that often our clients will not be able to place their cameras into aт optimal position. So, our system needs to work with the video captured at bad angles.
And we’re glad to report that our engineers managed to create the recognition algorithms that can capture vehicles from almost all imaginable angles.
And we’re glad to report that our engineers managed to create the recognition algorithms that can capture vehicles from almost all imaginable angles.
Keep working in low-light conditions
Running recognition at night is a challenging task. When the source video quality is not enough for the analysis the system turns on a cascade of image enhancement pre-processing steps that digitally improve contrast, reduce noise, de-blur objects. This technology works completely automatically and doesn’t require any manual tuning. As a result, the system recognize the objects that are just barely visible.
Recognize partly visible objects
Working outdoors in real-world environment means that we face all kind of obstacles. For example, one object may be blocked by another object. If the traffic is tight, we often can see only small parts of vehicles, overlapping each other. Our system is ready for such scenario: it needs just a fraction of a complete image to recognize the vehicle and mark it as a separate object.
We apply the same concept of maximum possible robustness to all of our recognition modules. If you see a half of letter ‘A’ on the number plate, the system still recognizes it as ‘A’ with high confidence.
We apply the same concept of maximum possible robustness to all of our recognition modules. If you see a half of letter ‘A’ on the number plate, the system still recognizes it as ‘A’ with high confidence.
Do measurements in a calibrated scene
You can measure objects’ dimensions and the speed in meters and km/h. To do this you need to set a number of points in the scene and provide real distances between them. The algorithm will do the rest and instead of pixels and pixels/sec you’ll see meters and km/h.
The system is smart enough to take into account the frame’s perspective. A car’s dimensions remain the same regardless of its position, whether it’s in the foreground or far away in the background.
The system is smart enough to take into account the frame’s perspective. A car’s dimensions remain the same regardless of its position, whether it’s in the foreground or far away in the background.
High processing speed matters
The recognition accuracy depends on the complexity of the neural network. Generally, the bigger network – the better performance it demonstrates. But increasing the complexity infinitely in order improve recognition quality is not the best approach. The larger neural model becomes, the slower it runs.
We developed complex neural recognition architectures that consist of millions of neurons. And in our products, we run them in combination with our ‘RTTFF-technology’ that allows you to enjoy high recognition accuracy and to process multiple ‘live’ video streams in real-time at the same time.
In addition to that, a number of our products allow you to select the neural network complexity manually. Instead of a default recognition model, you may want to choose a light-weight, very fast version that can process video streams in real-time even on a tiny edge device.
We developed complex neural recognition architectures that consist of millions of neurons. And in our products, we run them in combination with our ‘RTTFF-technology’ that allows you to enjoy high recognition accuracy and to process multiple ‘live’ video streams in real-time at the same time.
In addition to that, a number of our products allow you to select the neural network complexity manually. Instead of a default recognition model, you may want to choose a light-weight, very fast version that can process video streams in real-time even on a tiny edge device.
Get full control
We create our products to be as transparent as possible. We let the users see all the details of video recognition process through interactive visual interfaces. And we provide extensive controls over those processes. For example, you can adjust recognition sensitivity and immediately see the effect.
With the complete transparency of the recognition pipelines, you can fine-tune the parameters and achieve the best possible performance even at the complex scene.
With the complete transparency of the recognition pipelines, you can fine-tune the parameters and achieve the best possible performance even at the complex scene.
It has tons of technology inside.
We'd be glad to share more with you. Just drop us a message...