A new neural network has been developed in Russia that can recognize a person from 150m up in the air, even if only a part of the body is in the frame. The network processes the images on the drone itself in real-time.
This neural network, called Kaspersky Neural Networks, is developed by Kaspersky Lab. The specialists at the lab mention that the algorithm can also detect damage in case of emergencies, analyze data for geodetic, cartographic, and cadastral activities, assess the state of crops in the fields, and detect weeds, the company explained.
After detecting a part of the body, the network will mark the object in question and rate it on a scale of 100-points. The algorithm identifies the object by taking into consideration the points scored on the scale. However, the operator of the program gets to make the final decision. If the drone picks up the image of the entire body, it can also be able to detect if the person is walking, sitting, or lying down. As mentioned above, the images get processed on the drone, but it sends the ones with detected objects to the operators.
A mini-computer equipped with bus pilot is responsible for the operation. The processing results need to be rechecked by people on the ground if it runs on low-power equipment.
In the case of KNN, since this program has to use an industrial aircraft-type drone called the Albatross M5, it is going to be very expensive. This may cost several million rubles; on the other hand, another famous detection program called Lisa Alert which is very famous in Russia can equip a dozen commercial quadrocopters to essentially do the same job.
Kaspersky Lab recognizes EMERCOM of Russia as one of its main customers. EMERCOM has mentioned that the company is testing a neural network model of their own design which works to detect objects using remote sensing data of the Earth.