Systems based on artificial intelligence and neural networks can be adapted to perform a variety of tasks from complex mathematical calculations to object recognition and even video editing. A group of Russian scientists recently developed a special device that will allow blind people to re-orient themselves in space. The device was named SOMSI – a system of multi-channel information processing.
Responsible for the development of the company “Business Bureau” and, as reported by the publication “Izvestia”, is a gadget for analyzing the surrounding space using the resulting video and sound signals. The device consists of two parts: the first is all the necessary sensors for recording information like sensors, microphones and video cameras, and the second is the platform itself based on AI.
The principle of operation is as follows: having received the necessary data about the environment, the system transmits them to the cloud, where the neural network analyzes the information received, and then the result is sent back to the user, where it is converted into a voice or tactile message, or a virtual image (albeit in a rather primitive form). The latter option is available with an eye implant.
“The SOMSI system will work with modern bionic implants – this is a feature of the device. I really hope that all our plans will be realized when the product goes into series. ”- said Denis Kuleshov, one of the project leaders.
Today, the SOMSI system is a digital analogue of the retina, but there is not enough power and technology to process the image in all its details. At the same time, it is quite possible to recognize the large schematic contours of objects to the gadget. SOMSI has one more function: determining the distance to objects using ultrasound.
The user will need to clamp only one digging and the neural network will tell him where and what object is located. At the moment, there is an improvement in the performance of the device, and it is planned to enter the market in the second half of 2019. According to project engineer Vyacheslav Cheprakov,
“Our neural network is a very complex structure that requires long and detailed training. To recognize a single object, you need about 10,000 of its images in different angles. We have been training our network for about a year. For now, it recognizes basic things, such as a person, a car, or a phone, but does not yet distinguish between different faces or car brands. ”
The developers’ plans also include teaching the neural network to recognize the faces of specific people and transfer part of the program to the device itself so that it can work even in the absence of the Internet.