Cognitive Data Fusion

Cognitive Data Fusion is technology for use in model of computer vision of the combined data arriving from different sensors on computing block.

On April 6, 2017 the Cognitive Technologies company announced creation of Cognitive Data Fusion technology.

In solutions of many foreign developers of the systems of autonomous driving data processing is, as a rule, conducted by the principle of on-chip - directly on chips which are established on different sensors that complicates complex use and information processing, arriving from all sensors. It would be equivalent to if the sight and hearing at the person were not synchronized and did not supplement each other. Provide that operate the vehicle at the same time blind person who hears also able to see, but the deaf. And still at the same time they among themselves do not communicate. Yury Minkin, Head of Department of development of Cognitive Technologies unmanned vehicles

Data view about a road scene in Cognitive Data Fusion, (2017)

As a rule, this received from the radar allow to define exact distance to an object and its speed, but it is only a part of information on its type and arrangement. The object type, its sizes, other parameters is also important for making decision on danger of a situation. The radar can recognize rather precisely the machine and "begin to doubt" in recognition of the pedestrian. Also the radar, in fact, does not allow to define separate elements of a road scene, such as road signs, a marking, etc. Information obtained from the video camera, in turn, gives rather exact coordinates of an object, idea of its type, and a relative positioning concerning other objects. And also allows to detect a bigger class of objects of a road scene (traffic light signals, types of signs, a road marking, etc.)^[1].

Data integration, received from different devices, allows to fill information lacking for understanding of the operating road scene. For example, at the level of recognition of the image of the object received from the video camera of 80% given, received from the radar can add missing information and raise detection accuracy to 99% and more percent. Complex use of data also allows to integrate all information on speed, coordinates, distance to an object, its type, a relative positioning, existence of other objects in close proximity to it and their physical characteristics.

Problem of deep data integration, received from different sensors quite difficult. The operation principle of Cognitive Data Fusion technology is similar to how the brain of the person obtaining data from different sense organs functions: sight, hearing, tactile bodies, etc. at the same time. For this purpose information removed from each of sensors is synchronized and provided to a single system of coordinates. Then, "crude" data arrive in the calculator where they are in a complex processed mutually "enriching" each other.

This approach allows to implement compensatory function - when one of sense organs of the person refuses or performs the functions not in full, the activity of others amplifies. The same way the architecture of Cognitive Data Fusion provides a possibility of obtaining more detailed data on a road scene from other sensors in such cases. For example, if the signal from the radar detects an obstacle in a way of the car, and the video camera for any reasons accurately will not recognize it, the artificial intelligence will assess this situation as it problem and will request more detailed information from the video camera.

The Cognitive data fusion technology is already approved on thousands kilometers of the Russian roads. We are sure that its implementation will allow to reduce the level of accident rate of the unmanned vehicle by 20%-25%. Olga Uskova, president of Cognitive Technologies

Robotics

Notes