Tradition Group Ltd. presents URSA™ — its latest AI technical vision system.

URSA™ — is a project dedicated to development of a unified recognition and surveillance analytical system. The system has been initially developed with condition-trainable functionality and ability to classify the surrounding objects and environment, which allows users to receive reliable and timely information in required format about various physical objects recognized from the unified database of recognition results.

The system can be trained for use in the following areas:

• Traffic Management – to control, search, & track vehicles and traffic management at highways, motorways or control posts;
• Similarly, it can be used for railways, airways, sea and river transportation;
• Premises Security – manage, control housing complexes, buildings, manufacturing offices, or any other accommodation;
• Anti-terrorism – Locate People, left-over objects in public places, like the subway, railway stations, airports and any other public place;
• Operators of Internet search services for multimedia streams processing;
• Intellectual Soldier – security forces can utilize URSA in several aspects.
• Unmanned spies – intellectual recognition algorithms provide extensive possibilities.

Technology

Today most of the image recognition software applies object detection algorithms in descending order (running higher level processing steps first). Such strategy has been popular during the past decade, and yet an exclusive use of this «top-down» detection technique involves high computing complexity, thus making such recognition systems unsuitable for solving recognition tasks in real time for a mass customer without access to supercomputing resources.

In addition, the image processing local methods themselves have the possibility to «communicate» with each other and make joint decisions regarding the type of object to be recognized in case neither of the independent detection techniques can provide the accuracy specified. Such method allows increasing the range of recognizable objects, as well as the overall recognition accuracy.

Implementation