Russia Discovering a way to identify light weapons in a crowd of people using artificial intelligence

Russian specialists have developed a machine vision subsystem that recognizes small parts located at different distances from the camera.

Specialists from the Perm National Research Technical University were able to increase the accuracy of recognition of images captured by a neural network by 25%, which will allow video surveillance systems to detect dangerous objects in a crowd of people. According to the university's press service, the researchers have developed a computer vision subsystem. It recognizes small parts located at different distances from the camera.

“The 25% increase in accuracy occurs in individual test images due to artificial constraints imposed on class assignment and object localization in the context of the processed image scene,” said Andrei Kokolin, assistant professor at the Department of Automation and Telemechanics at Perm University.

Unlike classical neural networks, the proposed solution allows the system to highlight a region of interest on the screen and search in it for specific objects whose dimensions are determined taking into account the distance between the camera and the object. At the same time, changes in shooting conditions do not affect the result of content analysis.

A statement issued by the university’s press service said: “The proposed image processing scheme will help detect weapons or dangerous objects in a crowd of people when people are at different distances from the cameras. A classical neural network may not distinguish between weapons carried in distant or very close locations.” "But if you first detect the silhouettes of all the people on the photo, weapon recognition will be more accurate."