Korean scientists have taught the computer to recognize road signs in difficult conditions: at night, behind branches of trees and in bad weather.
The researchers at the University of Jones in South Korea named the "General Purpose Graphics Processing Unit (GPGPU)" for their development. The program can recognize objects with predefined characteristics - for example, road signs - with poor visibility. Using the traffic rules described in the Vienna Convention (implemented in Germany and South Korea), the developers tested the program and obtained convincing results.
To date, there are two fundamentally different systems for recognizing road signs for cars. One is used by the creators of the Advanced Driving Assistant System (ADAS) - assistant drivers who do not drive the vehicle, but give clues. The second (GPGPU) works in unmanned vehicles; She recognizes road signs, analyzing the video from the camera installed in the front of the car.
Modern software, on which the systems of recognition of signs work, is oriented either to color or to the shape of signs, searching in the surrounding world for the given patterns and color combinations characteristic for signs. The work of the color method is hampered by darkness, rain and snow, and the search for characteristic forms - visual obstacles, such as tree branches.
The software developed by Korean scientists uses deep learning technology. The computer compares the image of the road section with the reference images, and the neural network analyzes the shapes and interprets the symbols that are present in the image. According to the authors, the processing power required to operate the system is low, and the processing speed of information allows you to analyze the situation on the road while the car is moving.
