It determines the type of the vehicle with an accuracy of 91.1 percent.
American and Korean engineers have created a system for tracking traffic on the basis of two laptops with W-Fi and a neural network algorithm. Due to the change in signal level between the transmitter and the receiver algorithm is able to detect the passing car or motorcycle with an accuracy of 99.4 per cent and to determine its type with an accuracy of 91,1 percent, informs Rus.Media.
Tracking traffic on the roads, usually composed of cameras, radars and other tools to track the movement of vehicles, their speed, type and other parameters. This data comes in a single center, where experts or algorithms to analyze them. As a result of analysis it is possible to calculate the efficiency of the use of existing roads, and to determine the regions where necessary new roads. For a complete picture of the distribution of traffic should be equipped with tracking systems as many road sections, but it is very expensive when using existing equipment.
Engineers under the guidance of Kyung-Joon Park, Institute of science and technology Daegu-Kensuke showed that this problem can be solved with comparable accuracy, but using cheaper equipment. Created by the developers of the prototype consists of two laptops that are installed on different sides of the road and are used as receiver and transmitter Wi-Fi. Furthermore, along with these laptops was installed two laptops with attached cameras – they were used for video recording, from which then the developers have calculated the actual number and types of cars that drove past.
The system operation principle is based on the fact that during the passage of the vehicle between the receiver and transmitter, which exchange signals, signal characteristics change, and specific to each car. Through analysis of these changes can be understood as the fact of travel of the vehicle and its type. For this the developers used a neural network trained on data that the authors collected for 120 hours. The signal is pre-processed – turned it into an image in which rows correspond to the amplitude and phase of signals from different receivers (one notebook).
Engineers have taught a neural network to identify the passage of a car or motorcycle, as well as to determine its type. The developers have chosen five of the most popular types of vehicle classification: passenger car, SUV, pickup, truck and motorcycle. Testing of the system showed that it is able to detect passage of the vehicle with an accuracy of 99.4 per cent. When determining the type of vehicle precision below – from 83.3 percent for pick-up to 99.7 percent for the truck, the average accuracy amounted to 91.1 percent. Engineers found that the classification accuracy increases slightly training a separate neural network models for each lane of the road.