Korean scientists have created a system for tracking road traffic from two laptops
A combined team of Korean and American scientists created a traffic tracking system using only two laptops with active WiFi and neural network modules. The cost of such a system is much lower than the price of a standard network infrastructure for monitoring traffic, which includes cameras, radar, specialized software.
Accuracy of recognition of moving vehicles with a new system is quite high. Car type is determined with an accuracy of 91.1%. The difference between different types of cars and motorcycles, the system distinguishes with an accuracy of almost 100%.
The system requires two laptops, which are on opposite sides of the road and WiFi. True, the scientists installed two laptops with connected cameras, which made it possible to calculate the number of cars passing by and establish the type of vehicle.
The principle of operation of the system is the analysis of the changed radio signal, the characteristics of which are influenced by the moving vehicle. To determine the type of machine, developers used a convolutional neural network that was trained to recognize vehicle types for 120 hours. Data for training was collected already in the process, in the "field conditions" on the road.
The developers did not have a goal to learn to identify absolutely all types of vehicles. For a start, the neural network was taught to identify passenger cars, SUVs, pickups, trucks and motorcycles. As mentioned above, the accuracy of determining the passage on the road of various types of transport is very high - about 99.4%. The accuracy of determining the type of transport is already lower - from 83.3% to 99.7%. On average, the accuracy was 91.1%.
According to the authors of the project, the system can be trained in more complex tasks, after which it will be possible to use it, for example, by the transport police. The accuracy of the system is slightly lower than the standard infrastructure, but on remote sections of the road it is quite possible to use a neural network, this allows you to connect a large number of new sections to traffic monitoring.
The system is not suitable for fixing traffic offenders, but for determining the traffic density, the efficiency of using certain sections of the road is quite. “We believe that our system can be useful for evaluating the work of thousands of kilometers of distant roads from the center,” the project participants say. The price of the system, which includes the learning process of the neural network and the neural network itself is about $ 1000. The cost of the stationary observation point exceeds $ 35,000.