Feasibility study of the use of artificial neural network models in the automation of vehicle count and classification of public transport

Abstract

The analysis of vehicular traffic has become important in recent years due to the considerable increase of vehicles on the roads, which often is translated in traffic congestion. So, it is necessary to know the vehicle capacity of the roads, in order to take decisions to improve road traffic such as opening alternate roads, changing the timing of traffic lights, improving road signs, and so on. There are studies that perform the counting and identification of vehicles manually by observation, however, this technique can be inefficient since it must have people who perform this count on one or more roads during a certain period of time, leaving out information of several hours of the day. On the other hand, there are techniques that can be complex to configure or that interfere in the normal flow of traffic such as ramps with sensors and inductive loops. Therefore, there is a need to use a technique that allows the counting and identification of vehicles in a more efficient way without interfering in the flow of traffic. This study proposes the use of artificial neural network models in the automation of vehicle counting and classification. An artificial neural network model was generated that, through video, allows the detection and classification of public transport vehicles that travel on the roads with an accuracy of 94.29%. In addition, the model was tested in real time using an app and a smartphone camera, demonstrating that the model can be used if video is available without the need for calibration or system configuration. Therefore, based on the high level of certainty obtained and the tests carried out in real time, the evidence demonstrate that it is feasible to create systems that do not require calibration or complex configurations based on the use of artificial neural networks for the automation of counting and classifying vehicles