Identification of Collision Situations for Higher Efficiency of Traffic Control System

An integral part of modern cities within Smart City concepts is the development and related innovations in traffic control systems. New proposals need to be carried out in accordance with the applicable legislation and at the same time on a sufficient data base. This paper reflects the first conclusions of the research project SENDER, the aim of which is to develop, using deep learning methods, such a system that will warn drivers of impending danger in front of selected traffic intersections based on recognized data in the image from installed cameras. The paper mainly describes the process of selecting traffic situations in the area of intersections, which it is appropriate to warn the driver about. As part of the research, a state-of-the-art analysis was first carried out, which summarizes knowledge from current traffic control systems and the possibility of identifying collision situations, then the expert team defined collision situations that may occur in the node, including accompanying prioritization. On the basis of this identification, a specific intersection in the city of Brno was selected, which was fitted with cameras, and a test was conducted to determine whether the defined collision situations at the intersection actually occur. For the purposes of the developed system, this results in the specification of preferred collision situations, and these will then be simulated in variants with minor changes in the input parameters, in order to create a large enough database for learning the neural network.