Improvement of a Conflict Prediction Model on Highways using Deep Learning

Artificial intelligence has 3 high-priority fields of study: language processing, speech, and machine vision. The growing need for evolution associated with process automation requires increasingly effective and innovative solutions with replication capabilities. Artificial vision particularly has the ability to be applied within a wide range of processes, finding its maximum utility in tasks of detection, identification and tracking of objects. The application of solutions based on this technology to traffic analysis proposes highly flexible processing tools with a virtually zero degree of intrusion with the nature of the process. Reverse engineering procedures and the application of specialized algorithms allow the development of systems for the study of a problem closely related to the development of cities. Obtaining values such as the flow of vehicles that circulate on a highway is an essential requirement for the planning or readjustment of road works. Thus, the importance of systems based on artificial vision that allow this type of analysis is denoted. In this paper, a vehicle tracking algorithm is developed to determine the trajectory described, tracking time and average speed of the same. In addition, a vehicle counting algorithm is developed to determine the flow of different types of vehicles captured in the study frames. Also, for validation of operability and repeatability of the developed algorithms are established. Implement a program based on OpenCV - C++ for the analysis of vehicular traffic in offline mode.