Contribution to road safety through devices and applications based on artificial intelligence

Resumen

Currently, smart cities are becoming a reality thanks to the use of information and communication technologies, with transportation systems and road safety being examples of success. Several studies expose that road safety is a weak point in cities, showing that in Spain 12642 road accidents occurred in urban areas during 2018 or that 40% of road accidents occur when pedestrians cross at the right place. This is because pedestrians cross at designated crossings that are not always 100% visible due to different factors (e.g., poor maintenance of the roadway, obstruction of vertical signals or bad weather conditions). In these situations, the distance at which braking is initiated is a determining factor in the severity of the collision and its prevention. For this reason, this doctoral thesis presents different hardware and software solutions to help reduce road accidents, as well as a review of the state of the art of wireless communications used in the field of transport systems and road safety. The first solution proposed consists of an intelligent road signaling system capable of interacting with its environment, distinguishing between vehicles and pedestrians, as well as alerting drivers on the presence of pedestrians at a zebra crossing. To do this, fuzzy logic and sensory fusion were used with a set of various sensors as inputs. This system stands out for its autonomous power supply capacity, small size and easy installation on public roads without the need for civil works. Its functionality and viability have been tested in a real controlled environment, obtaining high performance and reliability. The second solution proposes an improvement on the first, which allows increasing versatility by generalizing vehicle detection using machine learning techniques instead of fuzzy logic. In order to determine which is the optimal technique for this problem, different approaches such as classifiers, anomaly detectors, prediction of time series and reinforced deep learning were used. For this, a data set was generated from samples collected in five different locations in Spain and Portugal under real conditions of fluid traffic. The computational models obtained after training and validation confirmed the possibility of replacing fuzzy logic with machine learning techniques. The third solution describes a mobile application that allows determining the crossing intention of a pedestrian over crosswalks and generating safe routes in cities. One of the novelties of the application lies in the ability to detect the crossing intention of users throughout the city and not only at specific points. The other functionality allows calculating and tracing safe routes through the city making use of pedestrian areas of interest (i.e., zebra crossings, pedestrian streets and elevated walkways). In this way, the road safety of the route is increased from the pedestrian's point of view. In addition, the application has the ability to dictate instructions on the route to users, as well as to include wireless communications to transmit the crossing intention of a pedestrian to the system developed in the first solution and alert drivers. Complementarily, the doctoral thesis makes a review of the state of the art to identify who, when and what is being investigated, placing the focus on vehicle-to-all, infrastructure-to-all and pedestrian-to-all wireless communications. In addition, the review establishes a taxonomy that aims to reduce the ambiguity of acronyms around the communications between vehicles, infrastructures and pedestrians, as well as to determine which are the future technologies that will give rise to novel applications.