Sumarização e análise de atividades práticas de direção para detecção e avaliação de estresse e ansiedade de motoristas

Abstract

This doctoral thesis presents an approach for detecting and analyzing the emotional states of drivers’ stress/anxiety from data collected during practical driving activities. The proposed method aims to identify these emotional states and streamline the necessary support to be provided to drivers. For this, the method combines techniques of prioritization and summarization of videos recorded during practical activities of direction. With the summaries produced, professionals in the field of psychology can identify and understand the behavior of the driver with stress/anxiety more quickly, without having to watch the videos in their entirety, let alone being present during the activities. This is because, with the prioritization technique, a list of practical driving activities is generated, ordered by the driver’s stress/anxiety level, enabling individual measures to be proposed to help drivers deal with or correct these states. The prioritization technique contributes to the summarization approach, as it speeds up the service to drivers who need this support the most. Both of the proposed techniques use data from Facial Expressions (EF’s), heart rate and driver movements — these characteristics have been selected because they have already been successfully explored in related works. Most related works focus on the automatic classification of stress or anxiety situations in drivers, therefore, the approach proposed in this thesis differs by presenting techniques to summarize and prioritize practical driving activities. The proposed summarization technique also uses GPS coordinates, allowing to identify the place where the behaviors were observed and, with this, to verify triggers for stress/anxiety in each driver. Heart rate was measured in terms of heartbeats per minute. As for EF’s, a set of 18 Action Units recognized with the Open Face 2.0 tool. As movements, it was considered driver’s habits that can occur in stress/anxiety situations: pressing, biting and licking his lips, biting his nails, and rubbing his face and/or hair. The proposed prioritization technique combines a classic ordering algorithm with a comparative function based on an Artificial Neural Network. The comparative function receives a list of parameters extracted from two driving activities and identifies in which one the driver had the highest level of stress/anxiety. As for summarization, the proposed approach is supported by a tool developed to allow the visualization of each of the management activities in three different perspectives, explored to facilitate and organize the analysis of the data: i) videos; ii) reports; and iii) summaries. In addition to the prioritization and summarization techniques, this thesis also presents a method for automatic detection of drivers’ hand movements, associated with stress/anxiety. The proposed method for detecting these gestures combines mathematical morphology with automatic detection of the driver’s face. To assess the proposed approach, data were collected in 60 practical driving activities for students at a driving school. The events of interest were manually labeled in the database, which was also prioritized, in relation to the level of stress/anxiety of the drivers, by a psychologist specialized in assisting drivers. The result of the prioritization method was assessed for cohesion and similarity with the reference set that was prioritized manually by the specialist. The video summarization support tool was qualitatively evaluated by professionals linked to the driving school in which the database was produced. The results of the proposed method for automatic movement detection were evaluated for accuracy, recall, precision, and F1-score, indicating that support for drivers can be streamlined, especially for those with a high level of stress/anxiety. It should be noted that the proposed approach is particularly relevant in the context of driving schools with a large flow of students, in which it is difficult for a specialist in Psychology to analyze, in a timely manner, all the data of the direction activities performed. In this context, this thesis contributes directly to improving the training of drivers and, consequently, to traffic safety.