This study is part of a PhD research project, aiming to analyse driver behaviour and pinpoint crucial factors contributing to dangerous driving manoeuvres.
Our research problem is to predict dangerous driver manoeuvres within a given context. Dangerous driver manoeuvres can be defined as actions taken by a driver that significantly increase the risk of collision, injury, or harm to others on the road. The first stage to achieve this is to identify the dangerous manoeuvres by learning the driving behaviour through the telematics data. The driver state is of high importance and influences the driver’s behaviour. Thus, the next step will characterise the driver state within the context. Two main states will be the scope of the study: stress and distraction. Context-aware driver behaviour analysis is the goal of this study. Therefore, the subsequent step will be to extract and derive features representative of the surrounding driving context. We will use deep learning approach to predict dangerous driver manoeuvres by using data.
In order to build deep learning models, vast amounts of training data are required. In our experiment, we will collect data using data collection tools that include the following: 1) a vehicle OBD II diagnostic and tracking system, 2) a road-facing dashcam, 3) a driver-facing dashcam, and 4) a Fitbit Sense 2 wearable. Data collected from these tools will be used to train the deep learning models.