Behavior Based Secondary Task Action Detection In Driver Monitoring Systems

Abstract

Driver distraction is one of the leading causes of road accidents and fatalities in traffic, both for novice and experienced drivers. Due to this, legislation has started to pinpoint the development and usage of systems to detect and prevent this kind of behavior known as secondary tasks in the context of driving. Some secondary tasks are particularly dangerous, such as texting. For the car system to effectively be able to assist the driver in reducing such behavior, driver monitoring systems are being researched and developed. While there are many different approaches to monitoring a humans behavior, the most common one is to use cameras that feeds the video stream into machine learning models trained to recognize and identify different behaviors. The scope of this thesis covers the steps of defining phone usage in the context of driving, collecting data in a simulator, preprocessing the data and training machine learning models to be able to predict the behavior of the driver. The research questions concerns the challenges in predicting human behavior, which signals are most important in doing so and how it is possible to model the dimension of time. The framework for the implementation of the project is a hybrid approach using the double diamond structure in combination with Human-Centred AI principles and a classical machine learning workflow.