Ajinkya Prabhune (Professor of Applied Computer Science and Data Analytics, on the left) and SRH alumna Ojashree Bhandare win Best Paper Presentation Award at IEEE Big Data & ML conference.
Maria Schmidt, Ojashree Bhandare (Alumna), Ajinkya Prabhune, Wolfgang Minker and Steffen Werner
Paper Title: Classifying Cognitive Load for a Proactive In-Car Voice Assistant
Cognitive load is the used amount of information a person can hold in his or her memory at a given time. High cognitive loads can occur for drivers as a result of fatigue, intoxication and secondary tasks, or even just due to daydreaming or getting lost in thought. These reasons hinder drivers from keeping their focus on the road. Truck or lorry drivers who spend many hours behind the wheel are particularly susceptible to high mental stress (i.e. cognitive load), which often leads to accidents or collisions. Therefore, our research aimed to analyse driver behaviour and then proactively assist drivers during driving.
We addressed this challenge in a paper entitled “Classifying Cognitive Load for a Proactive In-Car Voice Assistant” in which we designed, implemented and deployed an embedded Machine Learning (ML) workflow in vehicles to proactively assist drivers faced with high cognitive load. The assistance is provided through speech, because it is intuitive and does not distract the driver. The ML workflow carried out for this paper followed a two-step approach. In the first step, unsupervised learning techniques were applied to cluster the data in four cognitive load clusters. To identify the clusters in the initial stage, we used four signals from real-time driving data recorded by an electronic control unit via a CAN bus. From these four signals (braking velocity, steering wheel angle, steering wheel acceleration and vehicle speed), we generated 36 features by applying various feature engineering techniques. Out of these 36 features, we then selected 17 features to train our ML models. Finally, our ML workflow was tested and achieved 96.97% accuracy. In other words, our proactive assistance system was able to accurately classify the drivers’ cognitive load – and assist when required.