Non-intrusive Detection of Driver Distraction using Machine Learning Algorithms

Abstract

Driver's distraction has become an important and growing safety concern with the increasing use of the so-called In-Vehicle Information Systems (IVIS), such as cell-phones, navigation systems, etc. A very promising way to overcome this problem is to detect driver's distraction and thus to adopt in-vehicle systems accordingly, in order to avoid or mitigate the negative effects. The purpose of this paper is to illustrate a method for the non-intrusive detection of visual distraction, based on the vehicle dynamic data; in particular, we present and compare two models, applying Artificial Neural Networks (ANN) and Support Vector Machines (SVM) which are well-known data-mining methods. Despite of what already done in literature, our method does not use eye-tracker data in the final classifier. With respect to other similar works, we regard distraction identification as a classification problem and, moreover, we extend the datasets, both in terms of data-points and of scenarios. Data for training the models were collected using a static driving simulator, with real human subjects performing a specific secondary task (SURT) while driving. Different training methods, model characteristics and features selection criteria have been compared. Potential applications of this research include the design of adaptive IVIS and of “smarter” Partially Autonomous Driving Assistance Systems (PADAS), as well as the evaluation of driver's distraction.