Dynamic set point model for driver alert state using digital image processing

Abstract

The driver fatigue and lose of attention while driving are the most important causes of traffic accidents. Each year more than one million of deaths occur due to these facts. Thus, this problem has been converted into a serious social issue with high impact not only in economic terms, but also in the public health sector all around the world. Several approaches based on computer vision systems have been proposed to deal with this severe situation, but none of them have fully considered the non-fatigue state as a primary knowledge to detect an unusual event of a person while driving. In fact, typical approaches to deal with the problem of fatigue detection, are based on the analysis of behavioral features extracted with digital image processing such as frequency of blinking, yawning, among others. However, the huge limitation is the short interval of time between each analysis, that generally is few frames per second. Furthermore, all available methods are focus in modeling the fatigue, instead of representing the set point alert state of the driver, which is the main core of the proposed strategy. Hence, in this paper a dynamic set point model for alert state while driving using digital image processing and machine learning techniques is presented. The approach uses an embedded system build with a Raspberry prototyping board and a USB HD camera. Raspbian operative system controls OPEN CV libraries written in Python to detect face parts with an algorithm running Harr descriptors. The features extracted were the position and orientation of the head throw several minutes. Then, a mixture of Gaussians model with its learning and updating stages is used to represent the behaviour of features. Also, a dataset was built considering professional and non-professional drivers under two main scenarios: real and simulated conditions. Experimental results show the viability of the method for posterior analysis of unusual events while driving like fatigue detection, cellphone call or chat detection, or any other distraction not related to the driving process.