Human Activity Recognition in a Car with Embedded Devices

Detection and prediction of drowsiness is key for the implementation of intelligent vehicles aimed to prevent highway crashes. There are several approaches for such solution. In this paper the computer vision approach will be analysed, where embedded devices (e.g. video cameras) are used along with machine learning and pattern recognition techniques for implementing suitable solutions for detecting driver fatigue. Most of the research in computer vision systems focused on the analysis of blinks, this is a notable solution when it is combined with additional patterns like yawing or head motion for the recognition of drowsiness. The first step in this approach is the face recognition, where AdaBoost algorithm shows accurate results for the feature extraction, whereas regarding the detection of drowsiness the data-driven classifiers such as Support Vector Machine(SVM) yields remarkable results. One underlying component for implementing a computer vision technology for detection of drowsiness is a database of spontaneous images from the Facial Action Coding System (FACS), where the classifier can be trained accordingly. This paper introduces a straightforward prototype for detection of drowsiness, where the Viola-Jones method is used for face recognition and cascade classifier is used for the detection of a contiguous sequence of eyes closed, which are considered as drowsiness