Driver cognitive distraction detection: Feature estimation and implementation

Abstract

This article focuses on monitoring a driver's cognitive impairment due to talking to passengers or on a mobile phone, daydreaming, or just thinking about other than driving-related matters. This paper describes an investigation of cognitive distraction, firstly, giving an overall idea of its effects on the driver and, secondly, discussing the practical implementation of an algorithm for detection of cognitive distraction using a support vector machine (SVM) classifier. The evaluation data have been gathered by recruiting 12 professional drivers to drive for approximately 45 min in various environments and inducing cognitive tasks, i.e. arithmetic calculations. According to the prior knowledge and the experimental analysis, gaze, head and lane-keeping variances over a 15 s time window were selected indicative features. The SVM classifier's performance was optimized through exhaustive parameter tuning. The executed tests show that the cognitive workload can be detected with approximately 65-80 per cent confidence despite the fact that the test material represented medium-difficulty cognitive tasks (i.e. the induced workload was not very high). Thus, it could be assumed that a more challenging cognitive task would yield better detection results.

Action Units and Their Cross-Correlations for Prediction of Cognitive Load during Driving

“…In [27] the eye movements are analyzed to predict visual inattention using Neural Networks. The gaze information was used along with head movements and lane position of the vehicle in [28] to detect induced visual and cognitive distraction using a stereo-vision system integrated in trucks and passenger cars. For cognitive distraction, the authors achieve 68% on the truck experiments and 86% for the passenger car experiments.…”

Real-Time Detection System of Driver Distraction Using Machine Learning

“…In our opinion, the most representative works are [1], [9], [46], [48], [41] and [24], since more strongly related to our research and they have been a source of inspiration for us.…”

“…In particular, [9] and [38] suggest that there are basically three approaches to such a recognition problem: monitoring driver's perception; monitoring driver's steering and lane keeping behavior; recognizing driver's involvement in a given secondary task. Despite the fact that different classification methods can be found in literature to detect distraction or inattention while driving, nevertheless, since the mental state of the driver is not directly observable, no simple measure can weight distraction precisely and thereby all traditional methods show some limits [41]. In this context, the predominant approach is to use ML techniques, which seem to be much more appropriated for this type of classification problem.…”

“…Another promising approach can be found in [41], where SVMs are used to detect driver distraction based on data captured under real traffic conditions, resulting in accuracies of 65%-80%. Features are thereby computed from fixed-length time windows, i.e., the amount of context that is incorporated into the classification decision, is predefined.…”

Online Driver Distraction Detection Using Long Short-Term Memory

“…The predominant approach is to use static classifiers such as Support Vector Machines (SVM) [13], [18]. A promising approach can be found in [19] were SVM are used to detect driver distraction based on data captured in real traffic conditions, resulting in accuracies of 65 -80 %. Features are thereby computed from fixed-length time windows, i. e. the amount of context that is incorporated into the classification decision is predefined.…”