What are steering pictures are worth? Using image-based steering features to detect drowsiness on rural roads

Abstract

Drowsy driving is a persistent and significant problem on today’s roadways. Mitigation technologies may help resolve the problem, but their success depends on effective detection algorithms. Steering-based algorithms are a promising direction for such effective algorithms because of their low cost and ease of implementation. However, steering-based approaches are often limited by high false positive detection rates. The goal of this study was to assess if image-based steering features and convolutional neural networks can reduce these high false positive rates. The analysis investigated two methods for transforming steering wheel angle data into images, Markov Transition Field and recurrence plots. Area under the ROC curve and false positive rate results suggest that both approaches nominally improve detection performance and reduce false positives relative to a benchmark, with some evidence that recurrence plots have the highest performance.