Operator performance and safety that are both affected by the operator mental status (fatigue/alert) are basic requirements in work environments. The needs for practical and low-cost approaches for fatigue detection are therefore required by governmental, industrial and safety organizations. This paper proposes a new approach for operator fatigue detection that is based on biological data collection using accurate, low-cost and easy to use wearable devices. Three bio-data sensors for heart rate, wrist temperature and skin conductivity are adopted in this work for data collection and generation of fatigue-related metrics. Effective features of the collected bio-data are identified and labeled using the heart-rate variability metric that is measured by a wearable chest-strap heart monitor. The data collected from real subjects is used to train a dataset for fatigue analysis and classification using sub-classifiers based on artificial neural networks. Decision-level data fusion technique based on Bayesian combiner is then applied to enhance the accuracy and confidence of the obtained classification results. Performance of the developed alertness/fatigue detector is assessed experimentally and the obtained findings demonstrated acceptable performance in terms of modularity, accuracy, sensitivity and specificity when compared to individual classifiers.
If the inline PDF is not rendering correctly, you can download the PDF file here.