An fNIRS labeling image feature-based customized driving fatigue detection method
Lingqiu Zeng,Kun Zhou,Qingwen Han,Yang Wang,Gang Guo,Lei Ye
DOI: https://doi.org/10.1007/s12652-022-04325-7
IF: 3.662
2022-07-25
Journal of Ambient Intelligence and Humanized Computing
Abstract:The practicality of driving fatigue detection approaches largely depends on social acceptance. Physiological-based methods perform well, but they are rarely accepted by drivers, while subjective evaluations cannot support real-time detection. However, image-based approaches with relatively high take-up rates, ace detection accuracy problems, thus causing driver dissatisfaction. To satisfy the demand of logistic companies, in this paper, a customized driving fatigue detection method that integrates a subjective evaluation, physiological features and image features is proposed. As a result, a real-time driving fatigue detection system, RefreshingDrive, is designed. First, the relationships between questionnaire results, near-infrared spectroscopy indices, and oxyhemoglobin (HBO) and deoxyhemoglobin (HHB) values are investigated to express the individual differences among fatigue thresholds. Then, a long short-term memory network is constructed to find an individual-oriented HBO+HHB threshold, which is used to divide an image dataset into two subdatasets, namely, a fatigue dataset and a nonfatigue dataset. After that, a facial feature-based fatigue detection method, in which a multitasking cascaded eye convolution network is proposed to extract eye features while a 3D ConvNet is employed to recognize fatigue actions, is presented. Moreover, a fatigue evaluation rule, the Action-Blink Fusion Rule (A-B fusion), is proposed to recognize driving fatigue. Experimental results show that the proposed method achieves a fatigue recognition accuracy of 95.3%. Finally, an on-road test is performed to verify the effectiveness of the proposed approach.
computer science, information systems,telecommunications, artificial intelligence