IEEE Access (Jan 2023)
Classification and Prediction of Driver’s Mental Workload Based on Long Time Sequences and Multiple Physiological Factors
Abstract
The driver’s mental workload is closely related to driving safety. However, how to analyze the driver’s mental workload in a reasonable and correct manner remains an open question. As an important factor to evaluate mental workload, changes in physiology encounter two clear problems: (1) Physiological factor contains multi-characteristic indicators, there is a lack of reasonable means for synchronizing multi-dimensional tabular data, and the limits of tabular data processing in the evaluation of mental workload have a significant impact on the evaluation results. (2) The physiological data obtained during the driving process are of the time-series variety. The correlation of numerous indicators must be considered in time-series data correlation analysis. Mental workload should be the result of multiple indicators interacting over time, rather than a single instant. In this regard, we propose a model, that is the long time sequences and multiple physiological factors(LTS-MPF), for classifying and predicting multiple physiological changes in the time series. In contrast to previous methods of processing data in a single instant, LTS-MPF can directly analyze all time-series factors that may affect the driver’s mental workload during a time interval, such as Heart rate growth, Heart rate variability, and Electrodermal activity, and so on. Furthermore, LTS-MPF can predict the driver’s mental workload in the next 1s as well as classify the current sequence’s results. Specifically, we collect physiological data from drivers via sensors. These collected data are processed and transformed into tabular data. The table’s columns represent features, while the rows represent all feature data at one moment in time. The row order also indicates the forward and backward order of the different moments. We convert each row in this table into an embedding feature and feed the entire table into our proposed LTS-MPF based on the Transformer model. The LTS-MPF achieves time series correlation while eliminating column feature series irrelevance. The experiment results reveal that LTS-MPF exceeds earlier techniques in forecasting the driver’s mental workload, with an accuracy of up to 94.3%. And its accuracy in predicting mental workload in the future for one second can reach 93.5%. These findings suggest that LTS-MPF can be utilized to not only better evaluate a driver’s mental workload in the present, but also in the future, providing solid data for early warning of dangerous driving behaviors and enhancing driving safety.
Keywords
