THE INFLUENCE OF CHARACTERISTIC PARAMETERS ON THE PERFORMANCE OF FATIGUE LANE DEPARTURE RECOGNITION MODEL

Abstract

The aim of this study is to identify suitable characteristic parameters for building recognition models that can effectively detect driver fatigue. The parameters are divided into two main groups driver activity and vehicle motion state and further categorized into ten specific parameter groups (DRM No. 1 to DRM No. 10) to construct the corresponding recognition models. Based on Gaussian Mixture - Hidden Markov Models (GM-HMM) theory, we establish recognition models for fatigue lane departure (FLD) and normal lane changing (NLC) states. Using the basic principles of HMM theory, we propose the use of a Left-Right mixture Gaussian chain structure to build the recognition model. These ten groups of characteristic parameters are used as the observation chain for the GM-HMM model. After training, we obtain the characteristic parameters for the corresponding recognition models. Finally, the performance of these ten models is evaluated using three metrics: Accuracy, Sensitivity, and Specificity. The results demonstrate that the choice of parameters significantly impacts the performance of the recognition models. These findings underscore the importance of selecting appropriate feature parameters to improve driver fatigue detection systems, ultimately contributing to safer driving and the prevention of fatigue-related accidents.