Academic Journal of Science and Technology ISSN: 2771-3032 | Vol. 15, No. 2, 2025 63 Analysis of Driver Fatigue Eye Movement Characteristics Under Intelligent Supervision Environment Xianghong Li School of Energy Science and Engineering, Henan Polytechnic University, Jiaozuo, 454003, China Abstract: Reducing drivers' improper driving behavior in transportation is an important condition for ensuring safety. In today's intelligent era, many passenger and freight companies are gradually using intelligent monitoring platforms to supervise the behavior and condition of drivers and vehicles. The intelligent supervision platform is beneficial for fatigue monitoring and improving driving safety. However, the research found that the number of alarms on the platform is too high, and the smart regulatory environment exhibits different operational characteristics compared to traditional driving environments. Therefore, based on the characteristics of driving behavior in the context of smart regulatory information interaction. It is very necessary to explore the influencing factors and development patterns of driving fatigue under the intelligent supervision environment. This article uses principal component analysis to identify sensitive indicators of driving fatigue eye movement characteristics in the context of intelligent supervision platforms, with the sensitive indicators being pupil diameter and blink frequency. This study helps to reveal the formation and variation rules of driver fatigue under the environmental conditions of intelligent supervision platforms, improving the accuracy and reliability of fatigue detection while driving. Keywords: Intelligent monitoring environment; driving fatigue; eye movement characteristics; principal component analysis. 1. Introduction During the driving process, the driver needs to maintain a high level of concentration throughout, gathering and processing information both inside and outside the vehicle. This high-stress operating state keeps the driver's central nervous system in a state of high activation. However, long periods of operation can lead to driver fatigue, and as the state of fatigue changes, it can also cause changes in physiological data and driving performance. Due to the monotonous environment during driving, it is easy for the driver to experience visual fatigue. In order to intuitively and effectively reflect the state of visual fatigue, eye movement data has become an important indicator. Therefore, this article conducts research on the eye movement patterns of drivers when fatigued in an intelligent supervision environment to promote fatigue management for drivers under such conditions. 2. Research Review In recent years, an increasing number of researchers have used methods such as eye tracking and facial recognition to make objective judgments about the fatigue state of drivers. Currently, researchers mainly use CNN algorithms [1-5], MTCNN algorithms [6-7], SSD network algorithms [8-9], HOG algorithms [9-10], and ASM algorithms [11-14] for face detection. Wang Di[15] used an Adaboost cascade classifier for face detection and eye region segmentation, performing image preprocessing to reduce environmental interference, accurately locating the eyes through image processing and extracting eye feature parameters. Huang Z[16] extracted the blinking frequency, yawning frequency, and nodding frequency of drivers to create a fatigue detection sample library, and used the Naive Bayes algorithm to establish a model for classifying and judging the fatigue state of drivers. Wukun Liang [17] conducts a comprehensive assessment during driving by using the BFR (Blinking Frequency Ratio) algorithm and head pose algorithm: if the number of blinks exceeds the BFR threshold, posture collection is initiated, and a pre-trained head pose model is used for evaluation; if fatigue is detected, it alerts the driver. Dong BT[18] uses a single-shot scale-invariant face detector (S3FD) to detect faces in images, and then utilizes a face alignment network (FAN) to extract facial features. Wang XY[19] and others used the Dlib library to extract the coordinates of facial feature points, obtain the feature parameters of the driver's eyes and mouth, calculate the Euler angle parameters of head posture, and established a fatigue recognition model through an improved Multi-task Cascade Convolutional Neural Network (MTCNN). Xiao WC et al[20]. aimed to maximize the utilization of driver's facial feature information and temporal characteristics by establishing a Spatial Pyramid Fusion Multi-Scale Feature Output (SPP-MSFO) detection model, obtaining images of the facial region and calculating the aspect ratios of the left eye, right eye, and mouth based on the coordinates of key points, thereby forming a fatigue parameter matrix. From the review of domestic and international research on driving fatigue, it can be seen that current studies on driving fatigue cover multiple aspects such as fatigue evaluation methods and fatigue evaluation indicators. These findings provide a foundation for the research presented in this paper. However, there are still some issues that need further research, as outlined below: Currently, there is a lot of literature on fatigue driving analysis both domestically and internationally. However, there has been no dedicated discussion on the characteristics of driver fatigue eye movements in the context of the latest intelligent supervision environment. Nevertheless, this research is of great significance for monitoring driving fatigue in the context of intelligent connected environments. 3. Research Methods Principal Component Analysis (PCA) is a commonly used 64 method for data dimensionality reduction and feature extraction. It transforms the original data into a set of new variables, which are linear combinations of the original data. Each new variable contains a portion of information from the original data, and the new variables are independent of each other. These new variables are called principal components. By selecting the first few principal components, it is possible to compress and reduce the dimensionality of the original data, making data analysis and visualization more convenient. By the above method, the original high-dimensional data can be reduced to a few principal components. During this process, although the original feature space may be compromised, principal component analysis can yield principal components that contain the weight information of each original feature, thus still effectively reflecting the information of the original data. The corresponding principal components are arranged in order of the corresponding feature values from largest to smallest. Usually, the P (P≤0) principal components corresponding to the feature values with a cumulative contribution rate of over 85% or 95% can contain all the information of the original features. By using principal component analysis, redundant information in the original data can be eliminated, and the main components of the data can be extracted, allowing for a better understanding of the structure and patterns of the data. 4. Research Conclusion and Discussion The research content of this paper is the driving fatigue characteristics under the supervision platform environment, so the principal component analysis data is the eye movement data of the driver in the experimental group, and the eye movement data is obtained by the laboratory simulation experiment. 4.1. Analysis of research results First, it is necessary to determine whether the data meets the conditions for using principal component analysis. In this study, six original variables of eye movement data were used, including fixation time, fixation count, saccade count, average saccade speed, pupil diameter and blink count.In order to perform principal component analysis, these data need to be Z-standardized so that their mean is 0 and the standard deviation is 1. Next, Bartlett 's sphericity test is needed to check whether there is a spherical skewness between the original variables, that is, whether there is multicollinearity. The test results are shown in Table 3-2, p < 0.05, indicating that there is a correlation between the original variables. The KMO value is 0.889, which is within the general appropriate range, indicating that it is suitable for principal component analysis. Finally, two new comprehensive indicators were obtained by principal component analysis, which explained 59.27 % and 18.35 % of the variance of eye movement data respectively. The specific results are shown in Table 1, Table 2, Table 3and Figure 1. Table 1. KMO and Bartlett test of eye tracking data KMO sampling suitability quantity .889 Bartlett sphericity test Approximate chi-square 168.989 degree of freedom 21 significance 0.000 Table 2. Common factor variance of eye tracking data initial value extraction fixation time 1.000 0.723 the number of watch 1.000 0.755 Number of glances 1.000 0.756 average speed of saccades 1.000 0.736 pupil diameter 1.000 0.764 blink times 1.000 0.763 The common factor variance represents the degree to which the information contained in each variable can be represented by the extracted principal component, also known as ' common degree '.The ' initial value ' indicates that the demonstration information of each variable is 1, that is, 100 %.The ' extraction ' column indicates the degree to which the variance of the variable can be expressed by the principal component. It can be seen that the variance of the variable can be explained by the principal component by more than 70 %. Table 3. Cumulative contribution rate of total variance of eye tracking data Initial eigenvalue Extract the load sum of squares composition total variance proportion accumulation% total variance proportion accumulation% 1 4.132 59.265 59.265 4.132 59.265 59.265 2 1.322 18.349 77.614 1.322 18.349 77.614 3 .568 8.206 85.820 4 .495 6.939 92.759 5 .284 4.269 97.028 6 .096 2.972 100.000 65 Table 4. Eye tracking data component matrix composition 1 2 fixation time 0.715 0.454 the number of watch 0.619 0.413 Number of glances 0.743 0.471 average speed of saccades 0.474 0.223 pupil diameter 0.820 -0.353 blink times -0.254 0.933 Figure 1. Eye movement data lithogram According to the principal component analysis method, complex multidimensional data can be transformed into a few comprehensive indicators through dimensionality reduction, so as to better describe and explain the data. In this paper, after Z standardization and Bartlett spherical test of eye movement data, it is confirmed that the data are suitable for principal component analysis. According to the results of table3 , it can be seen that the cumulative variance contribution rate of the first two principal components is 77.614 %, so two principal components are extracted to represent the original variables. Figure 1 shows that the first two factors cover most of the information of the original variables. According to the component matrix table, it can be obtained that the index pupil diameter and blink times have higher loads on the first and second principal components, that is, the correlation is stronger. Based on the results of principal component analysis, the eye movement sensitivity index of driving fatigue under the intelligent supervision platform mainly chooses pupil diameter and blink times. 4.2. Discussion In this paper, the principal component analysis method is used to determine the sensitive indicators of driving fatigue under the environment of intelligent supervision platform. However, due to the limitation of time and space, this paper mainly analyzes the main indicators of eye movement. In the later stage, this paper intends to carry out the characteristics analysis of ECG and vehicle behavior parameters, and will carry out fusion analysis, in order to carry out the driving characteristics and laws under the environment of intelligent supervision platform in a refined and systematic way, and lay a theoretical foundation for the management and early warning of driving fatigue under intelligent supervision. Acknowledgements We want to express our gratitude to anonymous referees for their valuable and constructive comments. This research was supported by grants from Henan Province Science and Technology Key Project(Project Title: Research on Graded Warning for Fatigue Driving Based on Multi-source Information Fusion in Intelligent Connected Environment; Topic number: 232102240014). References [1] Liu Xiaoshuang, Fang Zhijun, Liu Xiang, Gao Yongbin, Zhang Xiangxiang. Fatigue Detection System Based on Eye and Mouth States [J]. Sensors and Microsystems, 2018, 37(10): 108-110. [2] Zhu Yubin, Yan Xiangjun, Shen Xuqii, Lu Zhaolin. Fatigue Driving Detection Based on Cascaded Width Learning [J]. Computer Engineering and Design, 2020, 41(02): 537-541. [3] Zhao Li. Design and Implementation of a Fatigue Detection System Based on Facial Features [D]. Jinan University, 2020. [4] Yang Xulin, Gao Kai, Lu Wanping, Wang Hongyan, Zhang Yu. Design and Implementation of a System for Monitoring Sleepiness of Urban Rail Vehicle Drivers [J]. Railway Locomotives and Cars, 2021, 41(03): 132-136. [5] Huang Jieyuan. Human Eye Localization and State Classification Based on CNN[D]. Beijing Jiaotong University, 2019. [6] Chen Zhikun, Chi Wanda, Gao Shang, Liu Xiaoxin, Fan Di. Fatigue Detection Algorithm Based on Deep Learning [J]. Computer and Network, 2020, 46(08): 68-71. [7] Wang Qi, Tang Yangshan. A Driver Fatigue Detection Method Based on MTCNN [J]. Practical Technology of Automobiles, 2021, 46(20): 49-51. 66 [8] Huang Ling, Hong Peixin, Wu Zerong, Liu Jianrong, Huang Zixu, Cui Zuan. A Method for Detecting Driver Fatigue Based on Regular Epidemic Prevention[J]. Transportation Information and Safety, 2021, 39(04): 26-34. [9] Li Jingxuan. Research on the Application of Driver Fatigue Detection Algorithm Based on Facial Feature Analysis[D]. South China University of Technology, 2020. [10] Zhu Mingliang, Li Dinggen. Research on Fatigue Detection Method Based on Facial Feature Points [J]. Computer Applications Research, 2020, 37(S2): 305-307. [11] Feng Wenwen, Cao Yinjie, Li Xiaolin, Hu Weisheng. Face fatigue detection based on an improved deep convolutional neural network [J]. Science Technology and Engineering, 2020, 20(14): 5680-5687. [12] Bai Zhonghao, Jiao Yinghao, Bai Fanghua. Driving Fatigue Detection Based on Active Shape Models and Fuzzy Inference [J]. Journal of Instruments and Instrumentation, 2015, 36(04): 768-775. [13] Chen Xin, Li Weixiang, Li Wei, Zhang Wenqing, Zhu Yuan. A Multi-feature Fusion Fatigue Detection Method Based on Improved ASM [J]. Computer Engineering and Design, 2019, 40(11): 3269-3275. [14] Zhao Xuejuan. Research on Driver Fatigue Detection Technology Based on ASM and Information Fusion. Science and Technology Innovation, 2020(30): 120-121. [15] Wang Di. Research and Application of Fatigue Detection Algorithm Based on Human Eye State[D]. University of Electronic Science and Technology of China, 2020. [16] Huang Z, Shi J. Research on Driver's Fatigue Detection by Multi-feature Fusion[J]. Mechanical Science and Technology for Aerospace Engineering, 2018. [17] Wu Kunliang, Yu Suping, Xu Wujun, Fan Hong. Simulation Research on Fatigued Driving Based on BFR Algorithm and Head Posture [J]. Computer Simulation, 2021, 38(08): 152-156. [18] Bing-Ting Dong, Huei-Yung Lin, Chin-Chen Chang. Driver Fatigue and Distracted Driving Detection Using Random Forest and Convolutional Neural Network[J]. Mdpi Ag, 2022, 12(17): 8674. [19] Wang XY, Chen LF, Zhang Y, et al. A real-time driver fatigue identification method based on GA-GRNN[J]. Frontiers Media SA, 2022, 10. [20] WC Xiao, HL Liu, ZJ Ma, et al. Fatigue Driving Recognition Method Based on Multi-Scale Facial Landmark Detector[J]. Mdpi Ag, 2022, 11(24): 4103.