Resaerch On Human Fatigue Estimation Based On Information Fusion

Fatigue is a phenomenon of mentally and physically change caused by long time working. Researches on fatigue estimation have great significance to the employee and the whole society. Based on data fusion alogrithm, traditional estimation methods require more calculation time. In this paper, we make use of the physiological and eye movement characteristics produced by people when working in front of a displayer and estimate their physiological status by using SVM method. This method has a better performace on calculation time and robustness by taking advantage of the relationship and complementarity between the two characteristics.The main work of this paper includes:1. Processing the video data by using computer vision methods. Firstly, train the cascade AdaBoost classifier by using the feature of haar-like. Based on the human face classifier we get, detect the landmarks on human face by using supervised descent method. Secondly, get the eye movement characteristic by computing the area of the eye with the information of those landmarks.2. Analysing the EEG signal. Firstly, collect data of different volunteers in different states by using Mindwave and remove the eye blink components in EEG signal by using wavelet denoise method. Secondly, we use the Welch method to estimate the power spectrum of the EEG signal and analysis the frequency domain of the signal because different frequency band of the signal has the different meaning.3. Obtaining the features in the eye movement signal and EEG signal. Firstly we get the PERCLOS’s P80 value from the eye movement signal and analysis the relationship between the value and the physiological status of the volunteers thus proved that this value can be the input data of the data fusion algorithm. Secondly, we use the energy ratio of different frequency bands in the power density spectrum as a indicator of people’s fatigue states by analysis it’s power density spectrum of EEG. What’s more, we make this ratio a input of our data fusion algorithm by testing the ratio with the fatigue label we set during the data collection.4. Fusing the data. Firstly normalize the input data and divide it into segments by time. Secondly fuse the data by using supported vector machine. During the fusion, we take the RBF function as the core function of the classifier and optimize the parameter of the classifier by using the grid method and the cross validataion method. Thirdly, we propose a new training method based on the relationship between the data segments instead of using only one segment to train. Finally we got a human fatigue estimating model and it works well.