Fatigue State Recognition System Based On Gait Information

With the improvement of living standards,hypertension and other diseases become younger,people’s awareness of physical exercise continues to increase,but excessive exercise may cause muscle and fascia strain,myocardial damage and other diseases.The current fatigue level quantification method for fatigue detection is relatively simple.It divides fatigue into two states of fatigue and non-fatigue,or divides the fatigue level according to the "Perceived Fatigue Rating Scale".These two methods are greatly affected by subjective and personal factors.In the current fatigue detection method,feature selection and fatigue recognition model parameter optimization are carried out separately,and it is impossible to achieve global optimal fatigue recognition accuracy.This thesis uses the percentage of maximum oxygen consumption to quantify the fatigue level,which weakens the influence of personal factors to a certain extent and has strong objectivity.In this thesis,the genetic algorithm is used for feature selection and fatigue identification parameter optimization at the same time,and the fatigue identification accuracy reaches the global optimum.In this thesis,a fatigue recognition system based on gait information is designed,which is mainly composed of a data acquisition module,a data transmission module and a fatigue level recognition module.The data acquisition module uses the pressure sensor to collect 8-way plantar pressure signals,uses the inertial sensor to collect 9-way signals at the ankle joint,and sends the signals to the fatigue level recognition module through the data transmission module.The fatigue level recognition module receives the data,uses the dynamic time warping algorithm to segment the gait sequence,and performs statistical feature extraction,interpolation and frequency domain transformation on the segmented gait sequence.The fatigue level identification module uses the machine learning classification model to identify the fatigue state of the time-domain signal,frequency-domain signal,and time-domain statistical features respectively.The best model for time domain signal and frequency domain signal is XGBoost,and its 5-fold cross-validation accuracy rate is 92.10% and81.72%,respectively.The best model for time-domain statistical features is random forest,and the accuracy rate of 5-fold cross-validation is 91.31%.After model prediction time evaluation,all model prediction times used meet the real-time requirements.After comparing the recognition accuracy and prediction time,this thesis uses time-domain statistical features to identify fatigue levels.In this thesis,the traditional feature selection method and parameter optimization method are used to select the signal type and statistical feature type.First,the optimal signal type combination is selected by the feature importance,and then the same method is used to select the optimal statistical feature type combination.This method further reduces the amount of data and improves the recognition effect.The best model is Extra Trees,with an accuracy rate of 94.74%.In addition,this thesis also uses genetic algorithm to perform feature selection and parameter optimization at the same time,and obtains better recognition effect than traditional methods.The genetic algorithm takes the model’s 5-fold cross-validation fatigue recognition accuracy as the optimization direction,and at the same time obtains the optimal signal type combination,the optimal statistical feature type combination and the optimal model parameters.Among the models with roughly the same fatigue recognition accuracy,the model with fewer sensors and fewer signals is selected as the optimized optimal fatigue state recognition model.Among them,Extra Trees has the best fatigue state recognition effect and uses the least number of signals and statistical features.In the case of,the fatigue level recognition accuracy is the highest,reaching 96.7%,which is improved compared with the traditional method.This system collects human ankle joint motion information and plantar pressure information in real time,and uses a machine learning classification model to objectively evaluate the fatigue state of human muscles.It is easy to use and does not affect the normal activities of subjects.It is simple and has good application prospects.