Research On Facial Feature Fusion Based Drivers' Fatigue Recognition Algorithm

Fatigue driving is one of the three major causes of traffic accidents.In particularly serious traffic accidents,the proportion of accidents caused by fatigue driving exceeds 40%.Therefore,how to effectively detect the driver's fatigue state has become the focus and hotspot of researchers.In recent years,researchers have attempted to identify fatigue driving from different sources of information,such as the driver's physiological state,facial features,vehicle behavior and so on.According to the source of fatigue characteristics,fatigue recognition technology can be divided into two categories.The first category is fatigue recognition technology based on single information,while the other one is fatigue recognition technology based on multi-information fusion.Among them,the fatigue identification technology based on single information is easily affected by environmental factors,information loss,etc.,and results in recognition failure or misjudgment.Therefore,in order to improve the effectiveness and robustness of fatigue driving recognition,this thesis adopts a multi-information fusion method for fatigue identification.The main work of this thesis is as follows:(1)A facial feature fusion based drivers' fatigue recognition algorithm is proposed.The algorithm consists of a pre-processing module,a feature-level module,and a decision-level module.The pre-processing module is used to detect and track the face,eyes and mouth.The feature-level module mainly utilizes the feature extraction algorithm to identify the fatigue state of each evidence source.The decision-level module combines the fatigue recognition results of multiple evidence sources in the feature level with the comprehensive discrimination result of the previous moment to make a comprehensive judgment of the fatigue state at this moment.(2)This thesis proposes two kinds of feature extraction based fatigue recognition algorithms: Pyramid Local Binary Patterns based Fatigue Recognition Algorithm(PLBP-FRA),and Local Binary Patterns and Reconstructed Histogram of Oriented Gradient Based Fatigue Recognition Algorithm(LBP&RHOG-FRA).And these algorithms can combine PERCLOS or yawn detection to determine the fatigue state of eye or mouth.The simulation results show that PLBP-FRA outperforms other feature extraction based recognition algorithms in eye state recognition,and the performance of LBP&RHOG-FRA is better in mouth state and facial fatigue expression recognition.(3)This thesis improves the Dempster-Shafer evidence theory based decision-making layer fusion method.Since the problem of obtaining basic probability assignment based onstatistical methods or expert experience in the past is too subjective,the posterior probability of the Support Vector Machine(SVM)output will be used as the dynamic basic probability assignment value.For the problem of evidence conflict,the distance-based evidence conflict handling method is adopted.The simulation results show that the recognition performance of the facial feature fusion based drivers' fatigue recognition algorithm is better than that based on single information source.