| With the rapid growth of the automobile industry and extending of transport routes, thecondition of driver fatigue is more serious. The survey found that driver fatigue has become amajor cause of traffic accidents. In order to improve traffic safety and protect the lives andproperty of the people, the development of driver fatigue detection system has a great practicalsignificance.This paper researched on the key technology of driver fatigue detection using the feature ofhuman eyes, as the feature of human eyes is one of most important feature that is able to reflectwhether the driver is fatigue or not. The paper’s main content is including the eyes detection, irislocation, eyes tracking, eyes states detection, etc. and the details are as follows:Firstly, this paper introduced a method of iris location for outer edge and eyes detectionbased on robust estimation for multi-structure. This method was on the basic of detecting thehuman face with AdaBoost algorithm, and then set the main area of the eyes, using the edgedetection operator of the Canny to obtain the edge of eyes. The paper was then proposed a irislocation method for the outer edge based on robust estimation for multi-structure, so we coulduse the prior information to detect the eyes. For200single human face image data sets, thesimulation experiments showed that more accurate location was obtained and the validhypothesis generation was accelerated by the proposed method which compared with theconventional hypothesis generation method of RANSAC and Hough Transform. In addition, theaccuracy of iris location for outer edge achieves94%.Secondly, the paper was proposed a robust eye tracking algorithm based on local sparseappearance model and adaptive template update strategy. We sampled larger overlapped localimage patches inside the eyes region, and made full use of the sparse coding coefficients of thepartial and spatial information of eyes with the alignment-pooling method to calculate the localimage patches inside the candidates region of the eyes to obtain the similarity. After that,tracking continued to use a Bayesian state inference framework in which a particle filter wasused for propagating sample distributions over time. In addition, we exploited a novel eyetemplates updating strategy based on incremental subspace learning and sparse representation,which employed both positive trivial templates and negative ones. This new eye templatesupdate strategy adapted the templates to the appearance change and reduced the influence ofoccluded eye templates as well. Through both indoor and outdoor face image sequences withdifferent environment, the experimental results showed the proposed eye tracking algorithmcould obtain a more accurate and robust tracking result by comparing with several state-of-the-art methods.Thirdly, the state of eyes detection method based on the fuzzy pattern recognition wasproposed in the paper. We fit the upper and lower eyelids and calculated altitude difference of theeyelids based on robust estimation for multi-structure, after we found the structure of eyes wassimilar to the ellipse. We then constructed the fuzzy membership function of the opened, closed,semi-opened state of eyes. And finally, we determined the state of eyes by using the directmethod of fuzzy pattern recognition. Through several of different face image sequences ofdrivers, the experimental results showed the proposed method could obtain a more accurate androbust results, as well as real time.Finally, a method of fatigue detection based on PERCLOS value and blinking frequencywas introduced in this paper. The PERCLOS value and blinking frequency could be calculatedon the basic of recognizing the state of eyes. We could set a reasonable threshold value step bystep to determine whether the driver was fatigue or not. In order to calculate simply, weconverted the ratio of time to the ratio of successive frames. Compared to the method using onlythe PERCLOS value or blinking frequency, more accurate results could be obtained in the paper. |
PDF Full Text Request