Research On Human Eye Detection Method In Intelligent Strabismus Diagnosis

Strabismus is a common and frequent eye disease,and about 4%of people suffer from strabismus.Strabismus not only has serious consequences for vision,but also has serious impacts on patients'mental and psychological,social,work and employment,and quality of life.In order to give appropriate treatment to patients with strabismus,timely and effective diagnosis is particularly important.However,the clinical diagnosis of strabismus relies on the execution and interpretation of the physician,and the diagnosis is subjective.With the continuous development of imaging technology,computer technology,image processing technology,etc.,especially the wide application of human eye detection technology,computer-aided intelligent strabismus diagnosis technology is increasingly popular with researchers due to its automatic,objective,and repeatable characteristics.At present,the research of domestic and foreign researchers on the intelligent strabismus diagnosis is still immature.The main problems are as follows:(1)The diagnosis system using eye tracking equipment(such as eye trackers)or the dedicated equipment for eye detection to detect and track eyes is expensive.And it generally requires a pre-test calibration process,making it unsuitable for young children and patients who cannot undergo the conventional sensorimotor examinations.(2)Existing equipment or research for the diagnosis of strabismus mainly gives conclusions about the presence or direction of strabismus,while the quantification of the degree of strabismus is less studied or only a rough estimate.(3)Most of studies focus on the algorithm design of intelligent strabismus diagnosis,while the strabismus specialist test are still performed and manually recorded with cameras by professional doctors.The intelligent diagnosis of strabismus is semi-automatic.A systematic introduction and research on key issues are carried out,focused on the human eye detection method in the intelligent diagnosis of strabismus in this dissertation.The main research work of this dissertation is summarized as follows:1.In order to achieve accurate and robust pupil positioning on low-quality eye images acquired by cheap consumer hardware in the less-constrainted environment in the context of intelligent diagnosis of strabismus,a pupil localization algorithm based on cascaded Haar features and modified Starburst is proposed.First,through the cascade of Haar features,and Histogram-Kmeans clustering segmentation,the rough pupil area is obtained;The pupil boundary points of different sectors are obtained by edge detection,edge filtering and edge sectoring;Finally,the ellipse fitting of the pupil is achieved based on the sectorized RANSAC strategy.The proposed algorithm and four state-of-the-art pupil localization methods have been tested on the datasets Pupil-5795,Swirski-600 and CASIA-581.The experimental results demonstrate that our algorithm outperforms the other four methods on the three datasets,and has achieved 92.1%,85.2%,and 80.2%pupil positioning accuracy within 5 pixel error.2.Aiming at the measurement of strabismus deviation only by pupil positioning cannot rule out the interference of slight head movement in the context of the intelligent diagnosis of strabismus,and the eye detection of individual feature cannot comprehensively reflect the overall characteristics of the eye,a convolutional neural network-oriented evolutionary parametric eye modeling algorithm is proposed.Firstly,an eye model is composed of four parametric curves.A convolutional neural network Dense Net-121 is adopted as the fitness evaluation to guide evolution.The parametric eye model is evolved towards optimization along with the GA evolutionary search to obtain the optimal model parameters.The proposed algorithm and five state-of-the-art eye modeling methods have been tested on the FAED-50 and CASIA-Iris-Distance datasets.This algorithm performs best on the FAED-50,reaching 0.62,0.61,0.27 and1.50 for the contour fitting criteria of each component of the eye and the comprehensive eye E_iris?E_pupil?E_socketand ECT_total,respectively,and 86%and 96%for the accuracy of the iris and pupil positioning within the error range of 5 pixels.On the CASIA-Iris-Distance,the algorithm also achieves comparable or optimal results.3.In view of the problem that the current strabismus specialist test relies on the execution of professional doctors and the automation degree of intelligent strabismus diagnosis is not high,an automatic cover and synchronous tracking hardware platform for intelligent strabismus diagnosis based on cover test is built.The operator sends instructions through the host computer interface and transmits them to the FPGA via USB3.0.After receiving the instructions,the FPGA controls the servo motor to perform corresponding actions;Near-infrared camera module synchronously records the near-infrared video of the patient.The system completes the automatic cover test of strabismus specialist examination within 50s.The statistical analysis of 24strabismus data verifies the feasibility of the proposed system for realizing the diagnois of eye strabismus.4.Aiming at the problem of insufficient quantification of eye deviation measurement in intelligent diagnosis of strabismus,we collected 24 valid strabismus data,and proposed an intelligent video strabismus diagnosis method based on automatic cover test,including eye region extraction,iris measure and template detection,key frame detection,pupil localizaton,deviation calculation and strabismus diagnosis.The intelligent diagnosis method can realize automatic strabismus diagnosis,especially for the accurate measurement of strabismus deviation.The accuracy was over 91%,in the horizontal direction,with an error of 8 diopters;and it was over 86%in the vertical direction,with an error of 4 diopters.In addition,the proposed method achieved 88.9%sensitivity,93.3%specificity,and 91.7%accuracy for the diagnosis of strabismus.