Analysis Of Eye Tracking Data And Algorithms Research On Data Fusion

The eye is an important organ of the human to obtain information from the outside world.The eye is the information transmission channel of a high quality and high efficiency.The research shows that the human brain's knowledge about 80% is obtained through the eyes.With the development of science and technology and the need of human life,the analysis of eye tracking data naturally has attracted a large number of scientific researchers' attention,and there are more and more researchers have joined the research team of eye tracking.The data fusion has attracted many researchers' attention,the results of data fusion are of high reliability,less ambiguity,and have been widely used in both military and civilian.Ensemble clustering is as an extension of data fusion in clustering having better performance than single clustering algorithm in robustness,accuracy,parallelism and scalability.Ensemble clustering has been applied in many areas,such as intrusion detection,mobile communication,and face recognition.Based on the theory of the analysis of eye tracking data,this thesis proposed a new ternary eye movement classification algorithm based on shape features using FCM.Based on the existing ensemble clustering algorithms using co-association matrix,this thesis proposed a new ensemble clustering algorithm based on the FCM.As a tool of the analysis of the eye tracking data and data fusion,FCM is also applied in many areas.As an extension of the classical FCM,this paper proposed a new adaptive fuzzy c-means clustering algorithms for interval data type.This new method is based on intervaldividing technique.When computing the distance between two interval values,it takes into account every point in both intervals.And there is a vector of weights which can make algorithm more adaptive for different data sets in the new method.The experiments shows that the new algorithm both in synthetic data sets and real data set has better performance.Many methods in eye tracking data classification use speed threshold to separate saccades.However for the data sets collected by the eye tracker with low sampling rate and poor performance,speed threshold is not ideal because of the losing frame of the eye tracker.Therefore,this thesis uses distance threshold to separate saccades.And then this paper uses the shape features and cluster center obtained by FCM to identify fixations and smooth pursuits.The experiments demonstrate that this method can obtain good classification results.There are many ensemble clustering algorithms based on co-association matrix.When constructing the co-association matrix,many of them use clustering members obtained by K-means.However this method can't show the similarity and difference between patterns very well.In view of this,this thesis proposed that constructing co-association matrix use the membership of the patterns.The experiments demonstrate that this method is ideal.