Hybrid Differential Evolution Algorithms And Their Applications In Gaze Tracking Technology

Gaze tracking technology establishes the relationship between human eyes and object regions by collecting the real-time information of eye movement,which has wide applications.There are some complicated mathematical models in gaze tracking technology.For example,the models of three-dimensional corneal curvature center and two-dimension pupil center are all the nonlinear mathematical ones,and they are difficult to solve directly.The classical algorithms often fail to obtain satisfactory results because of the high demands for real time,precision and stability.For some intelligent algorithms,such as differential evolution(DE)algorithm and shuffled frog leaping algorithm(SFLA),the premature convergence can easily take place,which leads to an unsatisfactory result in precision.Therefore,new hybrid differential evolution algorithms are proposed,and which are used to solve two mathematical problems of gaze tracking technology in this paper.The main work is summarized as follows:1.A hybrid frog leaping algorithm based on DE and Nelder-Mead simplex method(NMSM)is proposed.Because of DE exists a shortcoming that it easily falls into the locally optimal solution,a new hybrid algorithm is presented,known as DE-NM-SFL algorithm,which is based on DE and takes the advantages of NMSM and SFLA.The convergence of DE-NM-SFL is analyzed by using the infinite products theory.Numerical experiments on 10 Benchmark functions are designed to compare DE-NM-SFL with DE,NMSM,SFLA and DE-SFL,which show that DE-NM-SFL is superior to other algorithms for the accuracy and robustness of solution.2.A wolf pack search dynamic differential evolution algorithm is proposed.Firstly,an improved dynamic differential(DDE)strategy is discussed.Secondly,a new hybrid differential evolution algorithm called WPS-DDE is presented with a new self-adaption walking operator of wolf pack(WP).The asymptotic convergence of the WPS-DDE is proved by using the theory of stochastic functional.Comparisons and analyses between WPS-DDE and other four algorithms,namely DE,WPS,DE-WPA and DDE are made through the numerical experiments of 10 Benchmark functions.The results show that DDE-WPS has not only faster in convergence and stronger robust but also higher in operation rate.3.Applications in gaze tracking.The one is converting the nonlinear corneal curvature center model into unconstraint optimization problem,and DE-NM-SFL algorithm,a classical algorithm and four other widely used intelligent optimization algorithms are used to solve this problem.Comparisons and analyses of the experimental results indicate that DE-NM-SFL can achieve higher accuracy and stronger robust than the compared methods.The second one is to convert the two-dimension pupil center problem into two order polynomial fit,which is solved by WPS-DDE and three other algorithms.Experiments demonstrate that the results obtained by WPS-DDE have higher precision,and this approach runs faster.