Research On Mobile Terminal Eye Tracking Algorithm Based On Deep Learning

Visual tracking technology is an important subject in the field of computer vision.With the emergence of high-performance image acquisition and computing equipment and the emergence of related algorithms for image processing and analysis,visual tracking technology has gradually entered into various fields in our daily life.The application of visual tracking technology in the field of human-computer interaction can not only change the traditional human-computer interaction mode and bring new and efficient experience to users,but also bring great convenience to users with limb movement disorder.However,the traditional eye-tracking technology often relies on other hardware equipment such as eye-tracking device,which is not only expensive,but also brings certain inconvenience to users.Our project takes the process of using i Phone as an example,trying to apply deep learning technique to the eye tracking tasks and allows the user not to use eye tracker and other hardware devices,only rely on user's photo images when operating mobile phones taken by front-facing camera,using pure software way to predict the user view in focus position on the phone's screen,which achieve with your eyes instead of your fingers to touch screen operation.Firstly,this paper introduces the basic theory of convolutional neural network.Convolutional neural network is evolved from standard neural network.Convolution operation is used to realize interlayer connection,which can significantly reduce the computational cost and meet the requirements of local feature extraction of signals.It has been widely used in the field of computer vision.Secondly,considering the real-time and accuracy factors comprehensively,HOG and GBDT method are adopted to segment and locate the user's face and human eyes as well as blink detection.The design of blinking detection,face detection and human eye positioning module is completed to provide input information for line of sight focus prediction.Thirdly,the deep convolutional neural network is used to design the focus module of line of sight.In this project,we tried the multi-input regression-based line-of-sight focus prediction method,the multi-input category-based line-of-sight focus prediction method and the single-input end-to-end line-of-sight focus prediction method.In terms of network structure,the Alex Net-based network architecture and the residual neural network architecture with batch normalization are respectively used.It can predict the focus position of users' eyes on the phone screen more accurately and quickly,and then achieve the function of touching screen using eyes instead of fingers in theory.Finally,the real-time and accuracy of the line of sight tracking prediction method tried in this topic are compared,and the strategy of improving the performance is discussed.