Research Of Human Gaze Tracking Based On Manifold Learning

Gaze tracking technology,also known as eye tracking technology,is a hot topic in computer vision.The research content of this paper is to rely on monocular cameras to achieve non-intrusive line-of-sight tracking under natural visible light.The advantage of this solution is that it only needs a monocular camera,without auxiliary of infrared light source,in theory,the existing laptop,mobile phones and other mobile devices can use the program to achieve human eye tracking,without auxiliary additional devices.This paper first proposed a simple model of eye tracking based on monocular camera.The follow-up work of the article is derived from the parameters required by the model,the program uses Adaboost-based human faces or eyes detection algorithm to detect human faces and eyes with different poses.In order to improve the speed of face detection,A region segmentation algorithm based on color is proposed in this paper,in order to improve the speed of face detection,can detect the candidate region of the face in the image and speed up face detection.A local linear incremental manifold learning algorithm,called LLIMA,is proposed in this paper,which is an unsupervised learning reduction algorithm.The core idea of the algorithm is to discover the local linear structure of the sample set by the compact region extension algorithm,then extend the local linear region by compact region expansion algorithm and IPCA algorithm.When the region cannot be effectively expressed in a specific dimension,stop extending and reduce the dimension of the sample point in the region through the NMF algorithm.repeat the above operations until all the data in the sample set has been processed.When dealing with new samples,the Gaussian model can be used to calculate the probability that the new sample points belong to each region,and the H value corresponding to each dimension reduced by the NMF algorithm is obtained.After weighted,multiplied and merged,the final result can be directly used as the input layer of BP neural network.Exper iments show that the combination of LLIMA algorithm and neural network model can enhance the generalization ability of BP neural network.The article finally shows the effectiveness and feasibility of the program through a specific human eye tracking exper iment.At the same time,the problems existing in the program and the future direction of improvement is pointed out.