Research On Multi-view Generation And Fusion Display Technology Based On Human Eye Tracking

With the development of information society and innovation of information display,people no longer satisfied with the ultra-high resolution of the flat display,however pursue more intuitive and reality of the stereoscopic display.The stereoscopic display separates the different view images in the space or the time dimension,and the left and right eyes watch the different view images and form the stereoscopic effect through the fusion of the brain.The free stereoscopic display technology based on the optical grating has been the favor of researchers and users because no need to wear auxiliary equipment,which has a wide range of applications.However,there are inherent defects in the grating-type free stereoscopic display technology,that is,the location of the pupil is fixed and the phenomenon of "fake stereoscopic" occurs;which will cause the stereoscopic view to be discontinuous and affect the free stereoscopic display effect.In this paper,around the free stereoscopic display principle,stereoscopic image acquisition,stereo image fusion,adjusting the stereoscopic images according to human eye tracking,to introduce stereoscopic display,which make up an integrated software solution for acquisition and display and adjusting.The solution focuses on two problems,one is the "fake stereoscopic" phenomenon of grating-type free stereoscopic display technology,and another is the problem of less stereoscopic image resources.We use human eye tracking technology to track the human eye position real-time to adjust stereoscopic images,and always maintain a good stereoscopic effect.The second problem is solved by loading the stereoscopic model using the OpenGL rendering multi-view image to obtain a stereo image pair.The main work and content of this paper are summarized as follows:1.Research human eye tracking main technical points.In this paper,we use the Adaboost algorithm to train the strong classifier with Haar feature,and use the strong classifier to construct the cascade strong classifier to detect the human face quickly.And then use the ASM algorithm to train and accurately locate the human eye position.To speed up the detection of the program,use skin segmentation to reduce the candidate region for face detection,and use template matching to track the face region to reduce the frequency of face detection.2.Use OpenGL to render stereoscopic models to generate multi-view image.This paper mainly research the structure of PLY and OBJ,obtains the vertex array of the model by loading the model file,and OpenGL renders the stereo model according to the vertex array.We can usually obtain multi-view images by parallel camera array and convergent camera array.In this paper,the calculation method of the adjacent camera spacing in the case of parallel camera array is mainly introduced,so that the parallax of multi-view image is always kept at steady value.3.Reduce the "fake stereoscopic" by human eye tracking to adjust stereo image.The method proposed in this paper is to divide each exit pupil of the multi-view image into three parts.After the human eye position moves for a certain distance,the image mode is switched so that the human eye position is always at the center of the image exit pupil,while maintaining good stereo effect and minimize the brightness changes when image mode switching.This paper introduces how to measure the correspondence between the exit pupil region and the camera pixel in a single camera.