Study On Technology Of Relieving Visual Fatigue Of Stereoscopic Viewing Based On Vertical Parallax Elimination

In various fields, China has great and quick development. In the seventh decade oflast century, people still obsessed with black and white television. Then we hadanalogue color TV, digital color TV (DCTV) and smart TV. Now3D digital televisioncan be seen everywhere in the malls. It is quite easy for people to enjoy the strongvisual impact of stereo and stereoscopic images in the theater, just like we are in thesense. All of these are the effects of science that have pervaded our daily lives.Stereoscopic vision is widely used in various fields, not only in the aspect ofentertainment, but also in the areas of architectural features, image art, precisionmeasurement, virtual reality and even the area of medical.However, many kinds of technical problems follow with the popularity of3DTV.One of the most obvious problems is visual fatigue. Through investigations on largenumber of3DTV users and3D movie viewers, the result shows that long time watching3D videos generally causes different degree of symptoms, such as dry eyes, dizziness,binocular diplopia, blurred vision, nausea and vomiting. These short-time symptomsmay cause permanent damage to human body in a certain range of probability, whichlead to the3D video bottlenecks in the development of relevant industries.Through the in-depth study of3D visual fatigue, there are many factors which canlead to visual fatigue which are mainly in three broad categories, external reasons,physiological factors and psychological factors. The external reasons include electricalperformance of display device and so on. Physiological factors include vertical parallax,the inconsistency between convergence and focus adjustment and some other factors.External reason can only rely on more accurate artificial adjusting or continuousoptimization of electrical characteristics. It can not be avoided. There are someresearches on the inconsistency between convergence and focus adjustment, but fewresearches on the vertical parallax elimination.In real life, human beings form two virtual images in both eyes when watching thesame scene. The two virtual images form three-dimensional geometry information inthe brain. These two virtual images have a relatively small difference because of thespacing between two eyes. Simulating the process and the principle of formingthree-dimensional geometry information of human, we can use two cameras to obtainthe same scene from different angles, and then rebuild the three-dimensional shape andposition based on the parallax principle. Horizontal parallax plays an important role in rebuilding three-dimensional geometry information, which is determined by thedistance between the cameras and the angles among cameras and the scene. But thevertical parallax has no contribution to the three-dimensional geometry informationrebuilding, still can cause visual fatigue. Ideally, setting the two optical center positionsof the cameras to the same height can avoid the production of vertical parallax.However, in fact, the vertical parallax can not be avoided because of the stentdeformation, the inconsistencies of the camera internal sensor, and the impossibility ofcurrent precise processing. The vertical parallax directly affects the quality of the stereoimage synthesis, and causes visual fatigue.The stereoscopic image correction methods can be divided into the method withthe camera calibration and the method without the camera calibration. Obviously, themethod without the camera calibration has greater adaptability. There are more relatedresearches on this field abroad, but few in the domestic. In order to meet therequirement of real-time and accuracy of the3D video synthesis, we put forward to avertical parallax elimination algorithm without camera calibration parameters on thebasis of previous studies. Firstly, SIFT（Scale-invariant feature transform）featurematching algorithm is used to detect matching points of binocular images.3D projectivematrix is calculated according to coordinates and depth values of these matching points.Then3D projective matrix is applied to the left (right) view to get the right (left) one. Atlast, filter is used to fill the positions that have no pixel values.In this paper, the algorithm is implemented by C and Matlab. The multi-view videosequences, breakdancer and ballet, are used for test video sequences. The results ofusing previous algorithm and the proposed algorithm are analyzed and compared.Experimental results show that the performance of the proposed algorithm is better thanthe previous algorithm in the absence of camera parameters.