Research Of 3D Visual Comfort About Geometrical Distortions And Asynchronization

3D films have been fairly universal recently, which have occupied a large proportion among the fresh films. But there are also a part of the audiences saying they are suffering various degrees of uncomfortable symptoms, like dizziness, eye pain etc. For the sake of ensuring the comfort of the audience to watch a good 3D video, the problems related to 3D visual comfort and 3D film making need to be tackled. After solving the similar problems, the 3D industry will have a higher popularity rate in the future.During taking a 3D film, when two cameras have different coordinates with each other, there would be a great impact about 3D visual comfort. Firstly, the camera coordinate system is set up, using the baseline distance and the angles between the optical axes of two cameras as parameters marking the locations. Based on the theory about epipolar geometry, combining with inspection about disparity histograms, qualitative analysis about how the geometric distortion in all directions effects the horizontal and vertical disparity distributions will be given. The corresponding subjective experiments are designed based on the physical conditions, which are used to study the relationship between the geometrical distortions and 3D visual comfort. After checking the MOS of experimenters, the conclusion that the keystone distortions can be ignored in the physical conditions has been made. For the distortion having obvious effects on 3D visual comfort when test scenes changed, considering the disparity distribution, four features, which are the kurtosis ratio of horizontal disparity, the absolute variations of vertical disparity, the ROI angles between two views and the area ratio of the maximum ROI, are extracted. The Support Vector Machine is used to build the model about geometrical distortions, and the model is also been confirmed. Based on the statistic data, we can conclude that the bigger the absolute variation of vertical disparity is, the worse the 3D visual comfort is.In the process of taking a 3D film, when using two separate cameras for video capture, it is difficult to ensure that the frames in two views are exactly in the same time. 3D films tend to have rich sports elements, as the frames in two views are not synchronized, which will cause motion disparity, whose direction and size are related with the direction and speed of moving objects in the scene. Finally the motion disparity will have some impact about disparity distribution for the test scenes, which may have some effects on 3D visual comfort.In order to help solve the problem like this, based on classified geometrical parameters, this paper proposes a frame synchronization detection algorithm between two views. Firstly, through SIFT algorithm to select the key points set, then examine the motion characteristics of the key points via optical flow method, combing image segmentation via Mean-Shift algorithm in Luv color space, the key points are classified into two sets based on different targets and motion characteristics. Then Levenberg-Marquardt optimization algorithm combination with epipolar geometry is executed, combing with the matching information provided by two classified sets about keypoint pairs, the two types of geometric parameters are estimated respectively. Besides, a corresponding cost function about frame difference estimation is raised, and ultimately the frame difference between two views is estimate, whose motion features are synchronized. Based on input sequences, by the proposed method presented in this paper and manual selection of keypoints classification method are compared. Based on the cost function curves about two methods, the robustness and effectiveness of the proposed method are further validated.