Abstract
Introduction: Recovering depth from stereo image pairs is known to have difficulties (Hoffman et al., 2008; Kim et al., 2012). The main issue: Poor matching of the distance between the two eyes' images and the placement of the eyes' prisms. We describe results with a hardware/software method enabling the precise adjustment of the horizontal position of images and the viewing prisms of the haploscope. Methods: A two-piece apparatus was added to prism-and-mirror haploscope that was initially designed with a fixed interpupillary distance (IPD) suitable most viewers. We added 1) a 3D-printed slider for the precise positioning of the prisms. 2) an eye piece with separate viewing ports to ensure a fixed distance between the eyes and the prisms.The IPD of each observer was used to adjust the distance between the centers of the haploscope prisms. Observers used a program developed in MATLAB to adjust the degree of horizontal displacement of left/right images on the screen.We compared results with the original haploscope and with the improved apparatus. Stimuli included random-dot stereograms (RDS), computer-generated objects and pictures of human faces. In the RDS condition, the task was to decide whether the presented shape was located in front or behind the background. In the 3D object and the face conditions, the task was to report whether the object/face was convex, concave or flat. Results: Data indicates significant improvement in accuracy with RDS images with the customized setup. Surprisingly, even after the improvement, not all viewers were virtually 100% correct in perceiving convex faces as convex, whereas they were all virtually 100% correct with physical masks (Keane et al. 2013). Conclusion: Setting up the haploscope to accommodate the viewer's unique IPD distance improves accuracy of depth perception. Despite the improvement, performance with stereoscopic images is still inferior to that with physical objects.
Meeting abstract presented at VSS 2017