12/15/ Fall 2006 Two-Point Perspective Single View Geometry Final Project Faustinus Kevin Gozali an extended tour into the picture…
12/15/ Fall Two-point Perspective World Camera/eye at (0,0,h) h = dist(horizon, bottom of image) Camera vanishes at the center of horizon (v eye ) Image plane parallel to x and z axis Center at (0,f,h) Two vanishing points (vp l, vp r ) X vp r vp l v eye Image Plane x z y
12/15/ Fall Our Assumptions Horizontal Horizon line This holds for two-point perspective 1 VP to the left, 1 to the right Baselines below horizon Uniform heights for all walls X vp r vp l v eye Image Plane x z y base line
12/15/ Fall Vanishing Point Calculation User specifies parallel lines Compute intersection point with Least Square Method = VP Method described by Bob Collins Simply doing MLdivide doesn’t work Done separately for VP left and VP right X vp r vp l v eye Image Plane x z y
12/15/ Fall The Horizon Line Calculated based on 2 VPs Assumption: horizontal horizon in image plane Or, user can specify directly
12/15/ Fall Selecting Base Lines User specifies base-lines For each desired vertical wall Assumption: must be below horizon
12/15/ Fall Computing Depth (y-axis) Similar to 1-point perspective Based on height ratios Camera height as reference Recall: image plane at y = f v eye Image Plane y zSide View dy h base point depth (3D y )f Camera (0,0,h)horizon line
12/15/ Fall Computing Location in x-axis Using calculated depths (y) Compute horizontal distance from v eye in image plane Recall: left side is x-, right side is x+ Horizontal distance amplified using depth ratio to camera height Farther points are amplified more X v eye Image Plane dx 2 dx 1 2D x1 2D x2
12/15/ Fall Specifying Height After 3D location is computed Compute Height Scales Ratio for each base point to camera height Extend walls vertically upward Height of the base-point closest to the camera as reference User picks the desired height, then compute 3D height
12/15/ Fall Ground Surface Ground surface is a prefect rectangle at z = 0 Create Ground mask Up to the farthest base-point Warp texture (using Homography) Point correspondences have to consider 3D depth! Create 3D surface
12/15/ Fall Vertical Walls Similar to ground processing All walls are perfect rectangle No mask is needed Warp texture using Homography Consider distance of each base point as width Uniform height Create 3D model Intersects the ground plane correctly
12/15/ Fall D Model Generation Define the surfaces Based on (3D x, 3D y, 3D height ) Wrap textures Observe model
12/15/ Fall Fun with Texture Sources Texture Interpolation approximately evening view Blending Half morning half night Use this for 3D model
12/15/ Fall Texture Examples
12/15/ Fall Gallery
12/15/ Fall Gallery
12/15/ Fall Gallery
12/15/ Fall Gallery
12/15/ Fall References [1] Chu, Siu-Hang, Animating Chinese Landscape Paintings and Panoramas. A Thesis Submitted to the Hong Kong University of Science and Technology, August [2]Hoeim, Derek; Efros, A. Alexei; Herbert, Martial. Automatic Photo Pop-up. Robotics Institute, Carnegie Melon University, Pittsburgh PA, USA. [3]Single View Reconstruction Lecture slides /2006_fall/www/Lectures/SingleViewReconstruction.pdfhttp://graphics.cs.cmu.edu/courses/ /2006_fall/www/Lectures/SingleViewReconstruction.pdf [4]Perspective Drawing. An online tutorial. [5]Horry, Yoichi; Anjyo, Ken-ichi; Arai, Kiyoshi. Tour Into the Picture: Using a Spidery Mesh Interface to Make Animation from a Single Image. Hitachi, Ltd. [6] Collins, Bob. A guide to compute vanishing points.
12/15/ Fall Q&A Questions?