Build Your Own 3D Scanner: 3D Scanning with Swept-Planes

Session I Introduction The Mathematics of 3D Triangulation 3D Scanning with Swept-Planes Camera and Swept-Plane Light Source Calibration Reconstruction and Visualization using Point Clouds Course Schedule TexPoint fonts used in EMF. Read the TexPoint manual before you delete this box.: A A A A A A A A A A

Triangulation by Laser Striping M. J. Leotta, A. Vandergon, and G. Taubin. 3 D Slit Scanning With Planar Constraints . Computer Graphics Forum , 2008 Manually or mechanically translated laser stripe Per-pixel depth by ray-plane triangulation Requires accurate camera and laser plane calibration Popular solution for commercial and DIY 3D scanners

3D Photography on Your Desk: Bouguet and Perona [ICCV 1998] J.-Y. Bouguet and P. Perona. 3D photography on your desk . Intl. Conf. Comp. Vision , 1998 DIY scanner using only a camera, a halogen lamp, and a stick Per-pixel depth by ray-plane triangulation Requires accurate camera and shadow plane calibration

Assembling Your Own Scanner Parts: camera (QuickCam 9000), lamp, stick, two planar objects [~$100] Step 1: Build the calibration boards (include fiducials and chessboard) Step 2: Build the point light source (remove reflector and place in scene) Step 3: Arrange the camera, light source, and calibration boards

Swept-Plane Reconstruction Geometry

Convert from RGB to grayscale (for luminance-domain processing) Determine per-pixel minimum and maximum value over sequence Evaluate per-pixel “shadow threshold” as average of min. and max. Video Processing: Assigning Per-Pixel Shadow Thresholds

Video Processing: Spatial Shadow Edge Localization Select region of interest on each calibration plane (occlusion-free) Estimate zero-crossings to find leading and trailing shadow boundaries Fit a line to the set of points along each shadow boundary  Result: Best-fit 2D lines for each shadow edge (in image coordinates) 400 600 800 -60 -40 -20 0 20 40 60 column index I 60 (x,y) - I shadow (x,y) 540 550 560 570 580 -60 -40 -20 0 20 40 column index I 60 (x,y) - I shadow (x,y)

Video Processing: Temporal Shadow Edge Localization Tabulate per-pixel temporal sequence (minus shadow threshold) Estimate zero-crossings to find shadow-crossing times  Result: Use shadow-crossing time to lookup corresponding 3D plane 80 100 120 140 -40 -20 0 20 40 frame index I(x,y,t) - I shadow (x,y) crossing frame index for leading shadow crossing frame index for trailing shadow frame index 80 100 120 140

Session I Introduction The Mathematics of 3D Triangulation 3D Scanning with Swept-Planes Camera and Swept-Plane Light Source Calibration Reconstruction and Visualization using Point Clouds Course Schedule

Modeling Lens Distortion G. Bradski and A. Kaehler. Learning OpenCV . O’Reilly Media, 2008 radial distortion tangential distortion normalized image coordinates distorted image coordinates

Intrinsic Camera Calibration Camera Calibration Input How to estimate intrinsic parameters and distortion model? (unknowns: focal length, skew, scale, principal point, and distortion coeffs.) Popular solution: Observe a known calibration object (Zhang [2000]) Each 2D chessboard corner yields two constraints on the 6-11 unknowns But, must also find 6 extrinsic parameters per image (rotation/translation)  Result: Two or more images of a chessboard are sufficient 1 1 1 1 1 1 2 2 2 2 2 2 3 3 3 3 4 4 4 5 5 6 6 7 Estimated Camera Lens Distortion Map camera coordinate system world coordinate system intrinsic parameters extrinsic parameters

Demo: Camera Calibration in Matlab J.-Y. Bouguet. Camera Calibration Toolbox for Matlab . http://www.vision.caltech.edu/bouguetj/calib_doc/

Demo: Mapping Pixels to Optical Rays How to map an image pixel to an optical ray? Solution: Invert the calibrated camera projection model But, also requires inversion of distortion model (which is non-linear) Mapping implemented in Camera Calibration Toolbox with normalize.m  Result: After calibration, pixels can be converted to optical rays

Alternatives for Shadow Plane Calibration J.-Y. Bouguet and P. Perona. 3D photography on your desk . Intl. Conf. Comp. Vision , 1998

Additional Reconstruction Examples J.-Y. Bouguet and P. Perona. 3D photography on your desk . Intl. Conf. Comp. Vision , 1998 J. Kim and J. Wu. Scanning with Shadows . CSE 558 Project Report (U. Washington), 2001 P. Blaer, N. Hasan, C. Tripp, and L. Volchok. 3D Desktop Photography by Eclipse . Project Report (Columbia), 2001 J. Kubicky. Home-Brew 3-D Photography . EE 149 Project Report (Caltech), 1998

Visualizing Point Clouds: Point-based Rendering via Splatting *See the SIGGRAPH 2009 course: Point Based Graphics – State of the Art and Recent Advances by Markus Gross. Swept-plane scanner produces a colored point cloud : a set of 3D points Problem: how to render a point cloud to make it look like as a continuous surface? Splatting: render points as overlapping colored disks If normal vectors are measured as well, render points as shaded ellipses

Visualizing Point Clouds: Splatting with normal vectors and colors

Visualizing Point Clouds: File Formats No standard file format to store point clouds Point = (x,y,z) plus (R,G,B) and/or (Nx,Ny,Nz) It is easy to create an ad-hoc file format Scene graph based file format: VRML International standard: ISO/IEC 14772-1:97 VRML’97 PointSet node includes coordinates (x,y,z) and optional colors (R,G,B), but no normals PointSet { coord Coordinate { point [ 0 -1 2, 1 0 0, -2 3 -1 ] } color Color { color [ 1 0 0, 0 1 0, 1 1 0 ] } }

Visualizing Point Clouds: File Formats IndexedFaceSet node designed to store a polygon mesh can be used to store point clouds with optional colors and/or normal vectors Store point coordinates as vertices Store point colors as colors per vertex Store point normal vectors as normals per vertex Degenerate polygon mesh with no faces is valid VRML syntax IndexedFaceSet { coord Coordinate { point [ 0 -1 2, 1 0 0, -2 3 -1 ] } colorPerVertex TRUE color Color { color [ 1 0 0, 0 1 0, 1 1 0 ] } normalPerVertex TRUE normal Normal { vector [ 1 0 0, 0 1 0, 0 0 1 ] } }

Visualizing Point Clouds: BYO3D Java Viewer

Visualizing Point Clouds: Pointshop 3D [ Zwicker et al. 2002] M. Zwicker, M. Pauly, O. Knoll, M. Gross. Pointshop 3D: An Interactive System for Point-Based Surface Editing . ACM SIGGRAPH, 2002

How to Get the Source Code http://mesh.brown.edu/byo3d

For More Details http://mesh.brown.edu/byo3d

Break Next Session: Structured Lighting and Mesh Processing http://mesh.brown.edu/byo3d

Build Your Own 3D Scanner: 3D Scanning with Swept-Planes

More Related Content

What's hot (20)

Viewers also liked (19)

Similar to Build Your Own 3D Scanner: 3D Scanning with Swept-Planes (20)

Recently uploaded (20)

Build Your Own 3D Scanner: 3D Scanning with Swept-Planes

Editor's Notes