Full Body Scanning by Daniel Evora. Calibration Left & Right.

39
Full Body Scanning by Daniel Evora

Transcript of Full Body Scanning by Daniel Evora. Calibration Left & Right.

Page 1: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Full Body Scanning by Daniel Evora

Page 2: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Calibration

Page 3: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Left & Right

Page 4: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Mesh

101158 triangles

Page 5: Full Body Scanning by Daniel Evora. Calibration Left & Right.

3D RECONSTRUCTION USING STRUCTURED LIGHT by Stefanie Handojo

COMPSCI 117

PROJECT IN COMPUTER VISION

STEFANIE HANDOJO

Page 6: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Extracting 2D Points from the images

Decode Construct the 3D Points

Triangulation

Page 7: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Create Mesh Getting Rid of Long edges / Far away Neighbors

Filling Holes and Mesh Smoothing

Page 8: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Mesh Alignment Combining Meshes into Final Model Poisson Surface Reconstruction Software

Page 9: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Josh Tutwiler

• Goal: to construct a 3-D model of a bowling pin from 2-D images.

Page 10: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Smooth the Mesh

• nbr_smooth– Move each point to the mean of its neighbors

Page 11: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Meshes

Page 12: Full Body Scanning by Daniel Evora. Calibration Left & Right.
Page 13: Full Body Scanning by Daniel Evora. Calibration Left & Right.
Page 14: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Computer Vision Default Project by Patrick Flynn

Original Image

Page 15: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Computer Vision Default Project

Image Scans – 3 viewpoints

Page 16: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Computer Vision Default Project

Manually Cleaned Up

Page 17: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Computer Vision Default Project

Aligned with my ICP (lsqnonlin)

Page 18: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Default Project by Phong VuongRight

Left

Page 19: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Mesh Cleaning

Page 20: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Mesh Alignment

Page 21: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Poisson Surface Reconstruction

Page 22: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Default Project by Roula Lagaditis

Chosen Object: Bender-bot

Using structured light, recovered front, back, right and left

Page 23: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Default Project

Chosen Object: Bender-bot

Mesh Aligning - Front and Back scanned images, using “rigid-alignment”

Page 24: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Default Project

Chosen Object: Bender-botFinal recovered shape

Page 25: Full Body Scanning by Daniel Evora. Calibration Left & Right.

CS117 Final ProjectDanny Miller

3 objects

~ 7 scans per object

~ 2 GB of pictures

Idea – adding a green backdrop could make it easy to filter out the background

Green tablecloth from Party City - 79¢

Page 26: Full Body Scanning by Daniel Evora. Calibration Left & Right.
Page 27: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Green Removal

• Created a windows program in C# to remove the green from pictures.

Page 28: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Problems - Reflectivity

Page 29: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Scans

• 5 stage scanning process– Auto-Pruning

• Several passes

– User Pruning• 3 views in 2d• 30 views in 2d

– Smoothing• Several passes

– Normals– Colors

Page 30: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Alignment

• Using the linear algebra approach

Page 31: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Color?

Page 32: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Mesh Creation

Page 33: Full Body Scanning by Daniel Evora. Calibration Left & Right.
Page 34: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Object Centric Photo BrowsingTony Tran

Input

Page 35: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Part3: Estimating relationship between images.Image i’s sift points Image j’s.sift points

Input:

Find Correspondences(matches)

Compute Essential matrixand remove outlier matches RANSAC

E

Remove incoherent matchesbased on a triangulation heuristic

Remove bad matchesWith triangulation heuristic

Page 36: Full Body Scanning by Daniel Evora. Calibration Left & Right.

CS 117 Project: Motion CaptureBy Blake Atkinson

• Materials• 5 different colored sets of

appx. 3v LEDs

• Electrical tape

• Glove

• 9v Batteries

• Red and Black Wire

• More patience than you have

Page 37: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Since the epipolar lines are calculated using the Fundamental Matrix, which is calculated from your initial SIFT points, they too should land on the epipolar lines. If not you’ve done something wrong. Here we have the SIFT points (red & yellow) and the corresponding epipolar lines (blue) based on those points. The left image gets it’s epipolar equations from the right points, and vice versa.

Epipolar Geometry by Nick Schiffelbein

Page 38: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Automating Camera CalibrationSam Hallman

But how do you solve for a matrix??

Page 39: Full Body Scanning by Daniel Evora. Calibration Left & Right.

Try #2 with the MK symbol