A General-Purpose Platform for 3-A General-Purpose Platform for 3-D Reconstruction from Sequence D Reconstruction from Sequence

of Imagesof Images

Ahmed Eid, Sherif Rashad, and Aly Farag

Computer Vision and Image Processing Lab

Department of Electrical and Computer EngineeringSpeed Scientific School

University of Louisville

URL: www.cvip.louisville.edu

The purpose of this project is two folds:

1- Propose a vision platform that lend itself for acquisition of calibrated sequence of images, and concurrently obtain a direct 3D reconstruction by laser scanning.

2- Implement the stereo- and space-carving based approaches for 3D reconstructions on a sequence of calibrated images. These reconstructions will be evaluated against the 3D scanning generated from a laser scanner.

Abstract

The proposed system consists of a laser scanner (Cyberware, Inc, California), on which one or more optical cameras be mounted.

The laser scanner generates a 3-D scan of an object by rotating 360 deg.

The shaft over which the scanner head is mounted can be controlled in terms of speed and angle of rotation.

System Setup

The field of view of the camera(s) is set to cover the same size of objects as the laser scanner.

A sequence of referenced images can be acquired by a calibrated camera(s), mounted on the shaft.

Hence, by the this setup we generate a sequence of images {I1, I2, ... In} which we use in subsequent studies on objectreconstruction

Motion Mechanism

Find points in two or more input images that correspond to the same point in the scene.

Use knowledge of the camera locations and triangulation to determine the depth of the scene point.

One of the main problems of stereo is the correspondence problem.

To solve this problem, a large number of stereo techniques rely heavily on assumptions such as the existence of specific features in the images to produce satisfactory results

(I) 3-D Reconstruction Using Stereo Techniques

X

CLCR

eL eR

xLxR

Epipolar plan

Center of left camera

Center of right camera

3d point

xL, xR = two corresponding points

Left image plan

eL , eR = Left and right epipols

A right epipolar lineA left

epipolar line

Epipolar Geometry of Traditional Stereo

(II) 3-D Reconstruction Using Space Carving

Step1:

Initialize the volume V.

Step2:

• Determine the set of voxels Vis(V) on the surface of V.

• Project each voxel v on Vis(V) to different images where v is visible.

• Determine the photo_consistency of each voxel v on Vis(V).

Step3

• If photo_consistent detain the voxel

• If photo_inconsistent carve the voxel

Photo_inconsistent

Photo_consistent

(III) 3-D Laser Scanner

The 3-D scanner used in the proposed system is the Cyberware 3030.

Its dynamic Range accommodates varying lighting conditions and surface properties.

In operation, the 3030 shines a safe, low-intensity laser on an object to create a lighted profile.

A high-quality video sensor captures this profile from two viewpoints. The system can digitize thousands of these profiles in a few seconds to capture the shape of the entire object.

Simultaneously, a second video sensor in the scanner acquires color information.

Cyclograph Generation

m-cyclograph images (n x k)

x

y

=0=5

=360I1

I2

Ik

I1

k-Input images (n x m)

One Cyclograph image

C1 C2 Cm

x-y- Volume of input images

1

2

3

xo x

(IV) Panoramic Stereo

m-cyclograph images=m/2-cyclograph pairs

L2 L1 R2

xo

R1 Rm/2Lm/2

1

2

m/2

The two views of each pair should be overlapped to construct a cyclograph stereo pair.

Formation of Stereo cyclograph pairs

A Cyclograph Pair

Scanner

Space Carving

Stereo

Results: 3D Reconstruction

Scanner

Space Carving

Scanner

Stereo

Results: 3D Registration

Reconstruction

With Errors

Integration

With Space Carving

Results: Integration

Input Images

Complete Model

Scanner Space Carving