9
3D Reconstruction from Two 2D Images Ted Shultz and Luis A. Rodriguez ECE 533 – Image Processing Fall 2003

3D Reconstruction From Two 2D Images_shultzrodriguez

  • Upload
    gwaiche

  • View
    121

  • Download
    8

Embed Size (px)

DESCRIPTION

Develop an algorithm to reconstruct a 3D world scene from two static 2D images

Citation preview

Page 1: 3D Reconstruction From Two 2D Images_shultzrodriguez

3D Reconstruction from Two 2D Images

Ted Shultz and Luis A. Rodriguez

ECE 533 – Image Processing

Fall 2003

Page 2: 3D Reconstruction From Two 2D Images_shultzrodriguez

Goal

Develop an algorithm to reconstruct a 3D world scene from two static 2D images

Page 3: 3D Reconstruction From Two 2D Images_shultzrodriguez

Approach

• Acquire digital images

• Match up corresponding points in both images

• Estimate a depth mask

• Generate an image from a new view.

Left Image Right Image

Camera Camera

Page 4: 3D Reconstruction From Two 2D Images_shultzrodriguez

Original left image of room Original right image of room

Left image with dots showing 150 locations that were found in both images

Depth mask made from two images (brighter is closer)

Depth mask overlaid on left image to show depth of various objects

Page 5: 3D Reconstruction From Two 2D Images_shultzrodriguez

10 frames from included “Room Slide.mov” file.

Page 6: 3D Reconstruction From Two 2D Images_shultzrodriguez

Original left image of Computer Original right image of Computer

Depth mask made from two images(brighter is closer)

Depth mask overlaid on left image to show depth of various objects

Page 7: 3D Reconstruction From Two 2D Images_shultzrodriguez

10 frames from included “Computer roll.mov” file (animation must be

seen to appreciate)

Page 8: 3D Reconstruction From Two 2D Images_shultzrodriguez

Movies

Low resolution movies are currently online at:http://www.cae.wisc.edu/~shultz/movie

High resolution copies are on CD and available by request

Page 9: 3D Reconstruction From Two 2D Images_shultzrodriguez

Conclusion

• Movies are convincing

• Masks are not perfect, but adequate to fool the eye

• Currently manually intensive, point matching process could be automated.

• Could be modified to incorporate multiple views.