Building Intelligent 3D Map? Chia-Chen (Jennifer) Hsu Advisor: - - PowerPoint PPT Presentation

▶
building intelligent 3d map
SMART_READER_LITE
LIVE PREVIEW

Building Intelligent 3D Map? Chia-Chen (Jennifer) Hsu Advisor: - - PowerPoint PPT Presentation

Feb 14, 2024 950 likes •1.08k views

How can Robots Help Building Intelligent 3D Map? Chia-Chen (Jennifer) Hsu Advisor: Professor Qixing Huang 1 Fukushima 2011 Japan Earthquake How to explore this area? 2 How to control robots to explore this area? . Expectation:

slide-1
SLIDE 1

How can Robots Help Building Intelligent 3D Map?

Chia-Chen (Jennifer) Hsu Advisor: Professor Qixing Huang

1

slide-2
SLIDE 2

2011 Japan Earthquake

2

Fukushima

How to explore this area?

slide-3
SLIDE 3

3

……….

How to control robots to explore this area? Expectation: get a 3D map which contains the 3d reconstruction in this area

slide-4
SLIDE 4

How to build a detailed 3D map

Go Everywhere, Do not miss any corner

4

Where to go?

slide-5
SLIDE 5

How to build a detailed 3D map

Go Everywhere, Do not miss any corner Reconstruct Objects in Detail

5

slide-6
SLIDE 6

Detect Object of Interest

? How to Detect Object of Interest?

6

!

slide-7
SLIDE 7

Reconstruct Objects in Detail

http://www.robots.ox.ac.uk/~vgg/research/texclass/

7

  • 1. 3D Geometry

Reconstruction

  • 2. Texture

wood, plastic, Glass, metal?

  • 3. High Level Attributes

Deformable, weight, …

? ?

slide-8
SLIDE 8

3D Geometry Reconstruction : Scanning in Detail

8

How to scan it in detail without losing any part? How to do it automatically by robots?

? ?

slide-9
SLIDE 9

High-Level Attributes

Squeeze! Press!

9

Deformable

Weights Newton's second law F=ma

Push

Liquid? touch

slide-10
SLIDE 10

3D Geometry Reconstruction : Segmentation

Real Word Initial Reconstruction Result Push it Same Object ? Two Objects ! Fact: Two Objects Side by Side Push

10

slide-11
SLIDE 11

11

Tasks Mission

  • 3d Reconstruction
  • Input : A series of RGBD scans
  • Output: Reconstructed 3d geometry
  • Process: Use/Implement exist reconstruction algorithm
  • (Challenge) Try to come up the best route to scan an
  • bject (circle? Up/down?)
  • (Challenge) Try different objects
  • Detect Object of Interests
  • Input: Series of Scanned Image
  • Output: Find Object of Interests by Classification

Algorithms

  • Try several image-based object classification algorithms

and test the accuracy.

  • High Level Attributes, especially

deformability

  • Let Robot explores objects using different behaviors
  • Check “deformability ” : Compare the image of the
  • bject before and after robot’s behavior
  • (Challenge) Try to use other data collected from robot to

understand deformability.