1 / 16

3D Registration (Automatic Registration of 3D Data sets from Unknown Viewpoints)

3D Registration (Automatic Registration of 3D Data sets from Unknown Viewpoints) Sang-Hoon Kim, Jung-Kak Seo, Cheung-Woon Jho and Hyun-Ki Hong Dept. of Image Eng., Graduate School of Advanced Imaging Science, Multimedia & Film, Chung-Ang Univ., Seoul, KOREA. The Objective of Study.

mara-poole
Download Presentation

3D Registration (Automatic Registration of 3D Data sets from Unknown Viewpoints)

An Image/Link below is provided (as is) to download presentation Download Policy: Content on the Website is provided to you AS IS for your information and personal use and may not be sold / licensed / shared on other websites without getting consent from its author. Content is provided to you AS IS for your information and personal use only. Download presentation by click this link. While downloading, if for some reason you are not able to download a presentation, the publisher may have deleted the file from their server. During download, if you can't get a presentation, the file might be deleted by the publisher.

E N D

Presentation Transcript


  1. 3D Registration (Automatic Registration of 3D Data sets from Unknown Viewpoints) Sang-Hoon Kim, Jung-Kak Seo, Cheung-Woon Jho and Hyun-Ki Hong Dept. of Image Eng., Graduate School of Advanced Imaging Science, Multimedia & Film, Chung-Ang Univ., Seoul, KOREA

  2. The Objective of Study Modeling a real object and environments. In the early years of computer vision - Using camera image (2D projection of the 3D world) - Difficulties : lighting variations, complex shapes... In recent years, due to the advances in 3D sensing technology - Digitized 3D surface data have become widely available - Require registration techniques

  3. Introduction 3D Registration  Finding Rigid Transformation  Finding relative pose between two sensor. Projection Matrix x2=P2*X2 x1=P1*X1 λ[ R | t ]

  4. Previous Method • Feature Based Method • - Very difficult to extract feature • A Similarity-Measured Based Method • Require optimization • (ICP algorithm)

  5. Previous Method A Common Registration Method - Approximate registration (Mechanical positioning device, Manual assistance, Feature extraction..) - Optimize registration by ICP algorithm

  6. T ICP Algorithm Iteratively finding closest point pairs in two surfaces Find rigid transformation (Rotation, Translation) to minimize distance between closest point pairs. [Besl 92]

  7. ICP Algorithm • Advantage • - It does not require feature extraction • It can handle a reasonable amount of noise • Disadvantage • - Correct registration is always not guaranteed • - Require approximate registration

  8. Proposed Method ICP algorithm + Sensor projection constraint - Reduce computation iteration - Reduce possibility of falling into a wrong local minimum - More efficient and accurate registration.

  9. Proposed Method Sensor Projections Compute Sensor Pose Input Range data Find initial Transformation Transform Range Data & Sensor pose Find Overlaps & Closest point pairs NO Find Transformation E<threshold YES Registration

  10. Initial Transformation • Compute Covariance matrix - Centroid of points cloud C, Covariance Matrix M : • Compute Eigenvector - Eigenvalue decomposition of covariance Matrix. - Three Eigenvectors are column vectors of U matrix. M (Covariance matrix) = UDUT • Compute Transformation H

  11. Initial Transformation Find eigenvectors from each of data sets. Coarse registration using eigenvectors.

  12. Proposed Method Finding overlapped regions Occluded regions From 1st sensor Self-Occluded regions from 1st sensor 2nd Sensor 2nd Sensor 1st Sensor 1st Sensor

  13. Experimental Result Distance error of closest point pairs A common ICP Algorithm Comparison of common and proposed method Proposed Method

  14. Experimental Result Distance error of closest point pairs A common ICP Algorithm Comparison of common and proposed method Proposed Method

  15. Experimental Result Registration of multiple range data sets from unknown views

  16. Questions?

More Related