Project Description

This project addresses the problem of noise filtering using the tensor voting approach described in the lecture (or the algorithms described in the paper" N-Dimensional Tensor Voting" where you set N=3) to reject outliers inherent in a noisy 3D point set, which may be sparse (obtained from stereo matching, for instance) or quasi-dense (obtained from a laser range finder). The input consists of a set of noisy 3D points. The output of the tensor voting system is the corresponding denoised point set. We also provide the viewworld files of all the inferred tensor for the purposes of visualization.

Views

Below are the images showing different views of all the six testing data sets.

crown crown noisy crown: noisy_data_view1 crown: noisy_data_view2
crown inlier crown: inlier_data_view1 crown: inlier_data_view2
crown inferred crown: inferred_data_view1 crown: inferred_data_view2
mannequin mannequin noisy mannequin: noisy_data_view1 mannequin: noisy_data_view2
mannequin inlier mannequin: inlier_data_view1 mannequin: inlier_data_view2
mannequin inferred mannequin: inferred_data_view1 mannequin: inferred_data_view2
peanut peanut noisy peanut: noisy_data_view1 peanut: noisy_data_view2
peanut inlier peanut: inlier_data_view1 peanut: inlier_data_view2
peanut inferred peanut: inferred_data_view1 peanut: inferred_data_view2
Theo Theo noisy Theo: noisy_data_view1 Theo: noisy_data_view2
Theo inlier Theo: inlier_data_view1 Theo: inlier_data_view2
Theo inferred Theo: inferred_data_view1 Theo: inferred_data_view2
unknow1 unknow1 noisy unknow1: noisy_data_view1 unknow1: noisy_data_view2
unknow1 inlier unknow1: inlier_data_view1 unknow1: inlier_data_view2
unknow1 inferred unknow1: inferred_data_view1 unknow1: inferred_data_view2
unknow2 unknow2 noisy unknow2: noisy_data_view1 unknow2: noisy_data_view2
unknow2 inlier unknow2: inlier_data_view1 unknow2: inlier_data_view2
unknow2 inferred unknow2: inferred_data_view1 unknow2: inferred_data_view2

Viewworld files

viewworld utility

Choose the Viewworld files below for each sample input to see the demos.

peanut noisy              peanut inlier              peanut inferred tensor
crown noisy              crown inlier              crown inferred tensor
mannequin noisy      mannequin inlier      mannequin inferred tensor
Theo noisy                Theo inlier                Theo inferred tensor
unknow1 noisy          unknow1 inlier          unknow1 inferred tensor
unknow2 noisy          unknow2 inlier          unknow2 inferred tensor

People

  • Bin XU,
    HK University of Science and Technology, Fok Ying Tung Graduate School

  • Haiyan YANG,
    HK University of Science and Technology, Dept. ECE

    Download Testing data sets

    Noisy_peanuts
    Noisy_crown
    Noisy_mannequin
    Noisy_Theo
    Noisy_unknown_data_1
    Noisy_unknown_data_2

    Useful Links

  • COMP 5421 course website
  • N-Dimensional Tensor Voting and Application to Epipolar Geometry Estimation
  • A Closed-Form Solution to Tensor Voting: Theory and Applications