Computer Vision Assignment

Computer Vision Assignment. University of Amsterdam. 2014.

This repository contains the implementations for the Computer Vision assignments. There are 3 assignments located in Assignment1/, Assignment2/, and Assignment3/ directories. Go to the src/ directory of each assignment folder to run the script.

Author

Yikang Wang

Arif Qodari

How to use

Assignment 1 - ICP

Basic ICP

We can simply run the basic ICP implementation by running this following command:

    [R, t, Tr, S_homo, T_homo] = ICP(S, T, TOL, TIME, Tr)

• S is source point cloud
• T is target point cloud
• TOL is minimum tolerance value
• TIME is maximum iteration
• Tr is initial transformation matrix

It will return these following values:

• R is rotation matrix
• t is translation vector
• Tr is transformation matrix (combined rotation matrix and translation vector)
• S_homo is source point cloud in homogeneous coordinates
• T_homo is target point cloud in homogeneous coordinates

Merging Scene - Scenario 1

We can simply run the first merging scenario by running this following command:

    [points_merged, runtime] = merge_orderly(interval, TOL, TIME)

• interval is the number of interval images that will be merged

Merging Scene - Scenario 2

We can simply run the second merging scenario by running this following command:

    [merged_points, runtime] = experiment2()

Assignment 2

We can simply run the chaining procedure by running this following command:

    [M,PVM] = chaining(foldername)
    [M,PVM] = chaining(foldername, 'ransac')

• foldername is the data to use, it can be either TeddyBear or House
• 'ransac' it the parameter, specifying whether to use the Ransac method

It will return the following values:

• M is the patch view matrix for visualization
• PVM is the patch view matrix with a special format, which can be parse by the code in the first part of assignment 3.

Assignment 3

Structure from Motion

Once we obtain Point View Matrix from the Assignment 2, we can estimate 3D structure buy running the main function. Run this following command from the src/ directory:

    main(PVM)

While PVM is the obtained point view matrix. This script will show all partial 3D structures from each point of view.

Building 3D Model

In the folder /build, you can compile it with the command:

    make ../src

And then run the command:

    ./main XXX.pcd

to get the resule mesh mesh.vtk under /build. The vtk file can be easily viewed with ParaVeiw. Two pre-generated vtk files are already under the /data folder ready for visualization.