Home | PhD research | PhD work | Mouse trap | CNC milling | CNC milling 2 | CNC artwork | 360 Panoramic images | Camera pan tilt unit | Software | Computer vision

PhD related casual employment work (4 months)


Table of contents
  1. Introduction
  2. Documentations
  3. Power point presentation
  4. Screenshots from Register3D
  5. Screenshots from Join3D
  6. Screenshots from Walkthrough3D
  7. Screenshots from LaserLocalise
  8. Laser localisation

[Top]

Introduction

I was employed as a research assistant on a casual basis for 4 months at Monash University while waiting for my thesis reviews to come back. During this time I wrote a suite of software to perform point cloud manipulations. The suite consisted of 3 (user friendly to operate) programs: Register3D_wx, Join3D, Walkthrough3D. Registere3D_wx is used to register the individual scans to a global reference frame. Join3D is then used to combine the scans to a single (potentially massive) file. Walkthrough3D is then used to visualise this data.

The other task I did was implementing a laser based localisation using a Riegl and Velodyne laser scanner. I wrote a program called LaserLocalise which uses the data resulting from Register3D_wx and Join3D to localise a mobile platform equipped with a Velodyne scanner. I employed a brute force scan matching algorithm and sped it up by utilising the 4 cores on the Intel i7. We achieved a localisation accuracy of about 14cm and 1 degree using very simple techniques with real-time performances.

Below is a collection of various materials detailing what I've done during those 4 months.

  Video
  A 3D walkthrough of Pomonal.


[Top]

Documentations

User documentation for the suite of point cloud manipulation and laser localisation software. These documents show what the programs look like and how they operate together.

  • 3D software suite (1.5MB)
  • LaserLocalise (0.4MB)

    [Top]

    Power point presentation

    This is a power point presentation that I presented to my department. It gives the gist of the project but is missing the live demonstrations.

    laser_localisation.pdf (2.5MB)

    [Top]

    Screenshots from Register3D

    This is the main screen.
    Selecting the registration points between 2 scans. I tend to select corner features because they are easy to localise.
    This is the result of a successful registration. The top shows the combined scan and the bottom shows the overlap using false colouring.


    [Top]

    Screenshots from Join3D

    The user gets to see all the individual scans represented as spatial nodes. Join3D reads the transformation matrix from Register3D and figures out how all the scans are connected to each other in a global reference frame.


    [Top]

    Screenshots from Walkthrough3D

    A scene from Pomonal. The user can navigate the environment like in a video game, though the ability to shoot and blow up objects is missing.
    To achieve fast visualisation, Walkthrough3D converts the raw point cloud to a fusion of textured planes and points. This shows some of the options the user can tweak.


    [Top]

    Screenshots from LaserLocalise

    The main screen with a completed localisation.
    Various options to create the offline laser signature database from the 3D scan produced by Join3D.
    Various options used during the localisation.


    [Top]

    Laser localisation

    The green path shows laser localisation results for tests done at Monash University.
    A google map image of the test site at Monash University. This site was chosen because it represented a bush like environment.
    The 3D map of the test site at Monash University.
    Pomonal localisation results #1.
    Pomonal localisation results #2.
    An aerial shot of Pomonal.
    3D map aerial shot of Pomonal.
    4WD with a Velodyne mounted on top.
    An example of a Velodyne scan with false colouring.