ParaView Used in a Mining Research Environment

Within the Australian Centre for Field Robotics (http://www. acfr.usyd.edu.au/) many Kitware Open Source toolkits are used. VTK is often used for visualization. While open source tools have their advantages, there are times when we want to manipulate and visualize data received from various sources without going through the process of developing software to do so. One product that has become increasingly useful to us is ParaView.

An oft used technique is the conversion of data from various sensors into particles with associated scalar properties which can be visualized and manipulated. For example, laser and radar data in their raw form consist of range, bearing and often other data such as intensity. The data is manipulated though the use of special purpose software to produce Cartesian coordinates along with the associated properties of each particle. Associated properties can be scalars, such as color or intensity. Once the data is manipulated, it is a simple matter to convert the particle data to something that ParaView can use. Within ParaView, the user can build surfaces and or apply other filters as desired. The final results can then be saved as a new data file.

In ParaView it is possible to produce animations, for example an animation of the progress of surface mining. Examples of ParaView and VTK use can be found in (1),(2).

Since we often deal with particle data, it is necessary to provide ParaView with a format it can easily use. Another consideration is that we need to utilize the results from ParaView e.g., clipped or smoothed data. Often outliers or noise must be eliminated. To do this we need to take the ParaView data and convert it back to particles that are used in other applications.

In order to do this, a program was developed that reads the point data and creates a file of type VTP that can be used in ParaView. This program, called ReadXYZ, uses VTK to provide a view of the data in its GUI. It also reads files of type VTP and converts the points into the file to a text of x,y,z values and a scalar. There is minimal information in the GUI as it is essentially a filter. Its interface looks like this:

It has a relatively simple and clean interface, with a help screen and an information screen giving the dimensions of the image and the number of points.

One example of the utility of software like this is a reader that reads the dxf formatted mapping files from a mine. From this we can create a three dimensional map of the mine that can then be rendered in ParaView. Of course, a big advantage here is that you can selectively switch off selected parts of the pipeline to emphasise features like roads or buildings. In this illustration the lines comprising the roads and buildings have been thickened.

Once this is done we can superimpose tracks of vehicles and display properties such as GPS satellite visibility. There are many possibilities here.

As an example of quite sophisticated work using ParaView, one of us (Shrihari) (3),( 4) addressed the issue of large scale terrain modelling for a mobile robot. In this process, the two products, ReadXYZ and ParaView, played an important role.

The Rio Tinto Centre for Mining Automation (RTCMA) uses ParaView as the default data visualization and manipulation engine for some of their projects, in particular the large scale terrain modelling project. Data sources for this project typically include plain text files (consisting of lines of x,y,z, data) or VRML/WRL files containing coordinates and color data. ReadXYZ is used to handle the text files but ParaView is used to directly handle the VRML files.

We have found that the interface of Paraview is superb in comparison to other software. Most of the buttons, options, and menus are tailor-made for rapid visualization and contribute to rapid development in any application. The tools that ParaView provides (i.e. the filters) are extremely useful as well.

The figures below show some of the visualization and manipulation work for a large-scale terrain modelling project which employed ParaView. A visualization tool for such projects will need to provide at least the following capabilities:

• The software needs to load and handle large datasets (upwards of 1-2 million points per dataset).
• Tools for filtering out noise and clipping out chunks of the data for specific purposes.
• Visualization tools including coloring the points in different ways (by depth, distance from one end, etc.).
• Triangulation methods to generate surface maps from point clouds.
• Methods to overlay datasets / surfaces to enable a comparison.
• Convenient methods/interfaces for taking snapshots of the data in various views.

ParaView provided a single, complete and convenient visualization platform for doing all of these tasks. Hence it has been adopted as the data visualization and manipulation standard for this project. With each new attempt at using ParaView in our project, we discover new exciting features that enhance our work. Some examples of the terrain visualized using ParaView are found in the following illustrations:

Each of these datasets spans several kilometers squared in area and typically contains anywhere between 500K and 2 million points per dataset. We typically deal with several such datasets at any time in our project.

The first image above shows the output of a triangulation performed on the output of the research component for modelling of a mine pit, see (3),(4) for more details. This map was generated from a dense elevation map of about 1 million points.

The second image above shows the result of modelling a mine pit about 1.8 x 0.5 sq km in area. The significant aspects of this dataset are the characteristic features on the sides and the resulting complexity. Thus, ParaView provided a quick and convenient means of visualizing complex terrain data and a means for manipulating such datasets. In addition, the tools provided by ParaView are very useful for visually validating our experimental results – for instance, by a simple triangulation of point data one can easily verify, at least on a preliminary basis, if the data captures certain features.

We acknowledge that there have been major improvements in ParaView particularly in respect to the GUI and the inclusion of Python as a programming language within it. However one of the key uses of ParaView its ability to quickly visualize and manipulate data without the need to write filters, etc. Enhancements in this area will always be appreciated.

Acknowledgements
This work has been supported by the Rio Tinto Centre for Mine Automation and the ARC Centre of Excellence Program, funded by the Australian Research Council (ARC) and the New South Wales State Government.

Further Reading

1. Widzyk-Capehart E., Brooker G.M., Scheding S.J., Maclean A.J., Hennessy R.C., Lobsey C.R., Sivadorai M., Millimetre Wave Radar Visualisation System: Practical Approach to Transforming Mining Operations Mechatronics and Machine Vision in Practice, Springer-Verlag Berlin Heidelberg, 2008, p. 139-165, ISSN. 978-3-540-74026-1.
2. Brooker G.M., Hennessy R.C., Bishop M.V., Lobsey C.R. and Maclean, A.J. Millimetre Wave 3D Imaging for Industrial Applications Proceedings of the 2nd International Conference on Wireless Broadband and Ultra Wideband Communications 2007. Presented at the 2nd International Conference on Wireless Broadband and Ultra Wideband Communications, Sydney, Australia, August 27 - 30, 2007.
3. Vasudevan. S. Ramos, F. Nettleton, E. Hugh Durrant-Whyte, H., & Blair, A. Gaussian. Process Modeling of Large Scale Terrain, ICRA 2009.
4. Shrihari Vasudevan, Fabio Ramos, Eric Nettleton and Hugh Durrant-Whyte, Gaussian Process Modeling of Large Scale Terrain. To appear in Journal of Field Robotics.

Andrew Maclean is a Technical Officer who has been working at the Australian Centre for Field Robotics, University of Sydney. His main interests include: visualization of datasets, database design and the development of good, efficient and well tested code.

Shrihari Vasudevan is a Research Fellow at the Australian Centre for Field Robotics, University of Sydney. His research interests span Robotics, Sensor fusion, Machine Learning, Computer Vision and related fields. His current work is with the Rio Tinto Centre for Mine Automation and addresses issues pertaining to modelling large scale and complex terrain.

http://www.kitware.com/media/html/ParaViewUsedInAMiningResearchEnvironment.html