A Low Cost 3D Scanner for Navigation and Modelling
Introduction
The Tilt Unit
Data Processing and Fusion
Results
Discussion
Acknowledgements
References
Introduction
It is well known that navigating mobile robots is a problem that has been addressed many times in the last years and that perception of the environment is still a most relevant issue. Very often, what autonomous robots need to perceive is simply the space that they are expected to traverse. Perception of occupied and free space is therefore the main purpose of many systems developed, ranging from the early sonars up to laser range finders and vision, including stereo. For the pure purpose of locomotion, perception in two dimensions is normally enough since ground-based robots have planar locomotion directives. However, if further perception or more elaborate planning is required, then more information of the surrounding space has to be gathered. This yields the necessity of 3D perception and, for more advanced navigation, object extraction and identification; and this means not only object shape, but also patches and textures.
Besides the mentioned navigation requirements, a new field is now appearing that covers the 3D reconstruction of objects and environments. For objects of limited dimensions, solutions do exist for several years now, but for open spaces with larger dimensions (depths of tens to hundreds of meters) fewer approaches are available. Nonetheless, there are already commercial solutions to acquire 3D range, such as those furnished by some companies like Riegl Laser Measurement Systems, Cyra Technologies, Zoller & Frölich, Callidus Precision Systems, among others, but their price is still prohibitive (from 30 000 € to 100 000 €). Still, fusing generic, unregistered, intensity image with 3D laser range for full scene reconstruction is evolving and few groups have produced interesting results such as in Sequeira et al. [Sequeira99], El-Hackim et al. [El-Hakim98] and Stamos and Allen [Stamos00]. Indeed, creating 3D models of real world scenes is an important topic of research with applications in many areas such as virtual museums, game and entertainment, architecture description and modelling, virtual reality, robot navigation, archaeology, inspection, cultural heritage and many industrial applications like reverse engineering. The issue is very challenging due to acquisition of large-scale data, complexity of the geometry and difficulties to cope with reflectance properties of the objects and variations in the lighting in the scenes.
Having stated the interest of 3D range, the obvious approach is to try to adapt a traditional, much less expensive, 2D range finder (3-4 k€). Although this idea is not new, as can be seen in Surmann et al. [Surmann01] and its references, the concept was initially developed without the authors being aware of other approaches. The path to develop such a 3D unit covers the following problems: develop a mechanically stable tilt unit that would rotate the base 2D laser range finder, and synchronize the pan and tilt information to produce a spherical representation of points that would further be processed for navigation or modelling. The laser unit used in this project is a SICK LMS 200, which is an indoor version, but has, nonetheless, operated well also in outdoors, as shown later.
Back to Top
The Tilt Unit
To transform a SICK LMS 200 (a traditional 2D laser Scanner) into a 3D laser scanner, it was necessary to develop a Pan and Tilt unit that would rotate the base 2D laser range finder, and synchronize the pan and tilt information to produce a spherical representation of points that would further be processed for navigation or modelling.
Since navigation was the primary purpose of the approach, the laser was set in tilt position (scanning preferably in horizontal) because it allows faster and more efficient detection of vertical structures and obstacles, such as doors, chairs, tables, etc. Hence, the mechanical structure to support the laser range finder should have adequate robustness and stiffness and also it should not interfere with any mechanical part in the final device.
A control unit was also developed based on a PIC microcontroller in order to control and synchronise the step motor used to actuate the tilt unit.
Diagram of the developed scanner and communication protocols
The sensor mounted on a robotic platform (Robuter II)
Movie of the sensor at different acquisition speeds: Download Movie (9.2Mb)
Back to Top
Data Processing and Fusion
This section presents the reconstruction process used to compute 3D triangulated models from the cloud of points provided by the developed sensor. Two main issues are to be considered when computing 3D models from range points, first how to triangulate the points in order to obtain surfaces, and second how to register the several range images necessary to cover the whole model.
For the triangulation of the cloud of point, a 2D Delaunay triangulation is used. This solution leads to models with lots of redundant information and large size if all range points are used in the process and still need to be improved.
The fusion of range and intensity information used the technique presented in Camera Calibration and Texture mapping
Back to Top
Results
|
|
|
|
|
Snapshots of IEETA model |
Snapshots of single image of laboratory model |
Snapshots of single image of Deca model |
Back to Top
Discussion
When compared to commercial systems for 3D scanning, the proposed solution shows up some limitations namely in range and linear accuracy; indeed, some commercial systems have millimetre precisions, which can even be excessively high and not necessary to common navigation or modelling applications. On the other hand, the proposed device has a cost much lower than commercial solutions and uses a quite common range finder, which is almost an established industrial component, therefore easy to find and purchase. Moreover, the developed unit presents a few features and advantages not found in any commercial solutions, such as the large coverage of the environment in a single scan (180º x 270º) and its full versatility in adjusting tilt range and velocity.
As far as software and automatic reconstruction tasks are concerned, many issues have not been yet considered in this paper. The main one is related to the fact that all the results are presented with only a single range image. The first improvements to test extensively the 3D reconstruction capabilities of the sensor will consist in developing tools to register several range images and produce full models of real environments.
Many of the tools can also be improved, for example the triangulation algorithm used is very simple and produces large models. Processes to make the triangulation more efficient can be implemented resulting in models with a similar quality but a reduced number of triangles.
Range data must also undergo a minor correction procedure due to the fact that the tilt rotation axis does not cross exactly the centre of the laser emitter resulting in slight distortions for larger tilt angles, especially at further pan positions.
Independently of the needed improvements just mentioned, the proposed device has shown good capabilities for a fast low cost 3D perception of space usable in mobile robot autonomous navigation, and showed also enough accuracy to allow the reconstruction of real scenes when merged with intensity images.
Back to Top
Acknowledgments
This work was a co-operation between the "TEMA-Departamento de Engenharia Mecânica" responsible of the development of the 3D sensor, and the "IEETA-Departamento de Electrónica e telecomunicações" for the 3D reconstruction sofware.
Back to Top
References
|
Batavia, P. H.,
Roth, S. A., Singh, S. Autonomous
Coverage Operations In Semi-Structured Outdoor Environments, Proceedings of
the Intl. Conference on Intelligent Robots and Systems, EPFL, Lausanne,
Switzerland, October 2002.
|
|
|
[Dias03] |
Dias, P. Three dimensional Reconstruction of Real World
Scenes Using Laser and Intensity Data, PhD thesis,
University of Aveiro, September 2003.
|
|
El-Hakim, S. F., Brenner,
C., Roth, G. A multi-sensor
approach to creating accurate virtual environments, ISPRS Journal for Photogrammetry and
Remote Sensing, Vol. 53(6). pp. 379-391, December 1998.
|
|
|
Fischler, M. A., Bolles,
R.C. Random Sample
Consensus: a paradigm for model fitting with application to image analysis
and automated cartography, Communications of the ACM 24 (6), pp.
381-395.
|
|
|
Kurazume, R.,
Nishino, K., Zhang, Z., Ikeuchi, K. Simultaneous
2D images and 3D geometric model registration for texture mapping utilizing
reflectance attribute, Proceedings of 5th
Asian Conference on Computer Vision (ACCV). Vol. I, pp. 99-106, January 2002
|
|
|
Sequeira, V., Ng, K. , Wolfart, E.,
Gonçalves, J. G. M., Hogg,
D. Automated
Reconstruction of 3D Models from Real Environments, ISPRS Journal of Photogrammetry and Remote
Sensing (Elsevier). vol. 54, pp. 1-22, 1999.
|
|
|
[Smith99] |
Smith, P. W., Elstrom,
M. D. Stereo Based
Registration of Multi-Sensor Imagery for Enhanced Visualization of Remote
Environments, Proceedings of
the IEEE International Conference on Robotics and Automation, pp. 1948-1953,
May 1999. |
|
[Stamos00] |
Stamos, I., Allen, P. K. 3-D Model
Construction Using Range and Image Data, IEEE
International Conference on Computer Vision and Pattern Recognition, pp.
531-536 (volume I). South
Carolina, June 2000.
|
|
Surmann, H., Lingemann, K., Nüchter, A.,
Hertzberg, J. Fast acquiring
and analysis of three dimensional laser range data, 6th International Workshop on Vision, Modelling,
And Visualization, VMV 2001, pp. 59-66, Stuttgart, 21-23 Nov 2001, Germany.
|
|
|
[Tsai87] |
Tsai, R.Y. A versatile Camera
Calibration Technique for High-Accuracy 3D Machine Vision Metrology Using
Off-the-Shelf TV Cameras and Lenses, IEEE Journal of Robotics and automation,
Vol. RA-3, No. 4, pp. 323-344, August 1987. |
|
Willson, R. G. Modeling
and Calibration of Automated Zoom Lenses, PhD thesis, Department of
Electrical and Computer Engineering, Carnegie Mellon University, January
1994.
|
Author: Paulo Dias at IEETA/Universidade de Aveiro, Portugal - 07/07/2004