A 1.111Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE
Early on the morning of May 9, 2012, Oregon State University–OSU’s Civil Engineering Geomatics Lab teamed up with a research team from the USGS to perform terrestrial scanning for ecological research for a small section (approximately 100m x 100m) of rocky coast at Cape Arago, Oregon. While it was a sunny day, it was also very windy. In fact, a nearby NOAA weather station recorded winds up to 30 mph that morning (strong winds, just below the high wind threshold of where you would not want to be out in the field with equipment). The tripod showed no signs of instability, but we had someone positioned immediately next to the scanner ready to catch it if an extreme gust of wind came up.
A key to this research was to utilize an extremely low tide window in order to capture features that are normally below water. Airborne scans completed previously on the coast were not completed at low tides. The water at this site is also too turbid for bathymetric mapping. Given the nature of the rocky terrain, which resulted in a substantial number of occlusions, the scanner was setup at 10 locations. Ultimately the data were processed into DEMs with cell sizes ranging from 0.05m to 1.0 m using the Bin and Grid process that I wrote about in a previous article.
Scans were geo-referenced using methodology outlined in Olsen et al. (2009 and 2011), which constrains the scan origins to GPS coordinates, levels the scans from level compensators and uses a least-squares azimuth adjustment. Rapid Static GPS coordinates were obtained for every scan position processed in OPUS (although since it is the coast, the points were 2km outside of the polygon, so the solutions were slightly extrapolated). The quality of alignment was assessed by evaluating the distances between 40,000 to 100,000 sample points in areas of overlap between scans. The 3D RMS difference values for the scan pairs (filtered to 100m range) averaged 3 cm (maximum 3D RMS 5 cm), which is reasonable given the use of Rapid Static GPS for geo-referencing. So at a project level, the wind did not appear to really affect the general data quality (similar scans on the coast typically have an average of 2-3 cm 3D RMS). However, there were some internal biases observed, which warrant further analysis.
It is important to note that this article is a preliminary assessment and not a complete scientific study, which would require substantially more test cases. (Not sure who would want to take on that research given the risk to the scanner!). However, that aside, there are some very interesting observations on the impact of wind on scan data quality from this limited dataset. The following sections will describe the tests completed.
Change analysis of the point cloud data
At the first scan position, we completed 2 repeat scans without moving the scanner so that we would have the needed data to enable us to quantify the influence of wind on the scan data. The two repeat scans were compared using an open source package CloudCompare, which has incredible change analysis tools for laser scan and mesh datasets. Figure 1 shows the comparison results between the two scans. One can see that generally the deviations are more pronounced with range, as would be expected. Also, the scanner collected data by rotating clockwise. Hence, at the start point for the scans (labeled in the plots), the deviations are lower, but as it continued through its 360 rotation, the deviations became more pronounced towards the end.
Figure 2 shows the histogram for the calculated differences between the sample points in the two aforementioned scans from the same position. Note that the mean deviation is 0.012 m with a standard deviation of 0.013 m. Ninety five percent of the points were below 0.05 m in deviation between the two scans. The majority of those points were the result of returns from ocean waves and from vegetation.
Analysis of inclination sensor readings
Inclination sensor readings (roll and pitch) were taken every second and are plotted in Figure 3 for both scans. In these plots, the 360 sweep represents the scanner orientation relative to its starting point. The value on the radial axis represents the inclination sensor reading. For comparative purposes, Figure 3 also includes roll and pitch measurements for an indoor scan that was stable and did not have wind effects. The effects of the wind are immediately observable, showing a large scatter in the windy scans compared to the average scan. Silvia and Olsen (2012) discuss more about the utility of inclination sensorslevel compensators in scanners and reliability of values.
In static scanning, however, the average inclination sensor readings are used to determine the leveling for each scan as a whole rather than on a point by point basis as is done with an IMU in kinematic scanning. Table 1 shows summary statistics for the windy scans compared to a stable scan. Note that since the stable scan was not done from the same setup, the actual inclination sensor values are not comparable. However, the standard deviation and range would be directly comparable between the windy and stable scans. Note that the range of values (up to 2 for the windy scans) is much larger than the stable scans (0.3). The standard deviations are also 3 to 4 times higher in the wind. The actual average inclination sensor readings were very similar between the two windy scans (within 0.015) from the same location.
Quality evaluation of GPS data
OPUS provides statistics on the quality of the GPS measurements. For a detailed discussion of what these values mean, the reader is referred to the OPUS website. Overall, these values are generally good for OPUS rapid static GPS observations (see Table 2), indicating that the GPS data still produced good results in spite of the wind.
Concluding Remarks
While the wind did not make the data unusable for this project (5cm 3D RMS error was more than adequate for the project needs, particularly since the data were averaged and filtered for the final model), it can have a significant impact on the data quality depending on your application. This is one of the exciting aspects of scan data in that you can actually "see"
and analyze the influence of such effects, unlike a lot of other techniques.
If you are wondering why we didn’t wait for a non-windy day the answer is that the wind is like this nearly every day at this location and we had to coordinate with the tide. In general, however, I wouldn’t recommend scanning in this kind of wind!
Acknowledgements:
Special Thanks to Mahyar Sharifi-Mood (OSU) and Jeff Hollenbeck (USGS) for braving the cold wind and helping with the scans.
Michael Olsen is an Assistant Professor of Geomatics in the School of Civil and Construction Engineering at Oregon State University. He currently is a member of the ASCE Geomatics Division Executive Committee and is the Associate Editor for the ASCE Journal of Surveying Engineering. He earned BS and MS degrees from the University of Utah and a Ph.D. from the University of California, San Diego.
References
Olsen, M.J., Johnstone, E., Kuester, F., Ashford, S.A., and Driscoll, N. (2011). New automated point-cloud alignment for ground based LIDAR data of long coastal sections, ASCE Journal of Surveying Engineering, 137(1), 14-25.
Olsen, M.J., Johnstone, E., Driscoll, N., Ashford, S.A., and Kuester, F. (2009). Terrestrial laser scanning of extended cliff sections in dynamic environments: a parameter analysis, ASCE Journal of Surveying Engineering, 135(4), p. 161-169.
Olsen, M.J (2011). Putting the pieces together Laser scan geo-referencing, LIDAR News eMagazine, 1(2).
Silvia, E.P., & Olsen M.J. (2012). "To Level or Not to Level: laser scan inclination sensor evaluation," Journal of Surveying Engineering, 138(3), 117-125.
A 1.111Mb PDF of this article as it appeared in the magazine complete with images is available by clicking HERE