Change in the Point Cloud

Change detectiondeformation analysis is a topic of great interest when working with LIDAR and other remote sensing datasets. For the sake of clarity, change detection is referred to as solely figuring out if and where there has been change, or if things are the same and is not concerned with the magnitude, or how much change has taken place. Deformation analyses determine the magnitude of change.

LIDAR provides data at an exceptional resolution, which coupled with rapid and growing acquisition capabilities, enables unprecedented detail in change analyses. Change detectiondeformation analysis is applicable in a wide variety of applications including geomorphology, structural monitoring, and construction progress updates. Just as wide as the applications and scales of interest, there are a variety of approaches in order to complete change detection or deformation analysis. The appropriate technique and algorithm usually depends on the applications, scales, and objects of interest.

In this first article I will provide a quick overview and discussion. I will use future articles to get into more detail since there are a lot of interesting considerations when you look into the algorithms. In the meantime, for a review of types of change detection and deformation analysis techniques available, I refer you to the work of Lague et al. (2013) as well as a paper of mine that was just accepted in the ASCE Journal of Computing in Civil Engineering (JCCE).

Essentially, the techniques can be broken down into 3 main categories:

1. Model to model. The most basic version of this is comparing elevation differences between two DEMs. However, more complex variants exist that can incorporate any simple geometric primitives and combinations of model features.

2. Point to model. Here a 3D model exists as the comparative surface and a point cloud is acquired to determine the amount of change. These can often be used to validate as-built documentation (in the form of 3D or vector models) or for quality control to compare construction progress to 3D models created in design.

3. Point to Point. Here a point cloud is compared directly to another point cloud. It has the advantage of not requiring modeling (which can be a tedious process sometimes, particularly in complex scenes where there is a lot of noise in the data). However, it has the disadvantage in that the algorithm has to do a lot more heavy lifting with larger datasets as well as determining how things relate.

With each of these categories there are a lot of algorithms that organize the data for speed and include various metrics for accuracy. As an example, I would like to put in a plug for Cloud Compare, which is an open source software package that has very powerful, quick deformation analysis techniques. It provides great visual results. My students and I have found it to be a great resource and has several tools in addition to cloud comparison.

In a future article, I will provide a detailed review of Cloud Compare. There are also several other commercial software packages that do a good job as well. The Point Cloud Library also has some powerful change analysis capabilities, some of which can be implemented quickly for small windows in robotics scenes.

Traditionally, change detection is done in the office after the data is collected, geo-referenced (or registered to a local coordinate system), and processed. The aforementioned paper in the ASCE Journal of Computing in Civil Engineering presents a paradigm shift for performing in-situ change analysis compared to traditional post-processing methodologies for potentially improved site investigations. It also provides an efficient workflow (combining acquisition, geo-referencing, and change analysis) to perform these in-situ analysis, and describes a new algorithm to perform rapid change analysis on point clouds. I will put in a disclaimer here that the algorithm works well for its intended purpose and task, but other algorithms may work better for other applications and scales of interest. My approach is focused on Civil Engineering and geologic applications and has been customized for those scales and change levels of interest.

I believe that performing change detection in the field can offer several significant advantages to current post-processing workflows. First, the augmented reality provided by in-situ change detection enables field crews and researchers to see immediate results, in situ, so that they are able to make key observations while present at the site, instead of being reliant on their personal memories or notes. If your memory is as bad as mine, it can be very hard to distinguish what has changed since you last visited a site. Plus humans always interject bias (so do machines programmed by humansJ).

Second, it can improve the overall efficiency of the survey. When this information is available to the operator during field data acquisition, areas of minimal change can be quickly surveyed at coarser resolutions and areas of substantial change can be scanned at higher resolutions. This can also translate into reduced processing time and data maintenance, which are currently significant hurdles for analyzing point clouds. Finally, this method provides immediate validation and quality control of the laser scan and real-time kinematic (RTK) GPS data being collected, leading to more confidence in the acquired data and allowing any issues to be resolved directly in the field.

So while there are a lot of great resources, there remain several critical issues in change detection and deformation analysis:

1. Identifying change of interest. For example, in a geologic study, one would want to screen change from something passing in front of the scanner in only one of the scans to focus on change of the object of interest.

2. Filtering out scanner noise. Many times, this has to be done manually.

3. Determining the level of change actually detectable versus what is error from registration, systematic error, etc.

4. Handling gaps and occlusions in the data.

5. Statistical quantification. Most approaches provide visual feedback, which is helpful, but may not provide rigorous statistics. Often, items 1-4 need to be resolved for the statistics to be trustworthy.

I am excited to see future research that continues to develop in this arena in tackling these large problems.

References:
Lague, D., Brodn, N., and Leroux, J. (2013). Accurate 3D comparison of complex topography with terrestrial laser scanner: application to the Rangitikei canyon (N-Z). http://arxiv.org/abs/1302.1183

Cloud Compare
Powerful package to quickly compare point clouds, surfaces, etc.

http://www.danielgm.net/cc/

Point Cloud Library

http://pointclouds.org/

About the Author

Michael Olsen

Michael Olsen ... Michael is an Assistant Professor of Geomatics in the School of Civil and Construction Engineering at Oregon State University. He chairs the ASCE Geomatics Spatial Data Applications Committee and is on the editorial board for the ASCE Journal of Surveying Engineering. He has BS and MS degrees in Civil Engineering from the University of Utah and a Ph.D. from the University of California, San Diego. He has also worked as an Engineer in Training for West Valley City. His current areas of research include terrestrial laser scanning, remote sensing, GIS, geotechnical engineering, earthquake engineering, hazard mitigation, and 3D visualization. He teaches geomatics and geotechnical engineering courses at OSU where he has developed new, ground-breaking courses in Digital Terrain Modeling course and Building Information Modeling. Recent projects he has been involved with include: earthquake reconnaissance (following the American Samoa and Chile earthquakes and tsunamis), landslide analysis for the US 20 realignment, seacliff erosion mapping using LIDAR for San Diego County and Oregon, liquefaction hazard mapping for Utah, and modeling and studying historical buildings such as the Palazzo Medici and Palazzo Vecchio in Florence, Italy.
Contact Michael Article List Below