M3C2-EP: Error propagation for topographic 3D change detection
M3C2-EP is a change detection method for 3D point clouds. It extends the well-established M3C2 algorithm by Lague et al. (2013) by introducing error propagation for statistical signal-noise separation.
The separation of change signal from noise is crucial for topographic change detection. The amount of change that needs to happen before it can be reliably quantified (
Level of Detection) depends on different factors, including:
- Which sensors were used to measure the topography (sensor configuration, sensor uncertainties)
- How the data is processed (alignment of bi-temporal datasets)
- What the properties of the surface are (material and geometry)
Knowledge on these factors allows making statistically sound statements about change that is actually happening. M3C2-EP provides a framework to include this knowledge in bi-temporal point cloud change detection.
Process of M3C2-EP from data acquisition to final product.
Benefits of M3C2-EP:
- Make full use of the available data, adding more data will always improve the results (possibly more changes of lower magnitude)
- Reliable detection of smaller changes especially for objects that appear rough at the scale of analysis
- Comparison of data from different sources possible while making use of the individual advantages (e.g. ALS vs. TLS)
In M3C2-EP, we rely on the knowledge of the sensor system, the measurement process and the data preparation to propagate errors in distance and angular measurements as well as in the alignment of the bi-temporal datasets. This allows an alternative quantification of uncertainty without assuming locally planar objects.
The concept of M3C2-EP can be implemented for point cloud data from multiple sources:
- Photogrammetric point clouds often come with associated positional uncertainties
- Airborne/UAV-borne Laser Scanning uncertainties can be included when the trajectory is known
- M3C2-EP can also be used in combination with M3C2, e.g. when the required metadata is not available for one of the two epochs
When using terrestrial or airborne laser scanning systems, the uncertainty associated with individual measurements is often provided by the sensor manufacturer or can be estimated under lab conditions. The uncertainty introduced by the alignment can be quantified by analysing stable areas.
In addition to statistical testing, the uncertainties may be used for further analysis, e.g. volume derivation. Accepting that change values come with uncertainties also raises consciousness on the reliability of binary change maps, which highly depend on a user-provided level of significance.
M3C2-EP is featured in the following scientific publications:
- Winiwarter, L., Anders, K., Höfle, B. (2021): M3C2-EP: Pushing the limits of 3D topographic point cloud change detection by error propagation. ISPRS Journal of Photogrammetry and Remote Sensing. (forthcoming)
- Winiwarter, L., Anders, K., & Höfle, B. (2020): Herausforderungen in der Fehlerfortpflanzung von Laserscandaten für multitemporale Analysen zur verbesserten Quantifizierung des Level of Detection. 40. Wissenschaftlich-Technische Jahrestagung der DGPF. Vol. 29, pp. 373-380.
- Winiwarter, L., Anders, K., Wujanz, D., & Höfle, B. (2020): Influence of Ranging Uncertainty of Terrestrial Laser Scanning on Change Detection in Topographic 3D Point Clouds. ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences. Vol. 2-2020, pp. 789–796.