Pattern recognition in aerial wartime images and laser-scanning data sets

Projektpartner:

Landesamt für Geoinformation und Landesvermessung Niedersachsen (LGLN)

Motivation

The analysis of aerial wartime images regarding bellicose impacts, especially the evaluation of the hazards stemming from duds, is a central responsibility of the Kampfmittelbeseitigungsdienst (KBD). In Lower Saxony, the KBD is a department of the regional directorate Hamelin-Hanover of the Landesamt für Geoinformation und Landesvermessung Niedersachsen (LGLN). Currently, the analysis of the aerial wartime images is a manual task, which is highly time-consuming, despite a sole focus on selected areas. For many questions, a regional approach with information regarding the general occurrence of bellicose impacts in the form of “impact maps” would be sufficient. This information can be generated if a large percentage of all impacts can be detected with high reliability. For a cost-efficient generation of such maps, an automatic recognition of clues for bellicose impacts in wartime images, especially bomb craters, is essential. Furthermore, laser-scanning data should be incorporated aside the aerial images to derive information on impacts in forested areas.

Goal

The goal of this project is the development of a method for the automated detection of bomb craters in aerial wartime images and laser-scanning data sets. Craters often exhibit a regular shape, allowing for their detection using different image analysis methods like correlation techniques. By maximising the cross-correlation-coefficient, bomb craters can be found automatically. However, normally, these craters are of different sizes and image scales vary between acquisitions. In this project, marked point processes are used that can handle these differences between the images. The advantage of this technique is the concatenation of a strong object model with a stochastic approach to find the optimal configuration of the objects in the scene by minimising a global energy function, whose optimisation is achieved by combining Reversible Jump Markov Chain Monte Carlo Sampling and Simulated Annealing.

The aerial images and laser-scanning data sets are acquired from the LGLN and have the foci on rural and forested areas, respectively. Statements on the bellicose impacts on certain areas can be drawn and represented in impact maps. For this purpose a probability map is defined which is created from the automatic detections via kernel density estimation. By setting a threshold, areas around the detections are classified as contaminated and uncontaminated sites, respectively (Figure 1).

Publications

Kruse, C.; Neuberger, H.; Rottensteiner, F.; Hoberg, T.; Ziems, M.; Huth, J.; Heipke, C. (2017): Automatische Detektion von Bombenkratern in Kriegsluftbildern mittels markierter Punktprozesse. In: DGPF Tagungsband 26/2017, Würzburg, März 2017, pp. 245-261

Kruse, C.; Rottensteiner, F.; Hoberg, T.; Ziems, M.; Rebke, J.; Heipke, C. (2018): Generating impact maps from automatically detected bomb craters in aerial wartime images using marked point processes. In: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences IV-3, pp. 127-134.

Clermont, D.; Kruse, C.; Rottensteiner, F.; Heipke, C. (2019): Supervised detection of bomb craters in historical aerial images using convolutional neural networks. In: International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences XLII-2/W16, pp. 67–74.
DOI: 10.5194/isprs-archives-XLII-2-W16-67-2019

Kruse, C.; Rottensteiner, F.; Heipke, C. (2019): Marked point processes for the automatic detection of bomb craters in aerial wartime images. In: International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences XLII-2/W13, pp. 51-60.
DOI: 10.5194/isprs-archives-XLII-2-W13-51-2019

Kruse, C.; Rottensteiner, F.; Heipke, C. (2020): Using redundant information from multiple aerial images for the detection of bomb craters based on marked point processes. In: ISPRS Annals of the Photogrammetry, Remote Sensing and Spatial Information Sciences V-2, pp. 861–870.
DOI: 10.5194/isprs-annals-V-2-2020-861-2020