Image Processing Tools for Semi-automatic Minefield Detection

Airborne surveys have proved to be a very promising way to detect the boundaries of minefields, a requisite to any mine clearance procedure. Human photo-interpreters are, to date, unchallenged in the analysis of these images. Unfortunately in order to survey all suspected areas of a country and have the resolution needed to detect relevant minefield indicators, a tremendous amount of data is to be collected. This makes the analysis a very time-consuming task. This paper shows how image-processing tools can be used to speed up the work of photo-interpreters by detecting regions of interest or making some repetitive tasks automatic. This leads to a semi-automatic procedure where image processing is used both as a help to the detection of relevant indicators and a way to enhance some given parts of the images. Based on a real campaign performed first in a test site in Belgium and then on a real mine-infested country, namely Mozambique, in the scope of a European project co-ordinated by ITC, the work presented here involves two steps. First, some indicators have been selected for their relevances in minefield detection as well as for their easiness to be automatically detected on the available images. Then detectors devoted to each of these indicators have been implemented. This paper describes some examples of these detectors. Some specifications are also given to ensure the success of such a survey from an image processing point of view. They concern the constraints that minefield indicators and indicator detectors must fulfil in order to be able to be used in such a semi-automatic procedure. Other constraints about the images and the way to collect them to improve the detection of indicators are also given. In addition, some results about compression on minefield images are presented.