Apport de la classification automatique d’images sonar pour le recalage à long terme

This issue handles the ability of using segmentation results of sidescan sonar images for long term registration. This study take a part of AUV (Autonomous Underwater Vehicle) navigation problems, particularly to correct the drift of navigation sensors. Principle of images formation with this type of engines and the main properties are first reminded. Some images are shown, which allows us to understand difficulties of this type of data to realise AUV positioning. Then, we decide that segmentation results of sidescan images provide us landscapes on which registration will be able. Segmentation is then explained. It is supervised type on five classes, rocks, ripples, sand, mud and shadow. Using Gabor filters provides classifying parameters and classification is realised by the nearest neighbour. This is made step by step, refining step by step the segmentation. In order to know which landscapes may be extracted to enable the positioning, a statement is then realised on the obtained results, The registration methodology is then quickly explained and several results are commented. This allows us to realise a final conclusion on the ability to use results of segmentation of sonar images to make registration and to give possibilities and limits of this type of positioning system.