3P6-5 A Configuration-Conjunct Threshold Segmentation Method of Underwater Linear Object Detection for Forward-Looking Sonar

Lixin Liu,Hongyu Bian,Shinichi Yagi,Xiaodong Yang

2015-01-01

Abstract:Underwater operation, such as objects detection, obstacle avoidance, and target tracking, using forward-looking sonar have been widely applied because of the relative high frame rate and resolution. Image segmentation, as a basis of automatic visual understanding, contributes to underwater linear object detection such as suspending tubes, sustaining pipes, petroleum ducts, as well as parts of autonomous underwater vehicles, and so on. However, grating-lobe and multi-path disturb sonar imaging, and they might cause over-segmentation easily because conventional threshold segmentation methods search the segmentation threshold mainly based on gray level distribution without considering the structural prior probability. In another word, the grating-lobe and multi-path disturbance that with high similarity with the true object, are quite difficult to get rid of by de-noising methods and they intrude the gray level distribution which result in fuzziness in sonar images understanding. To seek for a more robust segmentation method and verify if it is possible to guide the segmentation threshold evolvement of forward-looking sonar images using configuration prior probability, a configuration-conjunct threshold segmentation method of underwater linear object detection is discussed. A series of segmentation experiments are implemented. Experimental results are compared, including the segmentation output and computational cost using different methods, which demonstrate the effectiveness of the proposed method.