Object Detection Based on Saliency and Sea-Sky Line for USV Vision

This paper proposes a fast and effective object detection approach for a monocular vision system aboard an unmanned surface vehicle (USV) under the ocean environment. By utilizing the improved saliency detection and the sea-sky line (SSL) as well as the way of centroid-judging, the object detection method is constructed for an USV on the ocean environment. The improved frequency tune (FT) saliency detection algorithm is proposed to capture the salient objects. Moreover, based on analysing the prior information for the attitudes of USV, Hough transform and Mahalanobis distance are adopted to identify the SSL, which is implemented in a parallel programming. The SSL is used to filter out the clouds, meanwhile the irregular background noises from sea surface are excluded by judging the number of centroids of potential objects below the SSL among the randomly chosen history frames. The proposed approach is applied on our 'Jiuhang490' USV's video, and results indicate its real-time performance and accuracy.