Inland Moving Ships Detection via Compressive Sensing and Saliency Detection

This paper presents an effective inland moving ships' detection method fused with compressive sensing and saliency detection to attack the challenge: when the ships detected suffer serious cavities due to their large size, relatively low speed, and uniform color. The background is composed of a K-SVD dictionary and a mean set of K-SVD coefficients associated to each pixel. To address the problem that the background and difference image are corrupted by the movement traces when ships sail into the first frame, the logical bitwise AND is performed between difference image and saliency map to get the exact result. Due to the use of K-SVD coefficients, the background is blurry. Then background update strategy is put forward to eliminate the movement traces and make the background more clearly. Finally, both qualitative and quantitative evaluations on several challenging inland video sequences demonstrate that the proposed algorithm outperforms several state-of-the-art methods in terms of efficiency and accuracy.