Inland ship detection based on dynamic group sparsity

Effective and efficient ship detection is a prerequisite in inland video surveillance, e.g. to track the target ship from frame to frame and to analyze the tracked ship to recognize its behavior. In this paper, a real-time inland ship detection algorithm called MEADGS is proposed, which is based on dynamic group sparsity and background subtraction method. The background and foreground image are reconstructed directly based on enhanced adaptive dynamic group sparsity (EAdaDGS) algorithm. The background dictionary is updated in real time by the current background image when current frame is a key frame. Meanwhile, the efficiency of our proposed method is further improved by multi-resolution detection procedure. The ships are located in low resolution, and detected accurately in the specified area of high resolution. The experimental results show that our proposed method obtains higher-quality, with a 7.6% margin over the second best method performance.