Small Target Detection Of Ship Based On Omnidirectional Vision

Small target detection is an important issue in computer vision domain, which was widely applied in the environmental monitoring field. It is conducive to reducing equipment and the budget expenditure that apply omni-directional vision to marine monitoring domain,because omni-directional vision has the advantages of 360° wild field and abundant information.The method of small ship target detection in the case of ocean background based on omni-directional vision is studied in this paper. The so-called small target is that the ship target accounts for only a few dozen or even several pixels in the image due to far imaging distances.The main task of this paper is to detect ship target as far as possible,and the basic idea is that extract sea-sky-line on the basis of image preprocessing, then detect whether there is small ship target around sea-sky-line,therefore,the core contents of this paper are preprocessing, sea-sky-line extraction and small target detection.Firstly, design the preprocessing algorithm on the basis of analyzing the features of omni-directional image. Image denoising methods-mean filter, middle filter and Gauss filter are researched, and the combination of nmiddle filter and Gauss filter is chosen to complete the denoising by comparing experimental results; then image sharpening methods——grads sharpening and laplacian sharpening are researched, and laplacian sharpening is selected by comparing experimental results; traditional Canny edge detection algorithm and self-adaption threshold Canny edge detection algorithm are also researched and realized by program, then according to the defect of the above edge detection algorithm, blocking self-adaption threshold Canny edge detection algorithm is proposed which overcomes deficiencies of traditional Canny and self-adaption threshold Canny when the image brightness is not balance, and experimental results vertify the effectiveness and superiority of designed blocking self-adaption threshold Canny edge detection algorithm.Secondly, methods of sea-sky-line extraction are researched based on omni-directional vision. The methods of sea-sky-line extraction can be divided into two classes: the one is that extract sea-sky-line on cylinder image of omni-directional image and the other is to extract circle sea-sky-line on omni-directional image directly. In accordance with the former, the method of sea-sky-line extraction based on dynamic programming is designed in this paper,which is done on cylinder image of omni-directional image, and it is confirmed that this method is effective under the circumstance that the breaking gap of sea-sky-line is not large.According to the latter, this paper proposes two algorithms——the algorithm of sea-sky-line extraction based on outermost edges search and the algorithm of sea-sky-line extraction based on the longest curve search, and the effectiveness and superiority of these two algorithms are proved by plenty of experimental results.Finally, the method of small ship target detection based on omni-directional vision is researched. Research on the existing methods of small target detection on visible image-frequency tuning and lifting wavelet. These two methods are great different from the ideas of this paper, so they can not be applied to small target detection suitably in this paper.Therefore,this paper presents the simple and practicable method of small ship target detection on omni-directional image based on single-window threshold, which includes the method of small target detection in single image and the method of small target confirming in sequential images. Experimental results show that small ship target which is protrusion of sea-sky-line can be detected by the method of small target detection in single image based on single-window threshold, and the only fly in the ointment is false drop; with the addition of the method of small target confirming in sequential images,the success rate is improved markedly, and the effectiveness and superiority is proved at the same time.