Research On Weak Small Target Detection On The Sea Based On Panoramic Vision

There is a long coastline and vast maritime territory in our country. Oceans rights enforcement and maritime search and rescue missions are very important. It is significant to carry out sea targets detection study and develop advanced monitoring equipments. In remote place on the sea, targets with large size can only has several pixels in the image, and detecting targets in this case is called small targets detection. Panoramic vision system has wide application prospects in maritime surveillance applications for its advantages of unified imaging, 360-degree view and rotation invariant. But it has weak support in techniques of target detection. So this paper carried out the study on weak small targets detection on the sea based on panoramic image.With the complexity of sea-sky background and the specificity of panoramic image, this paper carried out the small target detection on the sea based on panoramic image. In the process of detecting weak small targets, firstly the panoramic image is processed by haze removal method, secondly sea-sky line detection method is used to detect the circle sea-sky line, finally targets detection method is used to detect weak small targets around sea-sky line.So the paper has three core contents, which are image haze removal, sea-sky line detection and weak small targets detection.Firstly, image haze removal algorithm was studied. Single image haze removal method using dark channel prior was introduced to eliminate the effect of sea fog. The derivation of the method combines Dichromatic Atmospheric Scattering Model with dark channel prior. To eliminate the halos and block artifacts, guided image filter was introduced. The haze removal algorithm was proved effective in defogging of fog panoramic image.Secondly, circle sea-sky line detection of panoramic image was studied. Hough circle transformation based on gradient and sea-sky detection based on ellipse fitting were introduced to detect the sea-sky line in the panoramic image. But Hough circle transformation based on gradient failed in detection in result of the complexity of the image, and the second method can not work well in case that the sea-sky line is not complete. To detect the sea-sky line precisely, two methods were presented, which were sea-sky line extraction based on modified Hough circle transformation and sea-sky line extraction based on directions of the gradient.Thirdly, weak small targets detection based on sea-sky line extraction was studied. To assure the speed of the method, targets are detected around the sea-sky line. Two methods were proposed to assure the accuracy of targets detection, which were weak small targets detection based on local complexity of the image and weak small targets detection based on gradient threshold. Aforementioned algorithms were tested in weak small targets detection in panoramic image and compared with small targets detection based on single window threshold in experiments.Finally,the advantages and disadvantages are analyzed of old methods and methods proposed in this paper. The scheme of weak small targets detection system based on panoramic image was designed and weak small targets detection system was build based on Visual C++6.0 platform. Algorithms proposed in this paper were proved effective and superior through amounts of experiments and statistical analysis on different conditions.