The Research Of Ship Detection System Applied In Restricted Water Area Based On Improved Gaussian Mixture Model

With the rapid development of the national economy, the demand for water transport and railway transport is increasingly needed. We performed more and more bridge construction in the river. What is more, the rapid development of railway transport has made a great contribution to the increment of railway bridge. Meanwhile, the transport ship in terms of quantity, or in the tonnage, are also in constant rise. However, in recent years, the ship-bridge collision accidents are more frequent, which causes enormous property damage. In order to minimize the loss of the ship-bridge collision, pier crash facility has been applied, protecting the pier and the ship when the accident occurred in a certain degree. However, if effective measures to remind the ship to change direction in time are made before the accident, this method of early warning has a more important significance.We use image sequence to surveillance sailing ships, which belongs to the category of video surveillance and detection of moving objects. The method is mainly divided into two steps:first, with the establishment of an effective background model, we can detect and track the moving target by frame subtraction. Second, image segmentation and information extracting will be performed, which provide a basis for the subsequent calculation process. However, the application scene of this paper is the natural environment, including uncertain background like water wave, changing weather, day- night changing and so on, which Increase the difficulty of real-time surveillance.In this paper, we studied several common moving object detection methods, at the same time, we analyzed their advantages and disadvantages. Also, we studied the theoretical basis of Gaussian mixture model, and explored the effects of update rate a and the number of composition K in Gaussian model to the modeling with the cooperation of experiments. This paper proposed an improved Gaussian mixture model that self-adaptive update rate according to the limitation and the specific application scenarios. First, we choose the even image of the first 30 images in the video sequence instead of the beginning image to initialize the background model. We start modeling with large update rate and lower the rate with the number of images increasing, so that the speed of modeling is improved. In the moving detection period, we will judge the previous result to decide whether to change update rate, as a result, we can improve the effect of "Ghost Shadows". Then, we made specific work aiming to sudden illumination and following image processing, with this method we can improve the detection efficiency of steamship outline that exceeds the limit. This paper also introduces the construction of hardware system and realizes the algorithm with OpenCV based on VS2010. Our results achieve the expected requirements.