Research And Application Of Marine Movement Target Detection Method

Target detection as an important direction of computer vision,it is the follow-up target tracking and identification of an important basis,so the importance of target detection is self-evident.In recent years,with the rapid development of video technology,its application has been widely used in public places such as banks,hospitals,supermarkets and transportation.China has a vast territory,a long coastline and a vast sea area.The detection of maritime sports targets has important practical significance and application value.It plays a very good role in military affairs and civil affairs such as maritime rights protection,maritime regulation and marine environmental protection.Due to the complicated and varied environment of the sea scene,it poses a great challenge to the detection of the sea sport target.Based on the study of a large amount of literature and the research of predecessors in the detection of moving objects,this paper mainly does the following work:The basic concepts and algorithm flow of the three-frame difference method and the Gaussian mixture background model are introduced.The influence of the number of Gaussian distributions and the learning rate of background modeling on Gaussian background modeling is analyzed.Out of solution.(1)In the process of establishing Gaussian model,some Gaussian distributions decrease in priority,and do not contribute to the establishment of Gaussian background model.In the case of deleting Gaussian distribution,the article combines this strategy and proposes to add for each Gaussian distribution A survival time parameter,if the Gaussian distribution match,the survival time parameter is restored to its original value,if not match,the time parameter is reduced to 0,the Gaussian distribution to be deleted.This will accurately describe the actual scene state changes and reduce the computational complexity of the algorithm.(2)At the same time,aiming at the problem of fixed learning rate of Gaussian mixture background model,we use the inverse of the current frame rate in the initial stage of background modeling to accelerate the convergence of the background model.In the background stabilization stage,Rate is easy to introduce noise interference,requires a small learning rate to stabilize the background model.The article proposes to update the learning rate according to the reciprocal of the number of matched Gaussian distributions per pixel,so as to allocate a learning rate for each pixel,which can speed up the background model updating and stability.According to some difficulties existing in the detection of marine sports targets,this paper presents a detection algorithm combining three-frame difference method and Gaussian mixture background model.At the same time,in order to adapt to the changes of sea light,an adaptive threshold segmentation is set up.Then,the image is divided into background area,target area and interference area by using the block principle of image.Different background modeling methods and updating strategies are applied to different regions.Finally,the accuracy of this method in the detection of marine sports targets is verified by experiments.