Research On Water Surface Target Detection Technology Based On Vision

Unmanned surface ship senses the surrounding environment in real-time in order to perform tasks autonomously,and avoid obstacles by detecting obstacles in the foreground field of view.Vision sensors have the advantages of low cost and high resolution.Therefore,the visual sensor is adopted to obtain dynamic environment information around the unmanned ship,and detect water surface target based on image processing technology.The research on water surface target detection technology has important practical significance for autonomous navigation of the unmanned ship.Aiming at the problem of low accuracy and poor real-time performance of surface target detection,this paper designs a target detection method based on water-sky line.The method mainly studies three parts:First,the improved line segment detection algorithm is designed.The algorithm firstly detects the straight line by using line segment detection,then introduces length and angle parameter characteristics of water-sky line to remove straight line which does not meet requirements,so as to detect the real water-sky line quickly and accurately;The second is to design an defogging algorithm based on the physical model of foggy imaging,which first segment the sky area by means of water-sky line,accelerates the estimation of atmospheric light by using the quadtree subdivision method,then uses the contrast priori to design cost function,and estimates the transmittance with the guided filter,and finally the physical model of foggy imaging is used to achieve fogless image and improve the contrast of image;Third,the surface target detection algorithm based on optical flow clustering is designed,which first uses water-sky line to segment the potential region of target,to eliminate the noise interference outside the region,and then in the potential region,because optical flow of different targets is inconsistent,the optical flow vector of feature points is calculated by using improved LK algorithm,and DBSCAN clustering algorithm of grid division is used to cluster the amplitude and angle characteristics of optical flow vector to improve the efficiency of clustering and eliminate the interference of some water wave noise,and finally geometric constraints are used to further eliminate false targets.The proposed method of detecting water surface target is verified by experiments,and the method is compared with Gaussian mixture model and ViBe algorithm.It can be seen from the experimental results that the accuracy of the target detection method designed in this paper is higher than other two methods.Although it is not as good as ViBe algorithm in real-time,it has been improved compared with Gaussian mixture model.