Research On Nearshore Image Processing Method For Unmanned Surface Vessel Vision Task

With the increasing demand for resources of human society,the development of marine resources will become the trend of future development.Unmanned surface vessels(USV)are widely used in military and civil fields in the competition for marine resources because of their advantages of low cost,strong environmental adaptability and low labor cost.The vision system,as an important branch of environmental perception of USV,is the main way for USV to acquire peripheral information.At the same time,the information it obtains is also an important reference for the decision-making of USV.However,the special environment of water surface such as turbulence and fog will result in the low quality of image and video information.The unprocessed original images have the disadvantage of a low proportion of useful information,which can seriously affects the performance of the USV vision system.Considering the special environment of water surface,this dissertation puts forward some preprocessing methods for the USV nearshore images.By improving the quality of images and videos,as well as extracting useful information in the nearshore images,the performance of visual system is improved.The main research contents and innovations can be concluded as follows:In order to reduce the shakiness of image sequence caused by turbulence,this dissertation proposes a mechanical stabilization method based on PTZ platform control and an electronic stabilization method based on multilayer gray projection(MGP)to form a two-stage video stabilization system.In the mechanical stabilization,in order to deal with the defects of great inertia and lagging response of PTZ platform,a method to control the PTZ platform with different strategies in variant regions is proposed.After that the shakiness of camera is reduced effectively.In the electronic stabilization,the differential gray projection,ring projection and circular projection are adopted to estimate the translation,scaling and rotation separately.The proposed electronic stabilization method has two main advantages: one is to use statistical information to estimate the transformation between adjacent frames,which can resist the interference caused by meaningless texture,and has high stability for the water surface and nearshore videos;The other is that the transformation between adjacent frames can be estimated with only three times of projection,and the calculation is extremely fast.The proposed method has high speed,high precision and robustness.Compared with other current stabilization methods,the proposed method is more suitable for water and nearshore environment.In order to solve the disturbance of image quality caused by the fog on the water surface in the solar environment,a nearshore image dehazing method based on multi scattering model is proposed.The wet circumstance of the water surface has led to the fact that these areas are easier to have fog than the land areas,sometimes there still exists fog on the water surface even when it is a sunny day and it can influence the contrast and visibility of images.The physical model of traditional ideas cannot describe the fog environment which has the sun in the sky,so the effect of dehazing is not good.A new physical model is proposed to describe the special environment and applied in dehazing.The highlights of the proposed method can be summarized into two points: the first point is that the concept of atmospheric light map is proposed firstly,and the multi scattering model is adopted to estimate the atmospheric light map.For hazy images with the presence of sun in the sky,the atmospheric light map can better describe the distribution of the atmospheric light than the single atmospheric light value;The second point is that a precise refined transmission map estimation method based on gradient selection is proposed in this dissertation,which can greatly shorten the estimation time on the premise that the estimation performance is very close to the guided filtering method.The proposed method has a good dehazing effect for the hazy images with the presence of sun in the sky.The generated haze-free images are more natural than the results of the existing methods,and the contrast and visibility are enhanced more obviously.Finally,in order to solve the problem that the low proportion of meaningful information caused by the large areas of meaningless textures on both the water surface and the bank in the nearshore images of USV,a multilayer edge detection framework is proposed to detect the scene border for these nearshore images.Firstly,a multilayer edge detection method is designed for the special water surface environment.Next,the discontinuous edges in the detected results are linked based on the Gestalt law of good continuation,the distance constraint and the direction constraint are proposed to realize the law of good continuation.The discontinuous edges are linked effectively and the complete scene border is acquired.The proposed method is less affected by meaningless textures than the other edge and contour detection methods.The detected scene border is also more complete than the results of the other methods.At last,the proposed method is a fast and effective scene border detection method for nearshore images.In this dissertation,the preprocessing methods for USV nearshore images are studied,which not only optimizes the quality of the USV nearshore images and videos,but also provides a guarantee for the high-level task of USV.