Research On Three-dimensional Reconstruction Method For Underwater Scenes Based On Forward-looking Sonar

Seventy percent of earth's surface is covered by ocean containing abundant mineral resources,energy sources and biological resources,which has high development value and attracts great attention of the world.Sea detection technology as well as space technology becomes the focus of cutting-edge science and technology competition together.Sonar is the important underwater detection equipment and is the only visual perception tool with the long distance detection ability,equipped for long time by carriers liking ships,underwater robot,manned submersible,et al.Conventional two dimensional sonar images containing fewer information lacking high information,which have no intuitive expression,have great different from customary comprehension of perception for environment and are adverse for operators and underwater equipment cognize underwater environment.Three dimensional reconstruction technology is one kind of image processing methods,recovering feature information from two dimensional images to three dimensional space.It is a mature technology in optimal image field,but conventional optimal three reconstruction methods can't be transplant to acoustic image reconstruction directly because there is qualitative difference between optical camera's and sonar equipment's mapping theory.Studying special methods of sonar image three dimensional reconstruction was started latterly,which has not given a desired solution.One kind of three dimensional reconstruction method reconstructing sonar image feature points is described in this paper,in which corners in object's contours are regarded as to be reconstructed points.Reducing noise interference to guarantee validity of extracted contours,a novel method of image denoising and enhancement based markov random field and guided filtering is designed.Firstly,MRF is used to conduct pre-segmentation;then,median filtering arithmetic processes primal sonar image simply;at last,this resulting image is regarded as guided image to filtering the sonar image segmented by MRF,which realizes image de-noising and enhancing.The proposed method could remove noise in background and shadow effectively,restrains the noise within object,eliminate fake contour created by MRF,and has the ability of edge preservation and enhancing.For the reasons that sonar images lack high frequency information and most of features concentrate on edges of objects,corners on edges are selected for the object of sparse reconstruction,and filter method of contour and extracting corners method are built for extracting corners accurately.Studying mapping theory of sonar image and doing theoretical derivation and analysis for corresponding geometric relations between three dimensional space and two dimensional plane promote a three reconstruction theoretical method aiming at reconstructing sonar image feature points.A three-dimensional reconstruction method searching height feature in different segmental arc of every layer is built,which could realize reconstruction under the condition of knowing rotation and translation parameters.Corresponding to situations that do not know those parameters,particle swarm optimization algorithm is used to estimate parameters via using few feature points.Then more feature points are reconstructed subsequently by estimated parameters.In the end of this paper,the precision is improved drastically by appending rotation and translation parameters which are estimated by sensors and have some errors.