The Underwater Target Detection Research Based On Binocular Vision

Computer vision has always been a research hotspot.As an important part of the field of computer vision,binocular vision has been getting more and more attention for its simple structure and strong performance.Underwater binocular vision system can be used for underwater vehicle visual AIDS,helping underwater vehicle in underwater operation,is being more and more researched.This article uses a good waterproof binocular cameras,with stereo matching algorithm and dynamic underwater target detection algorithm,respectively,finished underwater target distance measure and underwater moving targets detection task.The main research work are as follows:Firstly,we have a study of the camera imaging principle,in view of the camera tangential distortion and radial distortion,mathematical model is seted up,and a model is established on the geometric relationship between binocular cameras.On this basis,calibration method of binocular cameras on the underwater environment has be studied,through comparing with the results of calibration on land,we found that underwater calibration result is ideal.In addition,on the basis of the calibration results,we have a study of stereo images rectify,and getted a good experiment results.Secondly,we have a study about the general steps of stereo matching,and some classic binocular stereo matching algorithm,including the stereo matching algorithm based on region and semi-globle stereo matching algorithm,through the program,we realized local stereo matching BM algorithm and semi-globle block match(SGBM)algorithm,obtained a good underwater dense disparity map.Then,we have a study of target distance measure based on binocular vision.Comparing the BM algorithm and SGBM matching effect and matching speed,this paper adopts the BM algorithm to obtain disparities as object has different distance from camera.Then,by matrix operations,obtain the spatial three-dimensional coordinates of underwater target,through comparative with the actual distance,we getted measurement error,and the error is small,the measure distance have a good accurate.Finally,we have a study of underwater moving targets detecton.As the underwater image has a poor quality,characteristic of strong noise interference and no-clear background,this paper adopts a method of building background model on neighborhood point sample,detected the underwater moving tagets accurately.Comparing with other traditional detectionalgorithm,like three frame difference method,background difference method and gaussian mixture model,our algorithm have a more accurate result,stronger inhibition ability of noise interference.