The Research Of Underwater Binocular Stereo Matching Algorithms

There is an inexhaustible supply of resources in ocean.If we can develop marine resources reasonably,the shortage of land resources could be alleviated.Compared with other underwater sensing technology,vision has advantage of high precision as it could present abundant information.Therefore,vision has become a necessary technology in underwater detection.Binocular vision technology enable three-dimensional reconstruction of environment,in which stereo matching is a key step.This paper focus on binocular vision system in air and underwater environment and its work is organized as follows.(1)Firstly,the transformation relation between each coordinate system of binocular vision is introduced.Then,imaging model in air and underwater environment are described respectively.Besides,the underwater imaging process is studied in two cases: non parallel system and parallel system.Principle of stereo matching is introduced and steps of stereo matching are described in detail.We also introduce several commonly used constraints and summarized and classified popular stereo matching algorithms.(2)In underwater imaging,light is refracted when it passes through different media,thus the most important constraint in stereo vision: epipolar constraint is no longer satisfied.To solve this problem,we propose a constraint based on projection area,which allow stereo matching without calculation of epipolar curve.Firstly,the projection area of reference point is calculated.Secondly,the projection area is used as searching constraint and similarities between reference point and points in projection area are calculated.Finally,we select the best matching point by using the Winner Takes All policy.To prove the advantages of our method,it is compared with classical algorithms.Experiments show that,the matching accuracy of our method is higher than classical algorithms.Furthermore,it is applicable for dense stereo matching,laying foundation for sequent application of stereo vision in underwater environment.(3)In stereo matching,similarity measurement is an essential step,however,there is always trade-off between accuracy and computational efficiency.To solve this problem,this paper proposes to accelerating matching cost calculation in two aspects: combining efficient method and effective method by hierarchical screening and reduce pixels in window by self-adaptive sampling.In this paper,we take Sum of Absolute Differences(SAD)method and Normalized Cross Correlation(NCC)method as an example to illustrate implementation of our accelerating strategy.Experimental results prove that the proposed strategies achieve a significant improvement in computational cost while maintaining the accuracy of effective method.