Underwater Stereo Matching Algorithm Based On Dynamic Programming

Rich marine resources have great potential for development. Visual detection of underwater exploration is a hot spot in recent research. It has the advantages of irreplaceable high speed and precision. Currently there are two main problems of underwater visual detection: underwater insufficient light makes image quality poor; multi-media imaging causes distortion.In the view of the underwater imaging of three-dimensional reconstruction, light refraction occurs when through water and air interface, making image distorted. This paper probes into the multi-media non-parallel binocular stereo vision system, and explores the dynamic programming stereo matching algorithm to match the multi-media images. The main work is as follows:(1) Firstly, it introduces the basic principles of binocular stereo vision, content and procedures; Secondly, on the basis of the three coordinate system, it constructs a pinhole camera imaging model; Thirdly, it focuses on the principles of stereo matching, concludes that the stereo matches each advantages and disadvantages of algorithms. Finally, it analyzes the evaluation methods of stereo matching algorithm.(2) This paper presents a model which can transfer image which camera shooting through a plat waterproof cover underwater into the image which has no refraction between camera and object. This model bases on the non-parallel binocular stereo vision system, and this paper presents some general conditions to use it. When the optical center is on the surface of the water, there is a one-to-one correspondence between the underwater image and the image-without-water. If the optical center is outside the water, this model is still valid when the distance between the focus and the object is much futher than the focal length, and the distance between the focus of lens and refractive surface along z axis. Three-dimensional reconstruction of the feature points is used in experiments to verify the validity of the model. It shows that this method transforms model firstly and then uses the air feature reconstruction method to obtain three-dimensional coordinates of the points. It is available to actually replace three-dimensional coordinates of feature points.(3) It is described a method to obtain GCPs which based both phase matching and area-based matching. Firstly, it employs principle of wavelet modulus maxima to obtain the proposed GCPs, and then use the area-based matching obtain high-confidence GCPs, and at last, optimize the path, forcing the DP algorithm to choose a path through the GCPs.It achieves the feasibility of dynamic programming stereo matching algorithm based on matching image after transformation. It obtains underwater images of dense disparity maps, verifying that the dynamic programming stereo matching algorithm is available for transformed under-water image matching.