Research On Stereo Matching Method Based On Underwater Binocular Vision

With the increase in population and the continuous reduction of terrestrial natural resources,the exploration and development of marine resources has accelerated the process.Binocular stereo vision is an effective underwater detection method.However,the complex and changeable underwater environment will cause light scattering and absorption,resulting in image quality degradation,and when shooting underwater,the imaging light has to pass through different media and be refracted.These interference factors make the work of underwater binocular stereo vision difficult.Stereo matching is the core step of binocular stereo vision,so it is of great significance to study the underwater stereo matching method.In view of the different binocular vision processing tasks,this paper separately studies the sparse stereo matching method and the dense stereo matching method in the underwater environment.The main research contents of this paper are as follows:Firstly,underwater image enhancement methods are studied.Aiming at the degradation of image quality in the underwater environment,the reasons for the decline of underwater images are analyzed,and the existing traditional image enhancement methods are summarized.In view of the existing traditional image enhancement methods that do not consider the imaging model when enhancing underwater images,there are still problems such as unclearness and color shift after the enhancement,so this paper improves the image enhancement algorithm based on color correction and dark channel prior,to solve the problem of underwater image degradation and lay the foundation for subsequent research.Secondly,the sparse matching method based on specific point disparity of underwater binocular stereo vision is studied.In the underwater environment,the camera refraction phenomenon occurs in the imaging process due to the waterproof sealing.The common method of dealing with the refraction of underwater binocular stereo matching is studied.Matching cost calculation is the basic part of stereo matching.The accuracy of the cost calculation will directly affect the final matching accuracy.A single matching cost cannot satisfy the underwater environment well.In response to these problems,this paper uses a multi-matching cost fusion method based on curve constraints to complete underwater sparse stereo matching.In a comparison experiment with a single matching cost matching method,this method has achieved the best performance in terms of underwater object measurement accuracy.Thirdly,the generation method of underwater binocular data set is studied.The binocular stereo matching data set in the air is relatively complete,which can meet the training needs of convolutional neural network.However,the binocular data set of the underwater environment used for convolutional neural network training is very scarce.In view of the lack of data set,this paper improves an underwater binocular dataset generation method based on style migration network,the improved network's generation results achieve the best performance in both peak signal-to-noise ratio and structural similarity indicators,providing data support for the binocular stereo matching network.Finally,a dense stereo matching method based on underwater binocular stereo vision is studied.The calculation of the matching cost of the traditional underwater stereo matching method is based on manual design,and the energy function needs to be continuously optimized to complete the dense stereo matching,which is very time-consuming and low accuracy.This paper designs an underwater stereo matching method based on convolutional neural networks to solve the problem of accuracy and efficiency of underwater dense matching.The comparison experiment with the traditional underwater dense matching method verifies the effectiveness of the method in this paper in matching efficiency and accuracy.