The Binocular Vision Guiding Method For Underwater Vehicle Target Grabbing

Entering the 21st century,the marine industry is developing rapidly,since the complex underwater environment,the underwater vehicle autonomous operation technology has become extremely necessary.Underwater binocular vision guided robot target grabbing technology is a key part of robot autonomous operation technology.The binocular vision system which plays the role of "eyes" could autonomously sense the position of the target and guide the manipulator to grasp the target.The subject background of this paper is ROV to accomplish autonomous underwater small target(scallop,sea cucumber,sea urchin)grasping,so underwater target positioning and guide robot to achieve target grasping based on binocular vision are researched.The main content is divided into the following modules:(1)Research on the calibration technology of underwater camera.Aiming at the problem that underwater light refraction propagation imaging no longer conforms to the perspective model,a two-step calibration method is proposed in this study.The first step is to convert the underwater image into the air image.The underwater refraction imaging model is established,compared with the air projection perspective imaging model,the positional relationship between the underwater imaging point and the ideal imaging point in the air is obtained.Then,the underwater image is mapped to the air image.The second step is to recalibrate the converted image.In this study,Zhang’s method is adopted for camera calibration.Experimental results show that,compared with directly using Zhang’s method for underwater calibration,the proposed method greatly improves the accuracy of underwater calibration.(2)Research on underwater image preprocessing technology.In this study,aiming at the problems of blurred edge details,low contrast and serious noise interference in underwater imaging,the image preprocessing is divided into two steps:denoising and enhancing.Typical denoising methods such as mean filtering,Gaussian filtering,and low-pass filtering methods,while removing noise,also eliminate a lot of edges and details,which makes the image fuzzier.The fast guided filtering method is applied for image processing in this paper.The output image and the guided image establish a local linear relationship to achieve the denoising and edgepreserving effect.Typical enhancement methods such as Laplace sharpening,histogram equalization and improvement methods have the problems of over-enhancement at the edge areas,insufficient sharpening at weak texture areas,and noise enhancement.The improved unsharp mask method is proposed for adaptive image sharpening in this paper.Considering both the frequency and degree of image gray mutation,local complexity and variance dynamically are adopted to control gain function together in this algorithm.At the same time,in order to reduce the time consumption,the image is down-sampled firstly.(3)Research on feature matching technology.Based on the real-time and robustness requirements of target position information for robot autonomous grasping,the improved ORB feature matching method is proposed in this paper.In the feature extraction stage,adaptive threshold is adopted to extract feature points,and non-maximum suppression is performed to remove feature point blocks,which reduces the number of invalid feature points and saves time.In the feature point description stage,the comparison of norm values of three pixel blocks is proposed instead of the comparison of two pixel gray values to build a binary descriptor,which improves the robustness of the descriptor.In the rough matching stage,the 2-nearest neighbor query algorithm is used to find the matching points,and epipolar constraint and maximum disparity constraints are added to reduce the matching time.An improved Random Sample Consensus(RANSAC)algorithm is present,which improve the rate of eliminate the mismatch points.(4)Binocular vision guides the robot to approach target and achieve target grasping.In this paper,an improved region growing algorithm is proposed to generate dense disparity map.Seed point intensity ranking and search window optimization is introduced to save time and improve the accuracy of disparity map.Based on the pseudo color disparity map,judge whether there are obstacles around the target.If so,give the target coordinates and the obstacle edge coordinates in real time to determine the safe area of the manipulator.If the size of the safe area is appropriate,give the decision that the target is suitable for grabbing,guide the hull planning path to approach the target,and visual servo control the robot to complete the target grabbing...