Research On Key Technologies Of Underwater Target Detection And Positioning System Based On Binocular Vision

The ocean is the origin of creatures on the earth, which has rich mineral resources.However, human vision is only in the visual band of the electromagnetic spectrum, limiting the exploration and development of underwater resources. The light vision of underwater robot almost covers all of the electromagnetic spectrum, it has good sensory ability of underwater environment and it can provide accurate target information for underwater robot autonomous operation. So it has the irreplaceable advantages, especially, when the target is close to the underwater robots. The study of underwater robot binocular vision detection and localization technology has important theory significance and application value for the autonomous operation ability of underwater robot.Concerning about the problem of underwater robot binocular vision target localization,this paper will focus on the characteristics of underwater target detection and extraction,underwater binocular vision localization and underwater moving target detection and target tracking technology, which provide technical support for the real-time stable working of underwater robot manipulator.In the study of underwater character target detection and extraction methods, the methods of image enhancement, edge detection and circular extraction are mainly discussed in this paper. Underwater images have noise and image distortion, since window size of traditional median filtering method is immutable, the flexibility is poor. and it can not effectively remove the large scale noise in the image. Therefore, a center weighted adaptive median filtering method is proposed in this paper, and the underwater experiments verify the validity of this method. Considering the discontinuity and weak anti-noise ability existing in the traditional algorithms of image edge detection, an improved edge detection algorithm is proposed, the noise of the image will be removed by morphology in this improved algorithm.The Otsu method is adopted to set the image threshold while connecting the edge of the image,the performance of this algorithm is verified through the contrast analysis. When making circular detection, traditional Hough transform method has the shortcomings of computational complexity and time consuming. Aiming at this problem, an improved Hough transform method is put forward, this method reduces the computational complexity and improves the operation speed, and the effectiveness of the improved method is verified by the underwater circular detection experiment.In the research of underwater binocular vision localization technology, according to the different positions of cameras, the binocular vision localization technology can be classified into traditional binocular vision localization and front-back binocular vision localization.Geometric distortion factors are put into the ideal pinhole imaging model to set up a nonlinear camera model in this paper. On account of the distortion of the underwater image and the errors during the camera installation process, the camera radial distortion correction is introduced to calibrate the camera, and the three-dimensional location information of target object is obtained through the parallax principle. Since the traditional camera calibration error is very serious, front-back machine vision method can not meet the demand of localization,therefore, a camera calibration method based on rotating transformer is proposed, and it can obtain the accurate target localization information. In this paper, after analyzing the advantages and disadvantages of stereo matching method,a stereo matching method based on the center of mass is proposed, and the validity of the proposed method is verified by the pool experiment.In the research of underwater moving target detection and target tracking method, the traditional Mean Shift algorithm for target tracking has a certain tracking deviation,when the target moves too fast, appears shade or multiple targets. In order to solve this problem,this paper puts forward an improved tracking algorithm,this algorithm uses color histogram model to get the color projection image of each frame, and adaptive adjust the window size and center position according to the tracking results of the previous frame, in order to obtain the current image size and center position. Finally, the goal of continuous target tracking is realized, and the validity of the method is verified through underwater experiments.