Research On Technologies Of Underwater Target Locating And Tracking Based On Binocular Vision

As the most important tools of underwater detection and task completing, underwater robots should have a good understanding of underwater environment. The light vision can provide abundant information of underwater scenes for robots; especially the binocular vision can provide the precise 3D coordinates of objects based on parallax theory. The study of underwater robot binocular vision location and tracking technologies have important theory significance and application value for the autonomous operation ability of underwater robot.The paper aims at establishing the binocular positioning system for spherical targets,studying on the underwater blurred image restoration technology, object extracting and feature matching technologies, binocular positioning and target tracking technologies,providing real-time information for manipulator task completing.In the study of underwater blurred image restoration, the paper researches on two kinds of underwater blurred image restoration, blurred images caused by the absorption and scattering of water and floating granules and blurred images caused by the relative motion between cameras and targets. In the first situation, as the traditional camera underwater imaging model has many parameters and is complicated to use, the paper compared the underwater environment and air environment uses the turbulence model and wiener filtering method to restore the image. Considering the model needs to change different parameters manually, the paper improves the turbulence model by drawing in the clarity evaluation and realizing the parameters automatic identification. In the second situation, the paper uses the motion blurred model and wiener filtering method to restore the image. Considering the motion model is not suit for the partial motion blurred images, the paper improves the motion blurred method by extracting the blurred area with color characteristic and restoring the blurred area only and then joining the two parts. The validity of improved methods is verified by the underwater restoring experiments.In the study of object extracting and feature matching, as the underwater robots have a complicated system, the images taken underwater exist heavy photoelectric noise, such as Gaussian noise, salt and pepper noise. The paper studies filtering noise by using gauss filter,neighborhood mean filter and median filter. The paper evaluates the filtering effects by calculating the objective image evaluation parameters after filtering experiment which is helpful for next image processing. In the circle extracting of objects, considering the traditional Hough extracting method has disadvantages of complicated calculation and long time consuming, the paper improves the least square circle fitting algorithm to decrease the time. The target extracting experiment proves the validity of the improved method. The underwater circle extracting experiments verified the improved method. In the feature matching, considering the locating purpose, the paper regards the center of circle object as the matching feature. The center based feature matching method can precisely match the center of objects in two images and get the pixel coordinates of object center which are necessary for the next object locating.In the study of target locating and target tracking, the paper first studies the camera calibration method and gets the internal and external parameters of cameras by calibration experiments and calculation. Considering the traditional binocular positioning model cannot satisfy the demand of locating accuracy, the paper proposes the pixel based data fusion locating method. The proposed method can increase the locating accuracy compared with the traditional binocular positioning method. The underwater target locating experiment proves the validity of the proposed method. In the research of target tracking, considering the traditional Meanshift tracking method cannot finish tracking well when the size of targets in images changes. The paper improves the traditional Meanshift tracking method by changing the size of tracking window and the color space. The improved method remedies the defects and makes the tracking results better. The underwater target tracking experiment proves the validity of the improved method.