Research On Monocular Vision Guided Method For UUV Docking

Vision sensor is the most important tools of environment detection, because the abundantinformation it can provide, computer vision has been widely used on the land. However, it isstill relatively rare in underwater applications due to the complex and changeable underwaterenvironment. So underwater computer vision is a meaningful and challenging research topic.This thesis addresses the problem of improving the monocular vision guided systemperformance according to the results of UUV docking tank experiment. It is primarily targetedat an efficient and accurate technology for providing the4-DOF position information in theprocesses of UUV docking based on visual cues.The guiding image source area is difficult to identify when UUV far from the dockingdevice and there are not enough guiding light in the camera’s field of view near the dockingdevice. In order to solve this problem, a so-called light intensity adjustable docking target isdesigned, which can adjusting the guiding light source luminescence area and strengthaccording to the variety distance between UUV and the docking device.The reference light marks the center of the docking module, its recognition and extractionresults decide the level of horizontal and vertical positioning accuracy and the matching timeaccounted for a large proportion of the entire visual processing in visual. The heart shape ofthe reference light is distinguished from the other circular guiding light sources. The quantumparticle swarm optimization algorithm and shape matching are combined to ensure the correctmatching and shorten the matching time, because this strategy can avoid the reference lightsource contour matching with each of the entire contour in the template library.Due to the error identification of the light source area, there are too much large errors inthe positioning result curve obtained from the UUV docking tank experiments. Aclassification method based on support vector machine is proposed to eliminate the false lightsource, using the seven morphological description operators as the light source characteristics.Classification experiments are carried out using light source areas extracted from the guidingimages as training samples.The semi-physical experiments are taken by using the multiple DOF shelves simulatesthe UUV motion. UUV move following the specified path in the horizontal plane and move in a certain depth experiments are designed to measure the accuracy of the monocular4DOFpositioning algorithm. Prove the monocular guiding system and the monocular positioningmethod to be reasonable and effective through the simulated of UUV docking. Theexperiments results show that the system and method proposed in this thesis is reasonable andpractical.