Research On Vision-based Guiding And Positioning For AUV Underwater Docking

Vision has been widely used in air due to its capability of providing abundant information. However, its application underwater is currently a challenging research topic. This thesis focuses on the needs of the future ocean space station to reclaim Autonomous Underwater Vehicles (abr. AUVs) that are sent out for tasks aiming at the development of an underwater visual guiding and positioning system who can guide the AUV to land into a dock of the ocean space station. The aspects of this vision system to be researched in this thesis are as follows:Design a visual guiding and positioning system according to the strategy of positioning to line. The landing-into-dock recovery style requires accurate directional positioning, so the philosophy of positioning to line was used as recovery strategy. Lights were adopted as guiding targets, and six visual guiding and positioning system schemes were put forward. Through theoretical positioning accuracy analysis, the scheme using the monocular-single-light ranging method was excluded. By the analysis of available guiding scope any other schemes were further excluded except the monocular and the binocular schemes who are simple and have high performance-price ratios. The monocular and binocular ranging experiments suggest that although the binocular ranging has higher accuracy, it has poor reliability and is more susceptible to the performance of the guiding lights and the error of camera calibration. The monocular visual scheme was finally selected whose hardware components were described and a linear light array was designed as the guiding target.Scattering-model-based underwater image restoration and camera underwater calibration. When a camera is calibrated underwater, the calibration images are blurred severely due to the scattering and absorption by the water medium causing lots of trouble to the extraction of calibration features. A segmented-linear-mapping image restoring approach based on scattering model was proposed. The scattering model used here was described and an experimental method was designed to calibrate the model parameters in a pool. Considering the restoration process as a segmented linear mapping from the measured pixels to the undegraded pixels, the relation between the histograms of images before and after degradation was used to set restrictions on this mapping and a set of discrete distances were derived from the restrictions by abduction. So the entire segmented-mapping function was constructed based on the scattering model. Restoring results suggest the approach is flexible and adaptable. Then the approach was applied to recover the calibration board images in camera underwater calibration and the contrast of the board areas was improved. As a result, the board areas and all the calibration dots were detected correctly and creditable calibration results were obtained.Extract image guiding features based on morphological analysis. There are three kinds of guiding features:all light areas, the reference light area and the two light areas that are the farthest away from each other on the guiding image. Light areas were primarily obtained by thresholding to the guiding image, but there are often pseudo-light areas within them. Seven BLOB morphological descriptors were adopted to describe the target areas and the conditional distribution of this 7D descriptive vector was approximated by multivariate Gaussian probability density function whose distribution parameters were estimated from samples. Bayesian minimum error rate rule was applied to classify the target areas. In the extraction of the reference light area that has distinct shape, use Canny edge filter to initialize the Snake curve (active contour), and detect the entire heart-shaped contour of the reference light area by the improved Snake model algorithm. Match degrees between the target contours and the template contours were computed on image pyramids to obtain the optimal match result. Experiments showed that the above approaches can remove pseudo-light areas effectively and extract the reference light area correctly which set base for the next position estimation.Monocular 4DOF positioning based on the linear light array. When AUV (camera) is parallel to the dock plane, methods was given to compute the height, sway, surge and yaw. Further, when AUV is not parallel to the dock plane, i.e. AUV rolls or pitches, a camera virtual rotation approach was put forward to build an interim state between states before and after the roll (or pitch). Positions in the virtual state were computed then transformed to the state without roll (or pitch) and precise 4DOF positions were obtained. Dynamic positioning experiments were carried out in a pool to analyze the available guiding scope and the anti-turbulence performance. SBL (Short Base-Line Sonar, abr. SBL) and altimeter were employed to measure the positioning accuracies of the system and compensate the positioning errors.Visual servo and AUV underwater docking trial. Different guiding strategies were researched for different docking phases. Position-based visual servo was used in vertical plane regulation and image-based visual servo in horizontal plane regulation. The computation of image Jacobian in horizontal plane 3DOF regulation was discussed and it proved that the servo system has no singular point. Simulation was carried out under Matlab. Finally, the AUV docking system was introduced and AUV underwater docking trials were performed. The AUV, the dock and the coordinating control system was described and a scheme for docking trial in water tank was designed. Data and results of the trials suggest that this monocular guiding and positioning system can guide the AUV effectively to fulfill the close docking and it proves that the approaches and methods used are sound and feasible.