Research On Binocular Vision Technology In AUV Guiding

Vision is one of the most important technologies of environment detection. For its advantage of the number of information sorts it can provide, It has been increasingly used in many fields. So far human has obtained much less information about the vast ocean than that about the land, while vision is a powerful sensing tool for human to enhance their understanding and make use of it. However, the application potential of underwater vision need to be further developed for the complexity to realize it. This dissertation did some research about a real-time binocular vision guiding system within the background of AUV docking.The paper analyzed the functional principles of the main components of the grabbing device. On bases of this, a set of grabbing device fit for the docking and guiding task was designed mainly according to the demand for range resolution. The later experiments suggested this grabbing device guarantees high ranging accuracy.First the image underwent Gauss filtering to decrease the noise. Then it was processed respectively by a neighbor-averaging-based enhancing filter and the histogram equalization algorithm. After compare, a conclusion had been drawn that the histogram equalization algorithm has better enhancing results. After the above filtering, a 2D-calibration-plate-based method was applied to calibrate the binocular cameras. The calibration accuracy was further improved by amending the experimental manner. Experiments suggested the above methods of filtering and calibration are feasible.Considering the great texture difference in underwater image, a texture-control-based NCC of pyramid images was adopted to match the stereo images and real-time extraction of interesting target or obstacle was achieved. The influencing factors of the accuracy and the real-time performance were quantificationally analyzed on the base of experiments. This helped to optimize the configuration of the hardware and the parameters of the algorithms. Besides, a pseudo-sonar model was introduced to simplify the expression of information in the depth map. The model expresses the information of range and position in form of an environment matrix, which can largely reduce the complexity of the use of depth map.Finally a binocular-vision-guided AUV obstacle avoidance (OA) simulation test was completed. The test was conducted on a 6 DOF AUV-motion-simulating platform in a simulating underwater environment. The test showed that the AUV can avoid obstacles of different shapes or one after the other. So it was primarily proved that this binocular-vision-guided system could provide valid real-time obstacle information for AUV.