Research And Development Of Vision System For Autonomous Underwater Robot

Vision system plays an important role in the design and development of underwater vehicle. The vision system is able to process the captured underwater image and the results can be used to guide the AUV’s activity in the underwater environment. Due to the particularity of the underwater environment, the main steps of the image processing include image pre-processing, image segmentation, object identification and distance estimate. The vision system in this project is able to meet the requirements of the SAUC-E competition, namely process the captured image in real-time, find the target efficiently and feedback the results to the control unit.The image pre-processing includes image enhancement, colour correction and image smoothing. Because of the effects of underwater environment, the captured image is not clear enough to extract the target for the processing. Image enhancement is able to employ homomorphic filter and colour correction to enhance the objects in the image and correct the colour degradation. At last, the image smoothing is employed to remove the noises for further image processing.The edge detection and thresholding are the main parts of the image segmentation. The Robust Colour Morphological Gradient (RCMG) is employed to extract the target and apart it from the background. Subsequently, the thresholding is used for edge quantisation to obtain the interesting areas.In order to find the targets required in SAUC-E competition, improvements are made for the existing target detection. Hough Transform (HT) is employed to identify targets including gate, ball and buoy. The target detection algorithm developed in this project improves the performance of object recognition in the terms of accuracy and computational speed. Subsequently, a distance estimate system is developed to measure the distance between the underwater target and the camera mounted on the Autonomous Underwater Vehicle (AUV). Pin-hole model is investigated to work in both a single frame and a sequence. Investigation in edge image helps to improve the accuracy of the distance estimate from a single frame. In sequence operation, an initial model is established to restrict the target detection, and the Kalman tracking algorithm is also employed to obtain more accurate and steady distance results.