Underwater robotics navigation using multi-sensor fusion

摘要

Image processing for underwater applications is quite complex when compared to its land based equivalent. For underwater robotics, for example, image processing is used to inspect the underwater structures and to determine the position of an ROV (remotely operated vehicle). This requires an advanced and sophisticated vision system which implements sensor fusion methods. A vision model for underwater navigation is presented here. The structure of the model uses higher interpretation levels to aid in the process of segmentation at an early stage, in order to overcome the weakness of underwater features. Image interpretation plays a crucial role in almost all the system processes: including consideration for the effects of a particular sensor, angle or distance, and in fitting the observations into the 3D global model. Fuzzy rules and morphological techniques are exploited in the processing. A constrained version of the proposed model is discussed here, and experiments on pipeline inspection are demonstrated. This illustrates, among other things, typical examples of the complexity of underwater image processing, the variety of image processing techniques, including techniques for “image interpretation”, required to obtain effective results. (9 pages)