UNDERWATER ACTIVE VISION Sample Clauses

UNDERWATER ACTIVE VISION. Traditionally only sonar sensors have been used for detecting targets underwater. This is because sonar can operate over long distances, and in water with poor visibility and lighting. Sonar, however, has two main drawbacks: it does not yield the albedo of the sensed surface and it is not reliable for targets closer than 5m, i.e. for tasks like docking and inspection. In recent years, vision has been investigated intensively for use with underwater robots and unmanned vehicles, in order to overcome the above shortcomings [V1]. One of the main issues is liaising the control (in case of sensory-motor control) of the vehicle with the information extracted from the images. Papers with emphasis on the control part, tend to use very elementary image processing (e.g. [V2] which is concerned with the automatic tracking of underwater pipelines), while papers that elaborate on the imaging part tend to ignore the importance of using this information for the control of the vehicle. Some of the works that concentrate on the imaging part are concerned with the design of landmarks painted on the targets, so that they are easily identifiable from a large number of distances (e.g. [V3, V4]). Using predefined landmarks to identify the target is appropriate for inspecting known targets, but not for identifying new unknown targets or targets belonging to un- cooperating agents (e.g. detecting and identifying underwater mines). The problems of using vision in underwater tasks are caused by many factors, including poor lighting conditions, vehicle vibration and drift due to underwater currents, and the presence of plankton that reduces significantly the optical depth of the water. To overcome the issue of poor lighting conditions and to be able to infer the 3D shape of the imaged surface, structured light has been successfully used for 3D shape reconstruction in industrial inspection applications [V5]. The use of structured light belongs to the category of active sensing, like sonar. In this case, however, a regular grid of light is projected on the sensed surface and imaged by a camera. The regular grid is deformed according to the shape of the surface, and from the deformation the 3D shape of the surface may be inferred. This implies that the camera has to be able to image the projected light pattern reasonably accurately. In murky waters, however, there is a very strong back-scattering component that reduces significantly the quality of the captured images. To overcome this ...