The Acquisition And Processing Of Optical Information For Intelligent Underwater Manipulator

Since the mid-90's of the last century, Autonomous Underwater Vehicle(AUV) has been Increasingly being used in oceanic scientific investigation and military field. AUV can independently finish underwater scout, drawing the map of seabed, underwater rescuing, geologic exploration and sampling ore and so on. Intelligent underwater manipulator is one of the necessary tools for AUV. But how to implement its intelligence is a hot difficult problem in the world. This paper discusses how to achieve"intelligent"through the vision information from the visual sensor at the background of"Intelligent underwater Robot Gripper". which is one of the"Eleventh Five-Year"exploration-oriented projects of 863 national plan, and presents a vision information acquisition and processing system in which the manipulator grasps the object based on monocular underwater vision in tailor-made way according to the structure of the manipulator and the configuration of the sensors.This paper first presents an intelligent working way which includes object tracking based on Particle Filter and Mean Shift, getting the position of the manipulator based on Simultaneous Localization And Mapping, which all through monocular vision, and then makes a grasping plan after fusing them. Firstly, To improve the visibility under water, two approaches are introduced, one is optical method and the other is Retinex. To improve the robust of this intelligent system, the system runs in HSV color space. To extract the contour of the object, Level set initialized with the tracking ellipse from the Particle Filter and Mean Shift is used. Lastly the feasibility of using the MonoSLAM method invented by Davison under water to get the pose and position of manipulator is discussed.