Method for Determining Grasping Position and Angle of Sea Cucumber by Rotatable Bounding Box

摘要

It is a challenging task for a manipulator to grasp sea cucumber on the seabottom, because it is difficult to determine the sea cucumber grasping angle of the manipulator. In this paper, a method for determining grasping position and angle of sea cucumber by rotatable bounding box is proposed. Firstly, an improved full convolution image segmentation network model is proposed. Secondly, the expanded sea cucumber image is used to train the improved full convolution neural network. Thirdly, we use the trained neural network to segment the image to get the sea cucumber target. Finally, the grasping position and angle of sea cucumber are determined according to the minimum enclosing rectangle of the largest connected region of the segmented image. Experiments show that the improved full convolution image segmentation network model can effectively improve the accuracy and speed of sea cucumber target segmentation. The proposed method can determine the grasping position and angle of sea cucumber at any rotation angle in real time and accurately.