Robot recognition, location and grabbing system based on binocular vision
                                                                         CHEN Liting, NIE Xiaogen
                              (School of Mechanical Engineering and Automation, Fuzhou University, Fuzhou 350108, China)

Abstract: Aiming at the difficulty of classifying and extracting threedimensional pose of multiclass stacked workpieces, experimental study on the robot recognition, three dimensional location and automatic grabbing of randomly placed of regular objects were carried out. A visual algorithm based on MLP classifier was used to realize workpieces recognition, and a perspective deformable template matching algorithm based on recognition result was used to separate the stacked workpieces from the background, the threedimensional pose estimation was further completed by combining stereo matching, depth calculation and coordinate transformation in binocular vision, based on the above research the software were carried out under the environment of Halcon and Visual Studio. Experiments to realize the pose extraction and grabbing of rectangular and cylindrical objects were carried out using the platform. The experimental results show that the system has good classification effect and high recognition and positioning accuracy, and can meet the realtime online grasping requirements.

Key words: multiclass stacked workpieces; threedimensional pose; binocular vision; MLP; Halcon