Research On Recognition And Location Technology Of Mechanical Parts Baseed On 3D Vision Of Suructred Light

In recent years,with the continuous development of industry towards the direction of intelligence and efficiency,3D object recognition has received great attention,especially the recognition and positioning of mechanical parts based on 3D vision.Based on surface structure light equipment to obtain 3D information,in-depth exploration of 3D vision parts recognition technology.However,the current recognition algorithm has the problem of slow recognition speed and low accuracy under the interference of noise,chaotic stacking and other factors.To solve these problems,based on the premise of the point pair feature recognition algorithm,this paper proposes to build a box for the field spot cloud,use unequal mesh to select evenly distributed feature points to form the point pair feature descriptor,and realize part recognition based on the point pair feature constraints of the point cloud center,effectively improve the recognition speed and accuracy of the point pair feature recognition method when there is background interference.The main contents of this paper are summarized as follows:(1)Build a 3D visual experiment platform based on surface structured light,and use fringe projection technology to achieve 3D reconstruction of objects.The multi-view point cloud of the object was obtained through the experimental platform,and the fusion of the point cloud in each view was realized by the point cloud target splicing method.Finally,the complete three-dimensional point cloud data of the object to be recognized was obtained.(2)The distribution of key points is uneven when the existing point pair feature recognition algorithm is building the feature description sublibrary of the cloud point pair of scenic spots.A method of uniform point selection based on heterogeneous element was proposed to construct an appropriate number of point pairs feature description sublibrary,so as to ensure that the key points of appropriate number and uniform distribution could be selected according to the size of the site cloud.The position and pose of objects are identified through matching and voting between point pair feature descriptors in the scene and model point pair feature database.(3)Aiming at the slow speed of feature matching process in point pair feature recognition algorithm,a three-dimensional point cloud object recognition algorithm based on center constraint is proposed.In this method,the approximate center point of the object in the scene is estimated by the intersection density of the normal vector in the field spot cloud,and the distance between the two selected points in the feature descriptor of the constructed point pair is constrained by the size of the point cloud of the object to be recognized,so as to reduce the amount of feature matching of the point pair.Then the feature database is matched and voted with the model points to identify the pose of the object.The hand-eye calibration method of Eye-to-Hand system composed of 3D imaging equipment and robot is introduced to verify the pose of object recognition.(4)The research algorithm is applied to 3D object recognition system of 3D vision.With matte white metal ball as the object to be recognized,object recognition is carried out in the scene where matte white metal ball and other objects are placed in confusion and matte white metal ball is stacked.And through the recognition and positioning experiments of mechanical parts,the experimental platform is built,and the 3D object recognition system based on surface structured light is developed to verify the effectiveness of the recognition algorithm.This paper mainly studies the rapid 3D object recognition technology based on 3D vision of surface structured light.It has been proved that the recognition and positioning accuracy of different objects in chaotic placement and similar objects in chaotic stacking scenes are 95% and 90%,respectively.It has certain research value to improve the actual production efficiency of robot.