Research On Target Recognition And Location Method Of Yarn Grab Robot Based On Binocular Vision

The bobbin yarn is widely applied in textile industry.Extensive manual operation is needed to complete when the bobbin is handled or loaded on the upper and lower yarn frames,with consequences of low efficiency.So advanced robot technology was introduced in textile industry to offset unnecessary cost from labor.It imposes competitive pressure on textile industry,which is labor-intensive now.For textile enterprises,it is an inevitable trend to replace manual operation with industrial robot.For the diversity in size,shape and color of the bobbin,it is difficult for to identify the operating target.In this paper,robust object recognition and location method are proposed based on binocular vision algorithms,which is applied to Yarn grasping robot efficiently.This paper is organized as follows:Firstly,calibration method of binocular vision system is provided by studying the Bumblebee2 camera.Based on the basis of imaging principle,the mapping relationship between spatial points and its corresponding two-dimensional image pixels are analyzed.Furthermore,internal and external parameters of camera is determined by the classical Zhang's three-step calibration method.Calibration results indicate that its error is within 1pixels which meets precision requirement of grabbing for industrial robot.Secondly,Impact from noise and uneven illumination on image of bobbin are eliminated by weighted average grayscale and filtered to reduce noise.Gamma ray is used to correct the image of bobbin and eliminate the uneven illumination.The Sobel operator is utilized to detect the edge of the bobbin image.Unexpected breakpoint can be supplemented by the edge connection operator.With the aid of topological analysis,the contour features of the bobbin edge can be extracted by the contour tracking algorithm and the bobbin is effectively recognized.The overlapping relation can be determined according to the angle between the surface normal vector and projection plane for the overlapping bobbin.Thirdly,bouguet algorithm is proposed to correct the left and right images of binocular vision and reduce the complexity of images matching algorithms.The matching point pair is constrained on a straight line.Due to weakness of the feature for the bobbin image,a stereo matching algorithm based on global minimum spanning tree is proposedfor image stereo matching,and the spatial coordinates of the bobbin yarn are obtained by combining the calibration parameters.Finally,An experimental platform is built based on binocular vision to validate target recognition and location method for yarn grasping robot.The overlap of single and multiple yarns is investigated experimentally to verify the recognition and location algorithm presented in this paper.In addition,44 figures,13 tables,67 references are included in this paper.