Binocular 3D Recognition And Positioning System Based On Line-laser Scanning

The demand for the recognition and positioning ability of industrial robots is increasing with the rapid development of intelligent manufacturing.Recognition and positioning technology is widely used in production chains such as palletizing,transportation,parts assembly,and welding.Compared with methods based on two-dimensional images,three-dimensional recognition and positioning technology can gets rid of the dependence on texture information and provides more valuable positioning information.The existing 3D recognition and positioning system faces the following challenges in industrialization:(1)ambient light interference,(2)fusion of 2D and 3D information in recognition and positioning,(3)rapid construction of 3D template.Focusing on these problems,this thesis first introduces an active binocular vision 3D measurement system based on line laser scanning to reduce the ambient light interference on 3D measurement.Second,the LineMod algorithm is improved,so that the 2D and 3D data can better complement each other in recognition and positioning.Finally,this thesis proposes an automatic template construction method suitable for fixed field of view.The main work of the thesis is as follows:(1)An active binocular vision 3D measurement system based on line laser scanning is proposed.The brightness characteristic of the line laser is used to reduce the interference of ambient light noise in the 3D measurement.The binocular stereo vision model is introduced to decouple the relationship between the depth calculation and the line laser pose.On this basis,a combination method of inter-column interpolation and inter-row filtering is proposed to interpolate sparse depth maps which can achieve rapid alignment of 2D and 3D information.The experimental results show that the relative error of the 3D measurement of this system is about 0.03%.(2)The thesis studies the development status of recognition and positioning algorithms,and introduces LineMod algorithm to fusion 2D and 3D information in recognition and positioning.The improved similarity scoring method effectively improves the accuracy of z-axis angle and the recognition ability when the object is occluded.The depth information is used to enhance the contour information.Experiments show that under different lighting conditions,the 3D coordinate positioning error of the experimental system is about 3 mm,and the attitude angle error is about 1°.(3)This thesis proposes to use the rotation and scaling of a single frame to realize the automatic construction of templates under the condition of fixed field of view.In order to improve matching efficiency,a main template set with an angular resolution of 10° is established for coarse matching,and a sub-template set with an angular resolution of 1° is established for fine matching.Additionally,the template set scale is bound to the object depth,and the prior information of the input depth map is used to index the corresponding template set to accelerate the matching process.Control experiments on dataset show that the template matching method in this thesis has obvious advantages compared with Halcon's shape-based template matching method.