The Research Of Binocular Stereo Vision System For Robot Carrying Tires

In the strategic plan of "Made in China 2025",China comprehensively promotes the strategy of a strong manufacturing power.The intelligent manufacturing is an important part of the strategy and also the fundamental way for China to transform from a big manufacturing power to a strong manufacturing power.As one of the top ten industries with priority development,robots have a huge demand in automobile,machinery,military,aerospace as well as home health,education and entertainment.In recent years,China's automobile industry has been developing rapidly,which also drives the development of the tire industry.China's tire output,export volume and enterprises' quantity are all ranked first in the world.In the process of tire production and transportation,using robot instead of human labor will greatly improve production efficiency and save cost.In this paper,a binocular stereo vision system of robot handling tire is studied.The main research work is showed as follows:(1)The binocular stereo vision system is set up.Due to the lack of texture on the tire surface,the laser point projector is added between two cameras to project laser points on the tire surface in order to increase characteristics of the tire surface to reduce the difficulty of stereo matching.In the stereo vision system,a longer baseline is set to improve the location accuracy.In order to increase the public field of view,left and right cameras are placed in a convergent manner.The internal parameters,distortion coefficients and the relative position between left and right cameras are calculated by the calibration software developed by the laboratory.Then these parameters are used in image correction and stereo rectify.(2)In the research of binocular stereo matching algorithm,According to acquired images,a corresponding stereo matching algorithm is designed.In order to solve the problem of large amount datum and long processing time in stereo matching,the corresponding solution is designed aiming at these two problems.First,down-sampling is carried out in original left and right images to generate new images.Then stereo matching is carried out in the new images.Left and right disparity maps of down-sampling images are acquired and the tire regions are extracted on the disparity maps.On the original images,matching cost computing,matching cost aggregation and disparity post-processing are only performed on tire regions,which greatly speed up the calculation of the stereo matching algorithm.Moreover,after obtaining the disparity maps on the down-sampled images,the disparity maps are up-sampled.And the disparity maps are further optimized in the size of the original image,which improves disparity maps accuracy.(3)In the process of point cloud registration,the programming based on PCL(Point Cloud Library)is used to register the generated three-dimensional point cloud on the tire upper surface with the tire template point cloud.The tire template point cold is generated by the laboratory's linear structured light scanning platform,and then the point cloud registration is carried out.First,the point cloud is down-sampled to accelerate the processing speed of point cloud.Then the key points are extracted by ISS(Intrinsic Shape signatures)algorithm,and key points'SHOT(Signature of Histograms of Orientations)descriptor is calculated.And the point clouds are roughly registered based on the SAC-IA(Sample Consensus Initial Alignment)algorithm and then the ICP(Iterative Closest Point)algorithm is used to precisely register the point clouds.Finally the tire position and pose are calculated.(4)The experiment is designed to verify the accuracy of our binocular stereo vision system.The rigid transformation matrix between the position of the tire after moving and initial position of the tire is calculated based on the data collected by the laser tracker,and the rigid transformation matrix is regarded as the true value.The tire template point cloud is transformed to the point cloud of the tire's initial position.Then the true rigid transformation matrix and the rigid transformation matrix calculated by our algorithm are applied to the tire template point to obtain two cloud points.The centroid coordinate differences and the normal angle between two point clouds are regarded as the relative measurement error.The experiment proves that our binocular stereo vision system has a higher precision.