3D Reconstruction And Measurement Of Workpieces Based On Binocular Vision And Linear Structured Light

The measurement of workpiece size parameters has important practical significance for its precision processing and production,and its results directly affect the working performance.Nowadays,the needs of batch production and increasingly complex processing technology put forward higher and higher requirements for modern measurement technology.In this paper,the measurement method of workpiece is studied,and a visual measurement method based on binocular vision and linear structured light is proposed,which realizes the three-dimensional reconstruction of the workpiece and the measurement of related parameters.The main work and contents include:1)By projecting the linear structured light to structure the auxiliary feature,the problem of difficult stereo matching with no texture on the surface of the workpiece in binocular vision is solved,and the design difficulty of the stereo matching algorithm is reduced.At the same time,a miniature carrying device of linear structured light emitter is designed to achieve the uniform scanning of the workpiece surface to obtain the point cloud of the entire surface of the workpiece;and through theoretical analysis,the selection of hardware equipment such as industrial cameras,industrial lenses,and linear structured light emitters has been completed to build a complete hardware system as an experimental platform.2)When extracting the stripe image of linear structured light,the previous background difference method and the single channel threshold method based on RGB image are combined to increase the available range of parameters and improve the antinoise ability of the algorithm.When calculating the center of the light stripe at the end of the linear structured light stripe image,the result of the gray barycentric method is used to replace the result of the Steger algorithm based on Hessian matrix,which solves the problem that Hessian cannot accurately judge the normal direction at the end of the stripe,and the point cloud of the workpiece surface is reconstructed by stereo matching the result of light stripe center location.Finally,a three-dimensional reconstruction experiment was performed on the back of the calibration plate with a production accuracy of 0.01mm to verify the reconstruction accuracy of the point cloud.The experimental results show that the proportion of point cloud with error less than 0.8mm is as high as 99.5748%,and the proportion of point cloud with error less than 0.6mm is as high as 96.3600%,which has high reconstruction accuracy.3)A method for measuring the geodesic distance of the workpiece surface based on the spatial point cloud is proposed.On the basis of obtaining the 3D point cloud of the workpiece surface,the geodesic point path search algorithm is used to determine the geodesic point path between two points,then the quintic B-spline curve is used to fit the down-sampled point path to calculate the geodesic distance.The experimental results show that the method is feasible,under the fast neighbor retrieval of kd-tree,the algorithm speed can reach 180ms.Finally,by using the moving least squares smoothing algorithm and greedy projection triangulation,the surface of the workpiece was reconstructed.4)By using tools like OpenCV,PCL and Qt,a 3D reconstruction and measurement system of workpiece based on binocular vision and linear structured light has been developed,which integrates binocular calibration,automatic control of scan platform,image acquisition and transmission,linear structured light stripe image processing,stereo matching and point cloud reconstruction,geodesic point cloud path search and other algorithms into the whole software.The interactive point location of workpiece,automatic measurement of related dimensions and the visual display of 3D reconstruction results are realized.By using the software and equipment to reconstruct and measure workpieces of different shapes with sizes between 80 and 660mm,the effectiveness of the algorithm and the stability of the system are verified.Experiments show that the algorithm can achieve a relative error of 0.755%,which can replace manual measurement.