Research On Surface Reverse Algorithm Based On Machine Vision

Surface reverse reconstruction is an important approach of advanced manufacturing technology.It can digitize the existing entity model in order to make it easier to revise and redesign the model.There are many methods for digital reverse detection of solid models,and machine vision detection is one of the most effective and common detection methods.With the rapid development of computer image processing technology,machine vision is widely used in various fields such as industry,agriculture,medical care,automotive,and aerospace.The core algorithm of machine vision is a research hotspot of domestic and foreign scholars.Machine vision is a reliable and effective method for 3D surface detection.Therefore,the study of key algorithms and techniques for surface reversal based on machine vision is of great significance for improving China's intelligent manufacturing technology.The focus of this study is the key algorithm for surface reverse reconstruction based on machine vision.The surface 3D point cloud data of this study object is obtained using binocular stereo vision in machine vision.The key technologies for the surfaces reverse reconstruction,such as "3D point cloud data acquisition","3D point cloud registration" and "3D point cloud surface fitting",are emphatically studied.At the same time,this paper expounds the overall technological process of surface reconstruction,studies the theoretical technologies and principle involved in the process,and optimizes and innovates the existing theoretical technologies.The main research content of this paper are as follows:(1)A method for quickly acquiring low-noise 3D point cloud data based on machine vision is proposed.The basic principle of binocular stereo vision for obtaining 3D point clouds on the surface of objects are expounded.This method includes image preprocessing operations,implementation of adaptive minimum matching area,background processing of matching images,and processing of disparity.The main factor affecting the efficiency of 3D point cloud acquisition is the process of calculating disparity through stereo matching.The matching images size and stereo matching parameters are the main factors affecting stereo matching efficiency.In order to obtain more surface details of the measured object,it is generally necessary to use a high-resolution camera,but the obtained images are also very large,and stereo matching will consume a lot of time.In order to solve the contradiction between images quality and stereo matching efficiency,a method of adaptive minimum matching area is proposed,which only performs stereo matching on the image area where the object is located.In order to verify the efficiency of this method,an efficiency comparison experiment was designed to calculate the time-consuming for three different sized objects by comparing this method and traditional method,the time consuming is 17.2%,26.3% and 53.9% of the traditional methods.Mismatch is the main cause of noise in stereo matching.Therefore,reducing mismatch is the key to obtain low noise 3D point cloud.In this paper,the background of the matching image is treated,and the area where the object is located remains the same to reduce the mismatch.The obtained disparity are processed appropriately according to the difference of the visual field between the left and right cameras,and the noise of the 3D point cloud is very few.This method can guarantee the quality of the point cloud and reduce the time,creating a good condition for subsequent point cloud registration and surface reversal.(2)A filter parameter setting method that can improve the effect and efficiency of point cloud registration is proposed.The basic principles of point cloud registration and the necessity of preprocessing for point cloud registration are expounded.The main factor affecting the effectiveness of point cloud registration is the similarity of the two point clouds.The higher the similarity,the greater the proportion of areas and features that are common,and the better the registration effect.At the same time it increases the acquisitions times of point clouds and the registrations times,which greatly reduces the efficiency and increases the difficulty of registration.In order to solve this contradiction,this paper removes non common area by using a pass-through filtering algorithm in two point cloud.In this way,the proportion of the common area are improved,and then the registration effect and registration efficiency are improved,under the condition that the acquisitions times of point clouds is not changed.(3)A method for splicing multiple points cloud is proposed.The effect of point cloud splicing depends on the point cloud registration result and the splicing method.It is inevitable that there will be errors in point cloud registration.If the point cloud are connected by the registration result matrix to the coordinate system of the reference point cloud,splice dislocation will occur due to the accumulation of registration errors.In order to reduce the number of matrix multiplications,this paper uses stitching on both sides of the reference point cloud.The results show that this method can achieve better splicing effect.(4)Several methods of surface fitting are expounded and the fitting effects of these methods are compared.It can be seen from the results that the 3D point cloud data obtained in this paper can accurately reflect the shape characteristics of the object by using the Poisson surface fitting method.It can provide a feasible means for the reverse reconstruction of the surface of the object.