Study On Sphere Detection Methods Based On Three-dimensional Point Clouds

As one of the key technologies in the field of pattern recognition,geometry detection of three-dimensional(3D)point cloud is widely used in machine vision,intelligent robot and other fields.Since there are a large number of spherical areas in the real environment,and the sphere model is relatively simple,accurate and efficient detection of the sphere in the 3D point cloud can improve the ability of intelligent robots to perceive the surrounding environment,which is one of the most important fundamental problems in machine vision.Therefore,in-depth study of the sphere detection methods based on 3D point cloud is of great theoretical significance and practical application value.In this thesis,the geometry detection of 3D point cloud is taken as the research background.Around the sphere detection problems in 3D point cloud,the sphere detection method of 3D point cloud based on the random sample consensus(RANSAC)algorithm is improved.Then,the calculated amount as well as the influence of points irrelevant to the detection target(i.e.,outliers)are reduced,and the speed and accuracy of sphere detection are improved,and the anti-interference ability of sphere detection is enhanced.The main contents of this thesis can be divided into the following three aspects:Firstly,sphere detection of 3D point cloud based on the classical RANSAC algorithm is introduced.In the detection,sample points are randomly selected in the point cloud in an iterative manner,and the sphere parameters are solved by analytic method.Finally,in order to minimize the error,the sphere parameters are corrected by least squares(LS)according to the set of inliers corresponding to the current optimal sphere parameters in the point cloud.The experimental results show that the introduced method can effectively detect the sphere in the point cloud.Secondly,an improved RANSAC point cloud sphere fast detection method based on the geometric constraints is proposed.Aiming at the disadvantage of random selection of samplepoints in classical RANSAC algorithm,which makes the calculation of sphere parameters uncertain,the nearest neighbor search algorithm is used to select the sample points.Meanwhile,the sphere parameters are calculated by two points and their corresponding tangent plane normal vectors.The sphere center and radius are determined by the midpoint of the vertical line between the two normals and the average distance from the two points to the sphere center,respectively.Then,the third point and its corresponding tangent plane normal vector are used additionally to verify the validity of the sphere parameters,so as to speed up the solution and verification of the sphere parameters.The experimental results show that the proposed method can effectively improve the speed of sphere surface detection.Finally,an improved RANSAC point cloud sphere detection method based on independent point marks is proposed.Aiming at the large-scale point clouds with numerous points and a large number of points independent of sphere,the proposed method uses the Delaunay triangulation criterion to divide the grid relationship between the points in the point cloud.And then marks the points in the point cloud that are independent of the sphere shape according to the comparison between the average distance of each point and its connected points and the standard distance determined by the global average distance and variance.The marked points are not used in the calculation of the sphere parameters,but only used when the corresponding sphere parameters and the corrected sphere parameters are given at the end.The experimental results show that the proposed method can effectively suppress the influence of spherical shape independent points in large-scale point cloud,and the spherical detection speed is fast.