Research Of Improved ICP Algorithm Based On Boundary Feature Points Of Point Cloud

More than 70 percent of the information people obtain rely vision.It has a great significance that using computer to help people get visual information.Computer vision is a technology that obtains 3D information of objects by simulating people's vision.With the high-speed development of computer science and technology,computer vision has been applied to various fields as a convenient method.Reverse engineering,which used to reconstruct 3D CAD model of objects,is an important application in computer vision.However,registration for 3D data is the key technology of reverse engineering,which is used in many fields.Point could data is a data structure that describe the 3D model of objects,which consists of points with at least a 3D coordinates in 3D space.It may also contain color or light reflection information.Usually,it has to carry on measurement from various angles to get a complete 3D model of an architecture or topographic features.Because of different viewpoint and coordinate system of these 3D models,we need match and align them and transform them into a unified coordinate system,which is the registration for point cloud.The principle of registration for point cloud is transforming point cloud iteratively so that reduces the distance or dislocation between two 3D models.In generally,process of registration for point cloud needs initial registration and accurate registration.The initial registration is to reduce the error of rotation and translation between tow models so that it can offer a good position to the accurate registration,which is used to improve the accuracy of registration.People put forward many solutions according to the registration for point cloud.The typical method is Iterative Closest Point(ICP)algorithm,which is widely used because it does not segment the point cloud.ICP method finds corresponding points in closest Euclidean distance between two models and uses these points to calculate transformation matrix and apply it to the point cloud.After several above steps,error between to point cloud may convergence to a given threshold.However,this method has low efficiency and poor accuracy.In order to overcome the disadvantage of the methods of registration for point cloud data,an improved iterative closest point(ICP)method based on the boundary feature points of the point cloud is proposed.First,an initial registration method based on the boundary feature points of point cloud is proposed.The method partitions the minimum bounding box of point cloud with grids in 3D space,and sets up the space grids model.Then,it applies boundary seed grids recognition and growth algorithms to extract feature points from the boundary of point cloud,and works out the transformation matrix using singular value decomposition(SVD)method to gets the results of initial registration.Furthermore,an improved ICP accurate registration method is presented.It weighs the corresponding points of the point cloud,eliminates the points whose weight is larger than the threshold,and introduces M-estimation to the objective function to eliminate the abnormal points.Finally,the point cloud is accurately registered by the improved ICP method on the basis of initial registration.Compared with original ICP method,the improved ICP method increases the efficiency by more than 70 percent and reduces the error to 0.02 percent.The experiment results indicate that the method proposed in this paper improves the efficiency and accuracy of point cloud registration greatly.