Research On Automatic Registration Algorithm For Scattered Point Cloud

With the deepening of research on 3D scanning technology,3D point cloud data has become more widely used in various industries.For example,3D processing technology based on point cloud data has been penetrate into aerospace industry,automobile manufacturing,mold manufacturing,electronics,the communications industry and other industries that are closely related to our lives.However,in the actual point cloud data acquisition process,the 3D scanner can only obtain the point cloud model data of a certain angle of view of the object.For obtain the complete point cloud model data of the object,it is necessary to register the point cloud data of multiple perspectives by the transformation calculation.Therefore,point cloud registration technology is an important part of 3D point cloud data processing technology.After researching and summarizing the various registration algorithms in recent years,this paper designed two point cloud automatic registration algorithms without manual intervention in the registration process.Firstly,This paper introduced the background significance of the point cloud registration technology and its future application prospects.According to our research on the point cloud registration technology at home and abroad,the point cloud automatic registration algorithms can be divided into three categories: the method of based on feature description,the method of based on statistical probability,and the method of iterative minimum error.we introduced the process of point cloud registration and the basic conception involved.Secondly,based on the summary of the current research results of point cloud registration,this paper designed and implemented a point cloud registration method based on SHOT(Signature of Histograms of Orien Tations)descriptor,it used the keypoints extracted by 3D Harris detector instead of the original point cloud to complete the registration that can reduce the calculation amount of the registration process.then,obtained the corresponding point pair sets by computing the SHOT descriptor of the keypoints.and obtain the final corresponding point pair sets by the Euclidean distance nearest criterion and the RANSAC algorithm to calculate the rigid body transformation matrix to complete the coarse registration,finally we used the ICP algorithm to performed the fine registration to improve the registration accuracy.After that,this paper innovatively designed a point cloud registration algorithm based on multi-scale descriptor.The algorithm has made innovative designed in the following four aspects: the keypoints extraction method based on shape index;designed the feature descriptor based on multi-scale eigenvalue normalized vector and normal vector deviation,the feature descriptor has stable rotation invariance and recognition and the dimension is less than the SHOT;The corresponding point pair sets query algorithm based on the nearest Euclidean distance criterion effectively eliminates the potential error corresponding point pairs;The global optimal transform selection method,it obtained the global optimal registration result by the global optimal transform select method.Finally,For verifying the validity and feasibility of the algorithm,this paper uses multiple sets of point cloud data from different perspectives to implement the two algorithms in the Microsoft Visual C++ environment.According to compared the experimental results of the two algorithms,it proved the superiority and robustness of the registration algorithm of the innovative design of this paper.