Research On Low Overlap Rate Point Cloud Registration Method Based On Fast Point Feature Histograms

As the current method of obtaining point cloud data becomes more and more mature,the application of point cloud based model reconstruction is also more extensive,such as map reconstruction of autonomous driving,reconstruction of cultural relics models,etc.With the increasingly mature way of obtaining point cloud data,the application of point cloud based model reconstruction is also more extensive,such as map reconstruction in automatic driving,cultural relic model reconstruction and so on.In all kinds of research on point cloud based model reconstruction,point cloud registration is an important and complex research content.Point cloud registration is a process of stitching together two or more point clouds with a certain overlap rate,which is essentially to calculate the rotation and translation relationship between point clouds.However,in point cloud registration,the overlapping rate between point clouds is usually low due to factors such as viewing angle limitation,occlusion,and dead angle.Therefore,this paper studies a low overlap rate point cloud registration method based on fast point feature histogram.The main research contents and innovative works include:(1)Aiming at the problems of small amount of valid point cloud data and low proportion of valid point pairs in point pair set when registering two low overlap rate point clouds,a paired point cloud registration algorithm combining fast point feature histograms(FPFH)and convex hull is proposed.The algorithm uses the FPFH descriptor to establish a point pair set of point clouds,and introduces a convex hull to obtain a rough transformation relationship to screen and verify the point pair set.Finally,the transformation matrix is solved through the screened point pair set,and the registration of two low-overlap rate point clouds is completed.The experimental results show that the algorithm has higher registration success rate and registration accuracy in the face of low overlap point cloud registration.In addition,the registration algorithm can also be applied to high overlap point cloud registration,which further expands the application scenario.(2)Aiming at the problems of heavier computational burden,low faulttolerance rate and large cumulative error caused by the number of point clouds being more than two when multiple point clouds are continuously registered,a continuous registration algorithm of multiple point clouds with low overlap rate is proposed.The algorithm takes the obtained first point cloud as the reference point cloud,calculates the transformation matrix between each point cloud and the previous point cloud,and reduces the error through fine registration,then calculates the transformation matrix from each point cloud to the reference point cloud according to the transitivity of rigid body transformation,and finally splices the whole point cloud after transformation in turn.The algorithm takes the first point cloud as the reference point cloud,calculates the transformation matrix between each point cloud and the previous point cloud,and reduces the error through fine registration,then calculates the transformation matrix from each point cloud to the reference point cloud according to the transitivity of rigid body transformation,and finally splices the whole point cloud after transformation in turn.At the same time,an optimization algorithm for the selection strategy of the reference point cloud is proposed.The experimental results show that the registration algorithm can effectively perform continuous registration of multiple point clouds with low overlap rate,and reduce the amount of point cloud data required for continuous registration and the cumulative error.In addition,the strategy optimization algorithm improves the fault-tolerance and accuracy of the continuous registration algorithm.