The Technology Research Of Point Cloud Filtering And Feature Descriptor

Pre-processing operations and feature descriptor are the key technologies ofthree-dimensional reconstruction which expresses the objective world. The filtering of thepoint cloud is more crucial in pretreatment. After filtering, the feature of the reconstructedmodel is clear and Surface is smoothing. Feature descriptor is the key in matching processbased on the feature. So this in-depth analysis the research current situation of point cloudfiltering algorithm and feature descriptor algorithm. Point cloud filtering algorithm andfeature descriptor algorithm have some shortage as ineffective filtering, poor filteringperformance and not enough precise in the description of feature points. Filteringalgorithm and feature descriptors are analyzed and studied.Firstly, Domestic and foreign existing point cloud data filtering algorithm is studied. Inorder to make direction and scale of such filter varies according to changes of sharpfeatures’, an method about anisotropy and adaptive point cloud filtering algorithm wasalso proposed. The algorithm is based on bilateral filtering. Through the analysis ofcovariance it adaptively draws direction and scale of the filter by the pending localinformation.Secondly, For domestic and foreign existing algorithms about feature descriptor arestudied. In order to consider the shape of features in the feature descriptor, RIFT algorithmbased on the optimal neighborhood radius was proposed. Algorithm describes theextracted feature points, and calculates the gradient information in the optimumneighborhood radius of each point. Algorithm is in two hemispheres in the neighborhoodmake the gradient information of feature points assign to the respective directions.Descriptors dimension is halved.Thirdly, the Experiment verifies the correctness and effectiveness of the algorithm.