Indoor Point Cloud Target Recognition Based On Joint Criterion And Support Vector Machine

With the rapid development of computer vision and artificial intelligence,3D scanning technology has been widely used,therefore,indoor scene segmentation and recognition based on 3D point cloud has also become an attractive research direction.However,due to the complexity of indoor scenes,defects of the three-dimensional scanning device itself and the incompleteness of related algorithms,there are still some problems in the process of point cloud segmentation and target recognition,such as poor robustness of point cloud segmentation,inaccurate description of feature and low recognition rate of similar objects.Therefore,studying the effective segmentation of point cloud and the accurate description of features is an important part of improving the target recognition result.Aiming at the problem of low recognition rate due to poor robustness of the segmentation algorithm and inaccurate or incomplete description of the feature description in the current research,this paper deeply studies the segmentation of indoor scene 3D point cloud and the accurate description of point cloud features.The main work is as follows:Firstly,based on the normal and color information between the sampling point and its neighboring points,a dissimilarity judgment model for neighboring points based on the joint criterions of local concave-convex perception and the Mahalanobis distance of color moments is constructed,and improve the traditional point cloud segmentation algorithm based on graph theory.It solves the problem that the segment result is too sensitive to noise or color change which caused by the traditional algorithm solely relying on the normal difference or color difference.The segmentation experiments of different point cloud data validate the effectiveness of the proposed algorithm,and at the same time,it lays a foundation for subsequent feature description and extraction of point clouds.Secondly,as the fast point feature histogram greatly increases the computational efficiency,it is more susceptible to the noise points and the different characteristic points with close distance,thus affecting the feature description results.This paper proposes a point cloud local effective neighborhood selection method based on the differential geometry consistency between sampling points and their neighborhood points,and improves the selection process of neighboring points in the fast point feature histogram.On the basis of the distance between the sampling point and the neighborhood point,the normal difference between the points and the distance from the neighborhood point to the sample point cutting plane are included in the neighborhood point judgment process.The experimental results show that the improved point feature histogram based on local effective neighborhood has more obvious improvement in feature description than the traditional fast point feature histogram.Finally,while improving the fast point feature histogram,the characteristics such as the color of the point cloud are calculated,and then multiple types of features are re-combined into a new feature vector to describe the point cloud characteristics.It overcomes the limitations of the single type feature and make the feature description more comprehensive and effective.At the same time,the appropriate point cloud data set is selected,and the recognition experiment of point cloud segments is designed based on the support vector machine algorithm.By comparing the segments identification results of different feature descriptions,the effectiveness of the target recognition method is verified.