Research On 3D Indoor Point Cloud Scene Segmentation Technology Based On Supervoxel

Three-dimensional point cloud segmentation aims to achieve accurate object segmentation consistent with real visual perception in point cloud scenes.By calculating the internal information of point cloud,point cloud scenes are divided into a group of regions with specific semantic meanings.It is a key step and an important research topic in the field of visual application such as scene understanding and target recognition.Aiming at the problems of low segmentation efficiency and poor segmentation effect in high curvature area of existing point cloud segmentation methods,this paper proposes a point cloud segmentation algorithm based on supervoxel concept,which combines point cloud concave removal to achieve accurate segmentation of general three-dimensional indoor scenes.For the three-dimensional point cloud scene after denoising and filtering pretreatment,the main work and innovation of this paper can be divided into the following aspects:A method for removing concave points in point cloud is proposed.Before point cloud segmentation,the concave points of point cloud are removed according to the set criterion of concave and convex points,which are usually located in different object boundary regions.Concave removal reduces the growth of supervoxels across objects in the subsequent segmentation process,and stops the growth of supervoxels when they encounter object boundaries in the growth process,thus maintaining the integrity of objects in the scene and laying a precise foundation for subsequent segmentation.Supervoxel segmentation is applied to the point cloud after the concave points are removed.Firstly,using Kd-tree principle to normalize point cloud voxels,extract voxel color,space and FPFH features,set seed voxels,and finally calculate the feature similarity distance of adjacent voxels.Then,voxels are clustered into larger supervoxels according to clustering criteria,which will be used as processing units in subsequent processing,thus shortening the operation time of the algorithm.The method of supervoxel region growth and fusion of adjacent surface sheets across convex edges are proposed.Firstly,we extract the normal vectors of supervoxel space after over-segmentation,and fuse the supervoxels with similar features into regions to get a cluster of adjacent surface patches.Then we set the criteria for judging the concave and convex characteristics of adjacent surface patches,and fuse the adjacent surface patches across convex edges to get the clustering segmentation results of supervoxels with missing concave points.Finally,the concave restoration algorithm based on radial basis function and simple region growth algorithm are used to recover and fuse the rejected concave points,and the segmentation results of indoor point cloud scene consistent with visual perception are obtained.Based on the idea of region growth,the proposed segmentation algorithm for three-dimensional indoor point cloud scenes achieves accurate segmentation of objects in point cloud scenes.It also has a good segmentation effect for high curvature regions and avoids the problems of under-segmentation and over-segmentation.Finally,the experimental results show that the segmentation effect of this algorithm is better than other algorithms,and the segmentation performance is good,the operation efficiency is high,which effectively improves the authenticity and accuracy of three-dimensional indoor point cloud scene segmentation.