Research On LiDAR Classification Based On Generalized Fractal Dimension

With the rapid development of the socio-economic and spatial information technology,it is imperative to use accurate and real-time spatial information as a data support which made more stringent requirements on the quality and the acquisition period of spatial data.LiDAR provides a quick,high precision and real-time way to acquire terrain’s 3D information as a new earth observation technology.Property analysis and classification has great significance to the effect of application of LiDAR,so feature classification becomes a research focus on the analysis of LiDAR.Currently,the study of LiDAR classification still does not reach the level of maturity and large-scale application,which limits the effectiveness of LiDAR in actual production.Aiming at feature classification of LiDAR point cloud,this study proposes a method of LiDAR feature extraction through space morphology characteristic of LiDAR point cloud in line with the technical roadmap of data organization and management,point cloud data denoising and filtering,feature space morphology analysis and category recognition.Specific study contents are as follows.(1)The data organization and management of LiDAR point cloud.LiDAR point cloud data’s volume is very large,and lack of space-related information between laser footprint.Point cloud data must be organized in a certain way before the analysis of LiDAR point cloud.This study proposes to establish mixing space index structure of two-level and two-index to organize and manage LiDAR point cloud.The data structure of two-level and two-index makes use of "non-null" rule cubic grid and three-dimensional K-D tree to organize point cloud in large and small scales respectively,and is unified as a whole structure.On the physical implementation,efficient data structures such as red-black tree and K-D tree are used as the foundation of realization.With this multi-division and multi-index type mixing organized way,it gets rid of inefficient and structural redundancy when a single type of spatial index used in LiDAR point cloud and provide effect data structure support for the actual analysis and process of LiDAR point cloud.(2)The denoising and filtering of LiDAR point cloud.The original LiDAR point cloud data contains terrain points and error points except for feature points,these non-feature points will cause interference on the classification of feature classification to some extent.The purpose of LiDAR’s denoising and filtering is to exclude noises and ground points and acquire feature point set.For the purpose of these two aspects,the present study combines denoising algorithm based on finite element analysis and filtering algorithm based on morphological gradient which have good process effects with point cloud structure based on established LiDAR’s data structure of two-level and two-index and designs a complete process to separate the feature points from others followed by the point cloud denoising and filtering process.The denoising and filtering process provides a data base for the feature classification of LiDAR point cloud(3)The three-dimensional morphological analysis and type attribute identification of LiDAR feature.Morphological structure of different types of features tend to have a unique feature,which determines the distribution structure of LiDAR point set in three-dimensional space.In turn,the distribution structure of LiDAR point set can also be an important basis for distinguishing feature’s type.Starting with studying the three-dimensional morphology of the feature point,this paper analyses the morphology of various types of features with fractal theory and use the generalized three-dimensional fractal dimension as a form of quantitative indicator to reflect the feature’s morphology.This paper also explores change rules of LiDAR feature’s generalized three-dimensional fractal dimension with the changes of point data conditions.The research shows that each feature type has its own unique feature of generalized three-dimensional fractal dimension of the distribution and variation interval.Using this research as a basis,it can achieve the effect of LiDAR point cloud feature classification ultimately with single feature segmentation and the three-dimensional fractal dimension analysis of each feature in feature classification process.It is found that using the generalized three-dimensional fractal dimension can separate different type feature from others relatively effectively,which indicates the feasibility and effectiveness of application on LiDAR point cloud feature classification of generalized three-dimensional fractal dimension.The proposed feature classification method in this paper has the characters of relatively wide application scope,integrated processing,effectively good classification effect and can bring enlightenment to the study of the feature classification of LiDAR point cloud to some extent.