| Road plays a major role in the national socio-economic development and people's daily life.With the fast development of society,traffic environment changes rapidly.It is therefore urgent to acquire road information timely,accurately and efficiently.Consequently,road information acquisition is an important part in the city informatization construction.The advanced airborne LiDAR technology provides a new means for the road information acquisition.This technology can obtain surface three-dimensional point cloud quickly and accurately.Moreover,it is less affected by the external environment change and could observe the Earth around 24 hours.In recent years,road extraction from airborne LiDAR point cloud has become a hot research topic and mainly comprises of the following four key technique parts,namely point cloud denoising,ground points filtering,road point cloud extraction and city road networks extraction.Denoising is a key stage of point cloud preprocessing.Due to the influence of instrument itself or changing external environment,the obtained point cloud always contains noisy data.The existence of these noisy data can decrease the accuracy of point cloud post-processing including but not limited to digital terrain model building and road extraction.Traditional point cloud denoising methods would usually filter effective ground points as noisy data at abrupt terrains thereby loosen useful information.Introducing new theoretical knowledge and proposing a new denosing algorithm which can improve the data quality by means of eliminating noisy data and protecting terrain information is the focus in the point cloud denoising research.Point cloud filtering is the process of removing non-ground points while reserving ground points from LiDAR point cloud.It is a fundamental step of realizing road extraction.As the mathematical morphology owns the strengths of simple principle and high realizing efficiency,it is always applied to point cloud filtering.However,there are two disadvantages of the traditional morphological filtering algorithms.On the one hand,the traditional methods need to set the slope as a constant,which makes the algorithm lack adaptability.In addition,the traditional methods cannot protect terrain details effectively while using larger filtering windows.These two shortcomings seriously limit the scope of traditional morphological filtering application as well as its filtering accuracy.A means to further improve the adaptability of morphological filtering in complex environments and filtering accuracy is a hot issue in point cloud algorithm research.Road points are generally included in ground point cloud.The ability for one to discriminate road points from ground points effectively is a difficult problem.Most researchers adopt intensity constraint to realize road point cloud discrimination.However,the existed algorithms always need to set intensity threshold manually.Too many manual work involved has greatly decreased the algorithm automation degree and make the algorithm non-universal.Thus,it is urgent to develop an automatic and accurate intensity determining method based on road point cloud intensity data characteristics.Road centerlines can reflect topological relation between roads.Thus,it is necessary to extract road centerlines from road point cloud to reflect specific information of city road network.Nevertheless,there are many narrow roads in city areas,such as aisles,corridors,etc.These narrow roads do not belong to city main roads and will make the extracted city network own lots of spurs.The removal of the influence of such kind of roads is still a vibrant research area.Besides,city areas contain some road-similar areas,namely parking lots,bare ground,courtyard,etc.These areas own similar elevation and intensity values with roads.Therefore,it is necessary to study how to remove the influence of these road-similar areas to realize the correct and complete extraction of city road networks.This study carried out a deep research and discussion based on the elaborated problems existing in the four key parts in road extraction.The main research contents and obtained outcomes are given as follows.1.This study introduced Empirical Mode Decomposition(EMD)technique to airborne Li DAR point cloud denoising and proposed a novel denoising method based on EMD.The proposed method calculated the elevation differences between row point cloud and reconstruction point cloud to automatically realize noisy data detection and elimination.This study used instance data and simulated data to test the proposed algorithm.The experimental results showed that the proposed algorithm can remove noisy points effectively and improve signal to noise ratio as well as data quality.2.This study proposed an improved morphological filtering algorithm based on progressive kriging interpolation.The essence of this algorithm is to combine the strengths of surface fitting filtering algorithm and traditional morphological filtering algorithm.By calculating the slope gradient of each level on the basis of kriging interpolation,the proposed algorithm can decrease the effect of flattening terrain when using larger window for filtering.This study adopted the test data provided by the International Society for Photogrammetry and Remote Sensing to test the proposed improved morphological filtering algorithm.The experimental results showed that the proposed method can protect terrain details effectively and sufficiently decrease the type?error.The obtained average overall accuracy for the filtering was 94.66%.3.This study introduced Skewness Balancing algorithm to road point cloud intensity threshold determining issue and proposed a novel algorithm.The proposed algorithm first assumed that pure road point cloud intensity values are in normal distribution.Due to the influence of non-road point cloud intensity values,all the point cloud intensity values were observed to be in positive skewness distribution.By removing the influence of non-road point cloud intensity values constantly until the point cloud intensity values are in normal distribution,road point cloud can be obtained.The experimental results showed that the proposed algorithm can get the intensity threshold accurately and parameter-freely,and improve the automation degree of threshold setting.4.An accurate,fast and high quality multi-level fusion and optimization method to extract city road networks has been presented.In the realization of this method,this study first proposed a narrow road recognition algorithm based on rotating neighborhood.By rotating the road neighbors at multi-directions,the proposed algorithm can discriminate and eliminate narrow roads automatically.And then a road-similar areas discriminating algorithm has been proposed on the basis of topological relation between roads.By calculating the chessboard distance between road crossing points,the areas where the distance is less than the threshold are discriminated as road-similar areas.Lastly,this study presented different fusion and optimization rules against road areas and road-similar areas.The presented study adopted the point cloud located at Vaihingen city,German provided by the ISPRS to test the proposed method.The results showed that the road extraction outcome owns 91.4% correctness,80.4% completeness and 74.8% quality.These three indices are all higher than those of the other existing road extraction models or algorithms. |
PDF Full Text Request