Research On Multi-resolution Hierarchical Interpolation-based Filter For Airborne LiDAR Point Clouds In Forested Area

Digital Elevation Model(DEM)of forest area is widely used in ecology,geology,hydrology and other important fields.Light Detection and Ranging(LiDAR)is less affected by factors such as light,temperature and seasons.It can quickly obtain high-precision,high-resolution real-world 3D coordinate data,and has become an important method for obtaining high-precision DEM in large forest areas.Point cloud filtering is the prerequisite and key step for all applications of airborne LiDAR data,and has been the focus of research in various surveying and mapping fields and computers.In recent years,the continuous development of various filtering algorithms has improved the quality and efficiency of airborne LiDAR data processing.However,most of the current filtering algorithms have better filtering effect when the terrain is relatively flat and the environment is simple,but the filtering accuracy is not high in the forest areas with steep terrain,discontinuous and complex environment.Therefore,it is of great significance to study the airborne LiDAR point cloud filtering method suitable for forest areas.Based on the full evaluation and analysis of the existing airborne LiDAR point cloud filtering algorithm,this paper proposes a filtering strategy that combines seed points and improve threshold adaptability for the reason why the classic filtering algorithm in the forest area is not applicable,thereby improving the new filtering algorithm feasibility and efficiency of the forest region.The research mainly starts from the following four aspects:(1)Acquire ground seed points.To solve the problem of fewer ground points in the data obtained by the airborne LiDAR system in densely forested forest areas,the morphological opening operation can be used to obtain the characteristics of relatively low points,and considering the influence of structural elements on the opening operation results,this paper is constructed a method for obtaining ground seed points using multi-scale morphology operation is proposed.The use of structural elements of different sizes to perform morphological opening operations on the point cloud can obtain a large number of ground seed points,thereby constructing an initial ground reference surface with higher accuracy,and reducing the accumulation of errors in the subsequent filtering process,while weakening the quality of the original data Impact on the performance of the filtering algorithm.(2)Elimination of non-ground points.Generally speaking,the obtained ground seed points may contain non-ground points(abnormal points).To solve this problem,a robust z-score method is proposed to remove non-ground points from seed points.In this paper,the standard z-score estimator in statistics is used as the outlier elimination framework,and two robust statistics are introduced to improve the collapse point of standard z-score,so as to improve the ability of z-score method to resist the interference of outliers.At the same time,it reduces the false elimination of ground points.(3)Establish an adaptive gradient threshold.Establish an adaptive slope threshold.Aiming at the problem that the classical filtering algorithm is easy to destroy the terrain structure and misclassify in the forest area with large terrain fluctuation and complex terrain,an adaptive slope threshold filtering method is proposed.In the filtering stage,the adaptive residual threshold considering the slope will change with the change of the topographic slope.On the one hand,it reduces the risk that the fixed threshold value will filter out the height of the complex terrain as a feature point,thus damaging the terrain structure.On the other hand,more ground points can be selected correctly,which improves the accuracy of each ground reference surface.(4)Experimental verification.In this paper,class I error,class II error,total error and kappa coefficient are used as accuracy evaluation indexes.The first part of the experiment takes six sets of ISPRS mountain benchmark data as the research object.Through single parameter and optimal parameter experiments,the insensitivity of parameter setting and high efficiency of operation of the algorithm in this paper are proved.By comparison with MHC and fifteen filtering algorithms proposed in the past five years,it is found that the new method proposed in this paper has the highest overall filtering accuracy,with an average total error of 1.88%,and is at least 35.3%more accurate than the traditional filtering method,which proves the reliability and robustness of the new method in mountainous areas.The second part of the experiment takes six sets of high-density forest data with different terrain features and vegetation coverage as the research object,and compared with four commonly used filtering methods(MF,PTD,IPTD and CSF)in detail.The results show that the new method has the highest overall accuracy,the average total error and kappa coefficient are 6.82%and 85.61%respectively.The accuracy of DEM constructed by the new method is about 40%higher than other methods on average,and is most similar to standard DEM visually,which proves that the new method is more suitable for forest areas.The third part of the experiment discusses the effectiveness of the three innovations with the help of ISPRS benchmark data.The results show that the three innovations enhance the adaptability,accuracy and robustness of the filtering method for various vegetation coverage and terrain features.