A voxel-based multiscale morphological airborne lidar filtering algorithm for digital elevation models for forest regions

In this paper, we proposed a voxel-based morphological filtering algorithm that can generate very accurate digital elevation models (DEMs) over forest regions because accurate DEMs are essential for forest mapping and other forest applications. Height distribution analysis, convexity constraints, morphological filtering, and moving-window voxel-based filters are exploited to detect object points. An object index is introduced and is computed by Otsu segmentation to label the classified lidar points i.e. indices above the threshold for objects are regarded as objects. To validate the proposed algorithm, multiple experiments, including the ISPRS benchmark datasets and ten forest datasets, are conducted and compared with several existing lidar filtering algorithms. The test results of the ISPRS datasets indicate that the proposed algorithm achieved low commission errors ranging from 1.53% to 6.91%. Also, the test results of the forest datasets demonstrate that the mean errors of the proposed algorithm are compatible with those from other algorithms.