A Novel Ground Filtering Method for Point Clouds in a Forestry Area Based on Local Minimum Value and Machine Learning

Abstract

Lidar point cloud filtering is the process of separating ground points from non-ground points and is a particularly important part of point cloud data processing. Forest filtering has always been a difficult topic in point cloud filtering research. Given that vegetation cannot be completely summarized according to the structure of ground objects, and given the diversity and complexity of the terrain in woodland areas, filtering in the forest area is a particularly difficult task. However, only few studies have tested the application of the point cloud filtering method for forest areas, the parameter setting of filtering methods is highly complex, and their terrain adaptability is weak. This paper proposes a new filtering method for forest areas that effectively combines iterative minima with machine learning, thereby greatly reducing the degree of manual participation. Through filtering tests on three types of woodlands, the filtering results were evaluated based on the filtering error definition proposed by ISPRS and were compared with the filtering results of other classical methods. Experimental results highlight the advantages of the proposed method, including its high accuracy, strong terrain universality, and limited number of parameters.