A two-stage algorithm for ground filtering of airborne laser scanning data

Abstract

A two-stage algorithm is proposed in this article for ground filtering of airborne laser scanning (ALS) data. Input ALS data are initially preprocessed for outliers removal. The first stage removes the non-ground objects from preprocessed ALS data based on the geometrical reasoning, which is applied over piecewise local neighbourhoods around selected points. The second stage retrieved the ground points falsely recognized as non-ground in the first stage using geometrical similarity of ground points in their surroundings. The proposed algorithm was tested and validated comprehensively using complex and heterogeneous landscapes of selected 10 ALS data sets and additional 15 International Society of Photogrammetry and Remote Sensing data samples. The ground points were filtered out in these data sets and data samples at average total error and kappa coefficient of 3.66% and 89.15%, respectively. The proposed algorithm performs satisfactorily in the complex terrain cases such as mixed vegetation and houses on sloping terrain, low vegetation, complex objects, low and small objects, scene border, and discontinuity. The proposed algorithm is straightforward and, consequently, computationally efficient. Thus, it has potential for wider use in industry.