Automatic Malleefowl Mound Detection Using LiDAR-based Ground and Habitat Features with Planar Terrain Modelling

Over the recent years, airborne Light Detection and Ranging (LiDAR) data has been used to detect the egg incubators (a.k.a. mounds) of Malleefowl on large conservation areas. However, the mound detection accuracy of the existing automated methods is not high enough to be implemented in the field. To improve the detection accuracy, this paper focuses on the mound and its habitat-based feature extraction and classification, using the distinctive geometric behaviour of and surrounding the mound. We propose a novel feature extraction method of mounds, which leverages the slope of the ground with a planar terrain model (a.k.a. the ground plane) obtained by least squares plane fitting of the ground points. By observing the features based on the ground plane, we extract structural and habitat features, and group the mound-like ground points using clustering. Finally, we apply several binary classifier models to detect areas with mounds, among them Medium Gaussian SVM performs the best (around 95 % accuracy). Our training and testing datasets contain LiDAR point cloud data captured from the Tarawi Nature Reserve by the New South Wales Government in Australia.