The Semi-supervised Classification Method of LiDAR Data Integrating with Aerial Images

Abstract

a new semi-supervised classification method is proposed by combining airborne LiDAR (Light Detection And Ranging) data with registered aerial images. Firstly, the algorithm filtered LiDAR data into ground points and non-ground points that were further partitioned into small planar regions based on local attribute estimation. Then these planar regions will be used as initial classes to obtain initial samples that were used as training samples in aerial images to perform classification process with the maximum likelihood. The proposed method can also revises misclassified building regions by using shape index. Every single LiDAR point can be labeled by comprehensively considering information like filtering results, intensity from LiDAR data and spectral features from aerial images. The experiment shows that the proposed can improve the classification accuracy of LiDAR points cloud in complicated urban.