Classification and Accuracy analysis of LiDAR and Aerial Images

Abstract:

In order to make up the insufficiency of the single data source in the classification, the paper presents an approach that integrating LiDAR and aerial image, and implements the multistage complex urban ground object classification based on object-oriented and single pixel. Aerial image and LiDAR are merged and split some homogeneous regions as the research object by combining spectrum and spatial characteristics; ground points and non-ground points are separated by filtering LiDAR, and the altitude difference is obtained based on twice return pulse heights, and vegetation data from false-color aerial image, the classification rules are established and the segmentation objects are classified by adopting object-oriented method; and then the mistake objects are defined again based on above information, the second-stage classification based on single pixel is finished; further classification is put up to eliminate the mistake buildings by extracting the line segments of buildings. The result shows that the approach can automatically effectively separate buildings, woodland, grassland and road, the producer-accuracy and user-accuracy is respectively 92.53% and 95.79%, the whole classification accuracy is 89.62%, the study was a successful step in developing classification method for integrating LiDAR and aerial images.