Automatic extraction of digital terrain model and Building Footprint from airborne LiDAR data using rule-based learning techniques

Topographic information such as feature maps and digital terrain models (DTM) has always been a basic requirement in many engineering sciences. This is even more important in urban environments, because it is very difficult to update city maps, and this is more difficult in large cities due to the high rate of change. On the other hand, there is an increasing need for up-to-date maps, and this need is greater in larger cities due to the numerous map-related land uses. However, updating the maps takes a long time and incurs huge costs, which will prevent it from being done in short periods of time. Therefore, in this research an algorithm has been created which can achieve the following goals. 1) To generate DTM only with LiDAR data without the need for layers and other information from the area 2) To create a building footprint from the LiDAR data by removing the tree cover effect 3) To create an automatic system that can perform the production process of DTM and footprint without the intervention of an expert. To achieve the first goal, the last reflection is separated from the LiDAR point cloud and the effective distance was calculated. In the next step, noise and roof errors were removed using KNN filter and a new network was created and re-evaluated based on the shortest distance in the LiDAR point cloud to create an integrated DTM. Finally, DTM that has been generated in this research compared by DTM that was created manually. In the next step, after taking the filtering steps, the Buildings Footprint was created and was saved as a vector file in the output path by keeping the first reflectance, filtering the nearest neighbor, filtering based on intensity, creating a new network, applying the height filter, filtering based on a closed range, applying the size filter, creating the initial boundary, performing noise removal at the boundary, correcting boundary fluctuations, and finally using the decision tree. Finally, the Buildings Footprint developed based on the algorithm was compared with the Buildings Footprint developed manually to assess the accuracy of the results. In the last section, to achieve third goal, all process was written in Python computer language and DB-creator program was created and in a fully automatic process, the DTM and the Buildings Footprint were created and saved. Based on the results, the RMSE value are ±0.62 meter for the urban and ±0.28 meter for rural buildings footprints. Also, Kappa coefficient that is 0.95. Considering the technical properties of LiDAR data used, the results could be considered completely accurate as compared to the accuracy of the available data. Therefore, it can be concluded that, although the DTM built in this study differs from the hand-built DTM in areas with synthetic structures, field studies have shown that this DTM can provide more details on synthetic environments and is more accurate due to the effects on the study areas of DTM made in this research.

Text 114838.pdf Download (1MB) Official URL or Download Paper: http://ethesis.upm.edu.my/id/eprint/18188

Actions (login required)

View Item