Building extraction for 3D city modelling using infused airborne LiDAR and high-resolution aerial photograph

Accurate and timely mapping of the urban building is crucial for proper planning for planners, managers, and even the government. Nevertheless, the urban environment is complex and heterogeneous, with different features such as buildings (houses), transportation, and vegetation. The extraction of urban features remains a challenge for planners and government due to the issues associated with the urban areas. In the past photogrammetric sensors were deployed. However, it was time-consuming, capital intensive and manual. The revolution of technology has made available Airborne light detection. The ranging sensor (LiDAR) has undeniably brought about detailed, speedy terrain mapping, although with the challenge of many weeks of building feature detection and modelling process due to its discriminate placement of elevation points on everything. It includes cars, houses, and trees. Hence, the focus of this thesis carried out urban building detection and, where possible, had minimal user intervention in its process. In the first instance, LiDAR derivatives were employed via an image algorithm to perform the detection of buildings. Our method achieved promising results over a large scene with completeness, correctness, and the quality matrix we have for the object-based evaluation average values were 97%, 99% and 99%, respectively. The second goal employs a deep learning(DL) algorithm to predict the best sensor for detection, either the LiDAR, optics or the fusion of the LiDAR and high-resolution aerial photography, to know which is most suitable for building detection with little or no user intervention. Whereas an acceptable range for good classifiers (TPR and TNR index) should be 100, none of those mentioned above was below the threshold of ninety. In contrast, we had 97%, 93%, and 91% for the pixel-based evaluation values, respectively, for the deep learning method. We tested on A1, A2, A3, and our discovery DSM had the highest accuracy compared to other sensors alone. For Area 1 (A1), a value of overall accuracy of 93.21%, with a kappa coefficient of 0.798. Also, the optics' overall accuracy value was 87.54%, and the kappa coefficient was 0.630. Whereas for the fusion, the overall and kappa coefficient here was A2(94.30%, 0.859).. in conclusion, the integration of LiDAR and Aerial photography outperformed all the optics and DSM. The weakness of the image and the LiDAR dataset has been compensated through their fusion. Moreover, the proposed model was evaluated on three building forms in different locations with different rooftops forms for this research; three forms of housing/building types were considered: the complex, high rise and single low detached apartment buildings only. The result was negligible over the study area by comparing LiDAR DEM heights and differential GPS. The.RMSE is 0.11 for the heterogeneous environment, and mixed building forms for high rise buildings form RMSE is 0.002 m for high rise buildings while for low residential apartments, our RMSE value Root means square error 0.003m. The studies show our models' capacity to improve urban building detection and automate building objects. It is an indicator of excellent performance. The proposed technique can help detect and solve urban building detection problems

Text OJOGBANE SUCCESS SANI - IR.pdf Download (4MB)

Actions (login required)

View Item