University of Twente Student Theses

Login

Building segmentation in oblique aerial imagery

Huang, Shan (2019) Building segmentation in oblique aerial imagery.

[img] PDF
2MB
Abstract:With the explosionof urbanization, the demand ofcity planning has been increased. These new challenges have to be facedin regard tothe planningand environmental sustainability or urban areas. To tackle these problems, the use of more detailed and complete geographic information is necessary. “Smart Cities” aim at delivering smartand complete information thanks to digital technologies.Andthe buildingis a sub-problem andit is a keycomponent to the reconstructing of LoD3 city modeling. In the past, the datato generate a 3Dbuilding model were almost basedon terrestrial views. However, with the development of imagematching technique, the airborne systems have been applied in many tasks to acquire airborne multi-view data. Compared to terrestrial views, airborne datasets can cover larger areasin the urbanareasand italso been found more convenientand economic. In my study, the oblique aerial images acquired from obliqueairborne systems are usedas a datasource for building segmentation. With the popularity of DeepLearning, tasks in the field of computer vision can be solvedin easierand effectiveways. Fullyconvolutional network is an end-to-end and pixel-basedneural network, itshows a good performance in semantic tasks to get a denseprediction result. In this study,we propose a method toapply deep neural networks to buildingsegmentation. In particular, the FC-DenseNet and the DeepLabV3+ networks are used to segment the building from aerialimages and get semantic information such aswall, roof, balcony and openingarea(window and door). Due to the limited computation resource, the patch-wise segmentation is usedin the training and testing process toget information at pixellevel. To address the problem of imbalancedclasses, the weighted loss function is usedin the experiment instead of the commonloss function. Softmax function is usedasto reconstruct from patches to original images. Different typologies of input have been considered: beside the conventional2D information (i.e.RGB image), we combined 2D information with 3D features extracted from dense image matching point clouds to improve the performance of the segmentation. Experiment results show that FC-DenseNet trained with 2D and 3D features achieves the best result, IoU up to 64.41%, it increases 5.13% compared to the result of the same model trained without 3D features(59.28%).The overall accuracy is increasedfrom 89.08% to 91.30%. Resultson roofin FC-DenseNet and DeepLabV3+ using 2D combined with 3D features arebetter than these two models only trained with 2D information: the class accuracy increased from 91.76% to 94.61% and 92.25% to 95.78% respectively. In conclusion, 3D features give benefit on the performance of segmentation. It can improve the performance of specific classes, for this thesis, the third component of the normalvector provides extra information to distinguish if the pixelon the same plane.
Item Type:Essay (Master)
Faculty:ITC: Faculty of Geo-information Science and Earth Observation
Programme:Geoinformation Science and Earth Observation MSc (75014)
Link to this item:https://purl.utwente.nl/essays/85876
Export this item as:BibTeX
EndNote
HTML Citation
Reference Manager

 

Repository Staff Only: item control page