University of Twente Student Theses


Fully convulutional networks for street furniture identification in panorama images

Ao, Ying (2019) Fully convulutional networks for street furniture identification in panorama images.

[img] PDF
Abstract:Panoramic images are used increasingly wide in the past years. They provide users with broader viewing angle than normal perspective images and the cost is relatively low, which makes them suitable in many scenes, especially in virtual reality and street view capture. However, they are new for street furniture identification which is usually based on mobile laser scanning point cloud data or conventional 2D images. This study proposes to extract street furniture information from omnidirectional images and the transformed images of them. The transformation is implemented from four directions using Gnomonic projection.To separate light poles and traffic signs from background, scene understanding methods are needed. In this study, semantic segmentation is performed on the images and the pixel-level task is implemented by pre-trained Fully Convolutional Networks (FCN). FCN is the most important model for deep learningapplied on semantic segmentation for its end to end training process and pixel-wise prediction. Then the focal loss function, which is known for solving the class imbalance problem, will be introduced in the FCN model to improve the results. In the experiment, we use FCN-8s model that pre-trained on cityscape dataset and finetune it by our own data. Then replace cross entropy loss function with focal loss function in the FCN model and train it again to produce the predictions. The evaluation of predictions shows that in all results from pre-trained model, fine-tuning, and FCN model with focal loss, the transformed images have better performance than panoramic images. And the average accuracy and mean IoU of the results are gradually improved in the three phases.
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:
Export this item as:BibTeX
HTML Citation
Reference Manager


Repository Staff Only: item control page