Date

2019

Author

Ateş, Himmet

Metadata

Show full item record

Item Usage Stats

264
views

145
downloads

Cite This

When asphalt defects are detected and not corrected, they can cause accidents and loss of property and lives. Potholes formed in such asphalt surfaces are one of the biggest causes of accidents. In order to minimize the loss of life and property, the potholes formed on the asphalt should be detected and corrected by the authorities as early as possible. The potholes formed on asphalt surfaces can be detected either manually or automatically. Automated methods can be more time and cost effective. Vibration-based data processing, 3D reconstruction and processing, and image processing in 2D images are the basic methods used in automatic detection systems. In this thesis, the aim is to develop a system which is easy to apply and has low error rate by using ”Convolutional Neural Networks" methods that will be applied on 2D images. In classical machine learning methods, fixed (unchanging) features are extracted and classification methods (either static or dynamic) are applied through these features. The success of these methods depends on the accuracy, structure and quality of the extracted features as well as the applied algorithms. A Convolutional Neural Network is constructed and compared with classical machine learning methods, which are already applied for pothole detection problemin the literature, in terms of success rate and failure rate using the asphalt image sets. The different parameters of the convolutional neural network method are tested on the existing image sets and the effect of these parameters is also analyzed

Subject Keywords

Image processing., Convolutional Neural Networks., Image processing

URI

http://etd.lib.metu.edu.tr/upload/12623164/index.pdf
https://hdl.handle.net/11511/28091

Collections

Graduate School of Natural and Applied Sciences, Thesis