Automated detection of pavement patches utilizing support vector machine classification
| Autor: | Hadjidemetriou, Georgios M., Christodoulou, Symeon E., Vela, Patricio A. |
|---|---|
| Přispěvatelé: | Christodoulou, Symeon E. [0000-0002-9859-0381] |
| Rok vydání: | 2016 |
| Předmět: |
Engineering
Feature vector ComputingMethodologies_IMAGEPROCESSINGANDCOMPUTERVISION 0211 other engineering and technologies 02 engineering and technology Mathematical analysis Automation Histogram 021105 building & construction 0202 electrical engineering electronic engineering information engineering Discrete cosine transform Computer vision Texture Electronics and Communications Abstracts (EA) Block (data storage) Support vector machines business.industry Binary image Process (computing) Pattern recognition Classification Roads Support vector machine Electronics and Communications Milieux (General) (EA) [90] Images 020201 artificial intelligence & image processing Artificial intelligence Pavements business |
| Zdroj: | The Institute of Electrical and Electronics Engineers, Inc.(IEEE) Conference Proceedings. |
| DOI: | 10.1109/melcon.2016.7495460 |
| Popis: | The efficient condition assessment of road networks is crucial to prevent pavement distresses which can cause a spectrum of detrimental effects. The need for automation of the underlying process is originated from the costly, time-consuming and dangerous current methods. Presented herein is the automation of the patch detection process, which is essential for pavement surface evaluation and rating. The method is based on Support Vector Machine (SVM) Classification. The road pavement images are divided into square blocks and the SVM is trained and tested by feature vectors generated from these blocks. The feature vectors consist of the histogram and two texture descriptors, using the discrete cosine transform (DCT) and the Gray-Level Co-Occurrence Matrix (GLCM). The output is a binary image, where each image block is classified as "patch" or "no-patch". The performance of the proposed MatlabTM implementation, which uses data collected from real-life urban networks, is rated by a detection accuracy of 81.97 %, a precision of 64.21 %, and a recall of 91.21 %. 1 5 |
| Databáze: | OpenAIRE |
| Externí odkaz: |
|