Date of Award
Doctor of Philosophy
Electrical and Computer Engineering
Dr. Ikhlas Abdel-Qader
Dr. Janos Grantner
Dr. Osama Abudayyeh
Dr. Jun-Seok Oh
Transportation departments are required to monitor the condition of road signs through appropriate condition assessment mechanisms to improve road safety and keep drivers properly informed. Typically, these mechanisms include visual inspection or specialized equipment such as retroreflectometers. These methods are costly, tedious, and risky since they need direct contact with road signs. Efforts to use emerging computer vision techniques for the assessment of road signs condition combined with the availability of road data inventories are allowing the automation of these processes, thus easing the inspection process, reducing costs of equipment, and decreasing the risks associated with the need for maintenance crews to be in direct contact with road signs.
In this dissertation, a system is developed for automating road sign condition assessment using imaging databases. The system integrates several local discriminative features with support vector machine (SVM) methods to generate condition information on road signs. This work has several contributions that have been demonstrated by experimental results, including: 1) improving the detection and shape recognition results even when a road sign is partially occluded or tilted; 2) the ability to identify tilted road signs; 3) the ability to evaluate the condition of road signs in terms of partial occlusion and partial breakage; 4) the ability to evaluate possible road sign vandalism; and 5) the ability to evaluate road sign deterioration in terms of quality and readability.
This system has been tested using images from three different road sign databases. The experimental results successfully demonstrated the efficiency of the automated road sign condition assessment system proposed in this dissertation.
Abukhait, Jafar Jameel, "A Discriminative Imaging-Based Framework for Road Sign Condition Assessment Using Local Features and SVM Classifiers" (2012). Dissertations. 43.