Advanced Road Scene Image Segmentation and Pavement Evaluation Using Neural Networks

Principal Investigators

Dr. Ezzatollah Salari
The University of Toledo

Co-Principal Investigators

Dr James Lynch
University of Detroit Mercy

Co-Investigators

Dr. Eddie Chou
The University of Toledo
Dr. Utpal Dutta
University of Detroit Mercy

Project Dates:

09/01/2009 to 08/31/2010

Project year: Year 3

MIOH-UTC Project Identifier: TS 18; Project 3

Focus Area:

Research: Intelligent Transportation Systems

Abstract:

Transportation infrastructure maintenance and repairs are a major source of expenditure for state and federal agencies. Due to the rapid deterioration of pavement structures, billions of dollars are allocated yearly for road maintenance and rehabilitation. One of the most important tasks in pavement maintenance is pavement surface condition evaluation distress measurement. To eliminate the tedious and unreliable manual inspection, image processing and pattern recognition techniques are used to increase the efficiency and accuracy while decreasing the costs of pavement distress measurements.

In this current phase of the project, the researchers have concentrated on the development of state-of-the-art image processing techniques to analyze and extract features from pavement surface images. Specifically, they have developed image processing procedures, including, nonlinear filtering, thresh holding, morphological filtering, skeletonization, and image transformation, to extract the surface cracks from the pavement images.

The proposed project is a continuation of our current MIOH-UTC project, extending the scope of the investigations in providing a more generalized solution in a less restricted environment. In the proposed study, the researchers consider a road scene containing grass, trees, buildings and other objects in addition to the pavement itself. In this context, they use both the color and texture information to extract the pavement regions. Afterwards, neural networks are designed to process the pavement images and then they are used as a decision tool to provide a classification for various types of cracks.