3D laser-based pavement imaging systems have been widely adopted by highway agencies in the last decade for automated pavement condition assessment while 2D imaging technologies and smart phone are also used to perform their pavement condition evaluation, especially for local transportation agencies. These collected pavement images are then used to extract pavement distresses semi- or fully automatically through various methods. Among these methods, models based on Artificial Intelligence (AI) with Machine Learning and Deep Learning (ML/DL) have gained high attention for pavement distress identification in the last several years. However, most AI models either have not yet fully connected with how highway agencies use the pavement distress data or have not been sufficiently developed to rely on quality 3D pavement image data.
The collected distresses such as cracking, faulting, flushing, and raveling are key indicators for triggering pavement maintenance and rehabilitation activities. Without clearly understanding the ultimate usage of this distress data by state agencies, the AI model development efforts for distress detection and/or classification, which includes AI model formulation, distress annotation, training, and performance evaluation, could be misguided thus failing to reach their full potential. For example, the AI-based models for automated crack detection using the classification of image blocks having cracking distress may not be able to output accurate cracking length and width information. Therefore, the produced model outcome may not meet the state agencies’ need for project-level applications such as planning crack sealing projects.
On the other hand, the performance of supervised-learning AI models for automated pavement distress extraction heavily relies on several factors including the used pavement image data quality, data size and diversity, the annotation quality (labeled ground truth distresses), the model formulation, model training, etc. However, the performance evaluation method used for many developed models is not always clear, especially for the diversity of the data used for that evaluation and its established ground truth. This ambiguity makes the comparison of the performance of different models challenging and unreliable.
The objective of this synthesis is to document state DOT current practices of automated pavement distress identification and AI (ML/DL) technologies for pavement condition evaluation.
Information to be gathered includes (but is not limited to):
• Requirements for automated pavement distress identification;
• Various applications of pavement distress condition information;
• Types of agency decision-making supported by pavement condition data;
• Artificial intelligence (e.g., machine learning, deep learning) technologies, tools, and models currently being applied to pavement distress detection and classification and to pavement condition evaluation; and
• Ground truth / reference / benchmark data used in AI-technique development, training, and evaluation.
Information will be gathered through a literature review, a survey of state DOTs, and follow-up interviews with selected agencies for the development of case examples. Information gaps and suggestions for research to address those gaps will be identified.
Information Sources (Partial):
• Pierce, Linda M., and Weitzel, Nicholas D. NCHRP Synthesis 531: Automated Pavement Condition Surveys. 2019.
• Hsieh, Yung-An, and Tsai, Yichang James. “Machine Learning for Crack Detection: Review and Model Performance Comparison”. Journal of Computing in Civil Engineering, September 2020.
First Panel: TBD
Teleconference with Consultant: TBD
Second Panel: TBD