Towards Human-Centered Pavement Quality Annotation with Crowdsourcing

Abstract

The maintenance and rehabilitation of road pavements are critical to ensure safe and efficient transportation. To achieve this, accurate and up-to-date information on pavement conditions is necessary. Professional pavement monitoring is a time-consuming and expensive process. In this paper, we focus on the process of identifying and labeling pavement distresses, which we term as “pavement annotation”. We explore the potential of engaging drivers and passengers in pavement annotation with crowdsourcing to provide a cost-effective and scalable solution. Our approach involves developing a mobile application prototype that collects both mobile sensor data and human perception data during map navigation. We use an iterative design and development process to create a user-friendly interface that enables the efficient and effective annotation of pavement distresses. We evaluated the reliability of physical model indices computed with the mobile sensor data, the quality of human-labeled road anomalies, and the alignment of the two. Our findings suggest that while challenging, there is a great potential to augment the sensor and human data to generate rich pavement-quality annotation with crowdsourcing. We highlight the advantages and disadvantages of sensor-based and human-driven pavement quality annotation and draw design implications for crowdsourcing software and artifacts to enable safe, scalable, and sustainable pavement annotation.