Pavement Crack Detection Using Deep Learning and a Combination of Smartphone and 3D Laser Data
Loading...
Author(s)
Roeser, Paul
Advisor(s)
Editor(s)
Collections
Supplementary to:
Permanent Link
Abstract
Automated pavement condition evaluation is essential for timely maintenance and management of pavement assets. 3D laser technology has become the mainstream technology in the US for state Departments Of Transportation (DOTs) to collect pavement condition data. However, it is expensive for local transportation agencies to collect them annually or bi-annually. Thus, this paper proposes the idea to combine high-resolution low frequency 3D data, collected once a year or two years with low-resolution high frequency collection from low-cost smartphone with 2D images and Inertial Measurement Units (IMU) data that can be collected quarterly. A case study was conducted using this vehicle, collecting 3D images, 2D images and IMU data simultaneously on a 4.12 miles roadway with interstate highways, city street and state-maintained roadways. This data was collected using Georgia Tech Sensing Van (GTSV) sponsored by US DOTs through the research project, which is using 3D laser and smartphone data on interstate highways and local streets. The architecture of a new automatic pavement evaluation pipeline using these low-cost sensors is proposed, as well as an initial focus study on automatic crack detection meeting the needs of these agencies. This study tests the limits of using transfer-learning of a model trained on accurate data and tested with images from a lower-cost smartphone camera. The results are promising and open the way for further research into the ideal methodology.
Sponsor
Date
2023-07-31
Extent
Resource Type
Text
Resource Subtype
Thesis