Automatic Detection of Pavement Surface Defects Using Consumer Depth Camera

The detection of pavement surface defects is a critical task in assessing and monitoring pavement condition. A number of image-based approaches have emerged in recent years to automate the process of surface defect detection. A prominent technology is the consumer depth camera, a camera that captures not only images but also the varying distance between scene objects and the camera itself in a format of 3D point clouds, similar to 3D point clouds obtained via light detection and ranging (LiDAR). The main challenge in applying a consumer depth camera to pavement defects detection lies in the low resolution and high random noise of the raw data. This paper presents a method that effectively addresses this challenge. It starts by filtering the signal noise using disciplined convex optimization and then enhances the resolution via moving least square up-sampling. This newly created method was tested in the field to detect different types of pavement defects including cracks, ruts, and potholes. It was found to be effective in detecting and quantifying pavement defects to provide quantitative information to support informed decision making in pavement management.