3D Reconstruction for Metrology and Visualization of Potholes using single camera

Abstract

Machine vision based evaluation systems possess great potential for automated quality inspection applications. High resolution vision sensors and robust vision algorithms have rendered reliable and low cost vision based systems. The proposed research work presents a simple and novel approach for 3D reconstruction of potholes for an automated pavement inspection and evaluation system. The technique utilizes Structure from Motion (SfM) based 3D reconstruction algorithm along with laser triangulation principle to generate metric 3D point clouds of potholes. The algorithm takes pothole images as input captured using a calibrated single camera. Laser triangulation is achieved through a calibrated laser pointer attached to the camera at fixed distance. Metric 3D point clouds are post processed in order to measure different pothole parameters like area, volume and average depth. Measurements of these features help in estimating road distress severity. The mean percent error for depth measurement of artificial potholes is found to be 3.36 % whereas mean percent error for natural potholes is found to be 10.32 %. The proposed technique shows promising results to generate metric 3D point clouds of potholes.