01627621

Component

Potholes cause diverse problems such as car accidents and damaged wheels. A pothole detection algorithm is an essential part of an automatic pothole maintenance systems. Typically, potholes have been detected by manual counting using humans, which is expensive and slow. Recently, pothole detection systems based on video cameras have been studied for fast and inexpensive pothole detection. In previous work, the authors developed an algorithm using video data that accurately detects potholes using motion and intensity features. The detection algorithm was highly accurate, but it could not detect potholes online because it was computationally heavy. Thus, the authors also developed a lightweight pothole detection algorithm for online operation, but it obtained incorrect detection results in certain situations such as potholes overlapped by shadow and those surrounded by similar intensity values. Thus, this paper proposes a pothole detection algorithm based on saliency map information in order to improve the previously developed lightweight algorithm with minimum additional computation time. Experimental results show that the proposed algorithm outperforms the previous lightweight algorithm with only a small amount of additional computation that is suitable for online pothole detection.

This paper was sponsored by TRB committee AFD20 Standing Committee on Pavement Condition Evaluation.

01618707

17-01514

English

14p

2017

Transportation Research Board 96th Annual Meeting

Digital/other

Figures; References; Tables

Algorithms; Detection and identification systems; Image analysis; Potholes

Data and Information Technology; Highways; Pavements

Transportation Research Board Annual Meeting 2017 Paper #17-01514

TRIS, TRB, ATRI

Dec 8 2016 10:29AM