Asphalt Course Cracks Detection in Pavement Images Using Frequency Based Techniques

Abstract

Roads are an essential part of a healthy civil infrastructure of a country. Degradation is a natural force which affects and creates cracks in paved road surfaces. Pavement cracks are important measure to assess the road condition for necessary asphalt course maintenance. Timely detection of cracks is essential in order to prevent further road surface degradation, hence, reduces the maintenance cost. Manual surveys for road distress evaluation are time consuming and tedious. The Gabor filter being closely related to human visual perception is proposed as a suitable choice for automatic crack detection. The proposed approach is then compared with other frequency domain techniques. The Gabor filter based technique produced better results with 96.2% precision as compared to Fourier and Wavelet Transforms that show 89.5% and 85.6% precision respectively. The Gabor filter is then optimized using Genetic algorithm for automatic parameter selection. In this regard; PSNR based method, PSNR based crack template method, and SSIM based simulated crack method are proposed. Optimal parameters thus used to detect real cracks. SSIM based simulated crack method shows better performance with an overall accuracy of 97.1% as compared to PSNR based method and PSNR based crack template method that demonstrate 92.1% and 95.3% accuracy respectively. Optimized Gabor filter shows a promising future for crack detection in paved surfaces.