Low altitude multispectral mapping for road defect detection

Abstract

Pothole’s defect is major damages indicated the road condition visually, and the structural defects due to some potential causes. Nowadays, new forms of remote sensing technique were widely used, but less studies in the application of low altitude multispectral mapping. The potential of multispectral images is its help better in resolution due to its spectral characteristic. Hence, it helps a lot in feature classification with proper training sample, classifier used, and spectral band composite. Thus, this study aims to extract the defective roads by using the multispectral image of Parrot Sequoia with low flight altitude. This study tries to detect a pothole's existence from band combination and supervised classification other than its common use which ultimately for agriculture purposes. The classifier used in this is Maximum Likelihood, Support vector machine (SVM) and Mahalanobis Distance. 15 different probability of band stacks of green, NIR, red edge, and red band were used as multispectral images. The comparison of the performance between the types of classifier and band combination was modeled and discussed in this study. Classifier algorithm maximum likelihood gives the lowest error of 0.108m² with a combination of NIR + red edge band. SVM gives the lowest error of 0.427m² with a combination of green + NIR + red edge + red band. While Mahalanobis distance gives the lowest error of -0.082m² with a combination of red edge + red band. Averagely, Mahalanobis distance gives the lowest error of 0.299m² of all bands used.