Development of Asphalt Paved Road Pothole Detection System Using Modified Colour Space Approach

Abstract

Asphalt paved roads in good condition are important infrastructure that contributes to a large extent the development of a nation. The technical cost and time required to manually assess asphalt paved road conditions have limited proper and timely planning in most developing countries. This paper presents an automated pothole detection system using modified colour space technique. The modified hybrid colour space model form was used to transform the original asphalt pavement image to grey level image. The contrast of the grey level image was then enhanced to improve its visual quality and detection accuracy. The enhanced grey level image was later segmented using thresholding technique. An evaluation of the accuracy of the developed system in the detection of potholes on asphalt paved road was conducted; the result of the segmented image was compared with the ground truth under two conditions. The accuracy of 92% was obtained in the first condition where the system was able to detect potholes in the acquired image, in comprising with related literatures the detection accuracies are not given in percentages except [18] that has accuracy of 93% and is not for detecting pothole but evaluation of road defects. In [19, 20, 21, & 7] pothole detection accuracy are given as reasonable accuracy but not in percentage. Further quantitative performance analysis of the detected region was conducted using Mean Square Error (MSE) metric. The results obtained shows that the detected areas gave minimum MSE when compared with the ground truth. The research reduces the complexity of potholes detection systems using an extremely intuitive colour space model that reduces the stages involved in image processing technique for pothole detection compared with other methods from related literature which uses; 3D reconstruction method, vibration method, video images using artificial neural network models, machine vision, histogram shape based thresholding and spectral clustering from histogram data obtain from gray scale images. Some of the aforementioned methods required high cost equipment, others are computational complex because they required features extraction and training phase. The use of extremely intuitive colour space model of this paper does not required high cost equipment, filtering and training phase, hence the reduction in computational complexity of the method.