ASPHALT ROAD POTHOLE IMAGE DETECTION USING DISCRETE WAVELET TRANSFORM

Abstract

Potholes have been one of the major problems faced by road users globally, contributing to vehicular accidents on roads, quick wear and tear of vehicles, among others. This is a big menace to the economic growth of any nation. Overtime, advancement in vehicular technology and sensors, has led to the establishment of automated approach for detecting road anomalies. Though, these approaches can be categorized into 3D reconstruction-based approach, 2D vision-based approach and the vibrational-based approach. However, the 2D vision-based approach has received wide acceptance among the academia and the industry due to its intrinsic advantages over the 3D reconstruction-based and vibrational-based approaches. In this regard, this research work focuses on the detection of potholes on asphalt roads which is one of the prominent road anomalies, by improving on the detection accuracy of 2D vision-based approach, by presenting a 2D vision-based pothole detection algorithm based on discrete wavelet transform. A total of 400 road surface images were captured, pre-processed and segmented, using discrete wavelet transform and canny edge extractor. A deep learning approach using a pre-trained Convolutional Neural Network (CNN) was adopted for detection and classification of the segmented road images. The results obtained showed that this algorithm is able to detect potholes more accurately even during light illumination and foggy weather conditions, than proposed 2D vision-based techniques in literatures, having an overall accuracy of 93.33%, precision of 91.67%, and recall of 94.83%.