Road networks in most developing nations like Nigeria, are characterized by the presence of anomalies such as potholes, speedbumps, rutting, cracks among others. This anomaly is usually caused by the poor drainage system, asphalt road exceeding their design life span, excessive traffic and atimes the use of poor-quality materials for road construction. Despite efforts by appropriate agencies to rehabilitate the anomalous road networks, the anomalies still persist particularly potholes. Thus, the need to equip vehicles with the capability of sensing and notifying drivers of the presence of this anomaly, towards making the appropriate decision of either slowing down before encountering the anomaly or avoiding it. In this regard, this paper presents the preliminary results obtained towards the development of a robust vision processing based approach for potholes anomaly detection that is independent of the illumination intensity during data acquisition. The proposed approach utilized the median filter for denoising the image, discrete wavelet transforms in deblurring and preserving the edges of the anomaly, while canny edge detection algorithm was used for segmenting the image and extracting features used in training a Convolutional Neural Network (CNN) for potholes anomaly detection and classification. The preliminary results obtained indicate the potholes anomaly were detected and classify accordingly with about 96% accuracy, 95% precision and low false alarm rate of about 5%. This indicate the potential of the proposed approach to be used for real-time potholes anomaly detection and notification system. Also, it can be incorporated into manned and unmanned vehicles towards aiding navigation in such an anomalous road terrain.
Key words: Anomaly, Canny-edge-extractor, CNN, Pothole, and Road.
|
