Efficient Transportation System by Automated Road Extraction Model from High Resolution Satellite Images

Abstract

A road extraction is a computer vision or machine learning technique created to recognize and classify routes or road sections within a given image or geographic information. Typical applications for these models include navigation, traffic monitoring, urban planning, map production, assisting businesses and governments in decision-making, transportation system optimization, and situational response. As a result, urban planning, transportation, and disaster response are all improved, but still need to improve the processing time and make autonomous process to use in real time. The proposed methodology uses the Mask RCNN Algorithm with improved Resnet backbone architecture to improve the result of road extraction. In this study, the mAP (Mean Average Precision Value) matrix is used to evaluate the accuracy of results. The proposed methodology results show good accuracy as compared to traditional algorithms or semiautomatic algorithms. This study uses the Massachusetts road dataset for result evaluation. The proposed methodology reduces the false detection rate and improves accuracy with processing time cost. The proposed methodology achieved mAP 0.90 for result accuracy, which is better than the traditional methods.