Segmentation and Detection of Road Region in Aerial Images using Hybrid CNN-Random Field Algorithm

Abstract

Road detection and segmentation is an important aspect in navigation system and is widely used to detect new roads and patterns in the region. These system has the main objective to help navigate the autonomous vehicle and robot on the ground. Road detection is very useful in finding valid road path where the vehicle can go for supportive vehicles preventing the collision with the obstacles, object detection on the road and other necessary information exchange. It has a variety of uses such as the disaster monitoring, traffic monitoring, crop monitoring, border surveillance, security and so on. There are several techniques used for detection and segmentation purpose of roads such as Artificial Neural Network, Support Vector Machine (SVM), Self-Organizing Map (SOM), Convolution Neural Network (CNN), and Deep learning techniques. In this paper, a new technique for road detection and segmentation is proposed which includes a combination algorithm of CNN and Random Field segmentation for road maps using aerial images. This road detection and segmentations give alternative solution for road classification and detection with a higher accuracy. In this system normally accuracy (ACC) have an average range of 97.7%.