Road detection by using gradient magnitude image and adaptive thresholding based Watershed

Abstract

In last decade, a lot of road detection algorithms are researched in remote sensing and aerial images, but they are not valuable for very dark and vague road images, hence, a new algorithm for road detection in a vaguer image is studied in this paper. It firstly uses a Gaussian filter to enhance the original image; then, an new edge detection algorithm is utilized to result in a gradient magnitude image by using different orientation information; subsequently, the road segments in the magnitude image is detected by applying a Watershed based algorithm, and a binary image is obtained; and last, in the binary image, the roads are finalized by removing noise in the image and filling gaps between the segments. To validate the proposed algorithm, the testing images were selected from both public dataset and our own dataset, and the testing results show that the new algorithm results are better than several traditional algorithms and recent semantic methods for the road detection, and the road detection accuracy between 92% and 95%. It is satisfactory for the road detection not only for normal images and also for the dark and vague road images. The new algorithm can be also applied into other applications, such as river detection and image enhancement, etc.