Automatic traffic monitoring using neural networks from satellite images

Considering the widespread problems of road transport, approach of the paper is a system to automatically control the roads by using images from satellite in night and day. Although no coherent system with appropriate performance has been yet introduced to achieve this goal, some methods has been proposed to estimate the road or recognize objects on the road, which have been more based on thresholding and color-based object recognition; and therefore, their efficiencies have direct relationship and a lot of dependence with the type of input image. In this paper, a complete and coherent system has been introduced to detect traffic by using satellite images, in which a special attention is paid to extraction of the road and vehicles on the road by using image processing and machine learning (including feature extraction, morphology methods, and algorithms of labeling); and rate of road traffic is estimated by using the obtained results and by using neuro-fuzzy network. Previous works has been introduced in the paper; and finally, the obtained results have been compared with the past appropriate methods. Higher accuracy and less dependence on the input image is among the results that have been explained in detail in last section of the paper in which and results for various images from satellite show an accuracy of about 85%.