Research on Vehicle Detection in High Resolution Satellite Images

Abstract

With the improvement of satellite resolution and the object-oriented detection method in satellite images, traffic data can be more quickly and widely acquired in large area satellite images compared with the traditional data acquired method. With the technology of image enhancement, the paper improved the image quality first, and then utilized the multi-scale segmentation technology and supervised classification method to detect the vehicle from satellite images. In the process, three classification decision trees for vehicles in different situations have been summed up. At last, the paper has achieved the empirical research using the remote sensing images of typical regions in the urban road from Worldview-2 and the GeoEye-1. Based on the precision analysis of the experimental results, it shows that the average accuracy is more than 90%.