Vehicle detection using improved morphological reconstruction for QuickBird images

Abstract

Detecting vehicles from satellite images is of great importance for both traffic monitoring and urban planning. However, it is still a challenging task, because vehicles are small and the background is complex. To deal with the issue, an improved morphological reconstruction approach is proposed to model the complex background in satellite images. Bright building regions are first removed by the normalized difference built-up index. Directional filters are next designed to generate the marker image for morphological reconstruction. As the directional filters can match any directional structure of the background, the newly proposed approach has a good ability to eliminate the complex background in satellite images. Finally, vehicles are detected using the Reed–Xiaoli algorithm. The proposed method is applied to real QuickBird images for vehicle detection. The results show that the proposed approach has strong robustness and high efficiency merits and can be used for vehicle detection in city areas.