The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
Publications Copernicus
Download
Citation
Articles | Volume XLII-4/W4
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-4/W4, 299–303, 2017
https://doi.org/10.5194/isprs-archives-XLII-4-W4-299-2017
Int. Arch. Photogramm. Remote Sens. Spatial Inf. Sci., XLII-4/W4, 299–303, 2017
https://doi.org/10.5194/isprs-archives-XLII-4-W4-299-2017

  27 Sep 2017

27 Sep 2017

A FUZZY AUTOMATIC CAR DETECTION METHOD BASED ON HIGH RESOLUTION SATELLITE IMAGERY AND GEODESIC MORPHOLOGY

N. Zarrinpanjeh1 and F. Dadrassjavan2 N. Zarrinpanjeh and F. Dadrassjavan
  • 1Department of Geomatics Engineering, Qazvin Branch, Islamic Azad University, Qazvin, Iran
  • 2School of Surveying and Geospatial Engineering University College of Engineering University of Tehran

Keywords: Car Detection, Fuzzy Inference System, Geodesic Morphology, Satellite Imagery

Abstract. Automatic car detection and recognition from aerial and satellite images is mostly practiced for the purpose of easy and fast traffic monitoring in cities and rural areas where direct approaches are proved to be costly and inefficient. Towards the goal of automatic car detection and in parallel with many other published solutions, in this paper, morphological operators and specifically Geodesic dilation are studied and applied on GeoEye-1 images to extract car items in accordance with available vector maps. The results of Geodesic dilation are then segmented and labeled to generate primitive car items to be introduced to a fuzzy decision making system, to be verified. The verification is performed inspecting major and minor axes of each region and the orientations of the cars with respect to the road direction. The proposed method is implemented and tested using GeoEye-1 pansharpen imagery. Generating the results it is observed that the proposed method is successful according to overall accuracy of 83%. It is also concluded that the results are sensitive to the quality of available vector map and to overcome the shortcomings of this method, it is recommended to consider spectral information in the process of hypothesis verification.