Research On Vehicle Targets Extraction Method Of The Urban Road From Satellite Images

With the development of remote sensing technology and the improvement of the spatial resolution, satellite-based object recognition and extraction of the vehicle and its potential application gradually become the forefront of international attention. However, the existence of shadows in the satellite images, especially the shadows formed in the high-density urban environments due to tall buildings and other surface features blocked the sunlight, greatly reduced the optical information of the vehicle targets in the images, reduced the vehicle targets extraction efficiency and the extraction accuracy. Based on this, with WorldView-2satellite images data as an example, on the basis of the existing conventional road vehicle targets extraction technology, combined with shadow processing technique and image enhancement methods of remote sensing images, the vehicle targets extraction of the urban road included shaded area can be divided into the detection of shadows and extraction of the vehicle targets two stages to study.First, based on the HIS color enhancement technology, the image is converted from the RGB color space to the HIS color space, analyzed the hue(H), intensity(I) and the saturation (S) of the shaded aera, and combined with the band calculation method build the shadow index (NDUI), the ratio r, the subtraction SD, three features to describe the image features of the shaded area;Secondly, make full use of the spectral information, form information and texture information as well as information about the vehicle neighbor relations of the unshaded aera and the shaded aera vehicle targets, vehicle targets are divided into brightness vehicle targets within the unshaded area, dark vehicle targets within the unshaded area, brightness vehicle targets within the shaded area and dark vehicle targets within the shaded area four types to analyze its images features;Thirdly, according to the images feature analysized results of shaded area and vehicle targets, combined with object-oriented remote sensing image object extraction technology to build a vehicle targets extraction process of the urban road included shaded aera, analyzed and determined the multi-scale image segmentation parameters, build the image object classification hierarchy, determined the rule set of shaded area detection and vehicle targets extraction;Finally, take a WorldView-2satellite remote sensing data in Zhengzhou city as an example, to verify the vehicle targets extraction process of the urban road included shaded area has been established in this paper and through the establishment of confusion matrix for the extraction results calculated the overall accuracy, Kappa coefficient, precision of producers and users of the extraction results to analyse and evaluate the accuracy.