Research And Implementation Of Rapid Registration Of Aerial Flood Images Based On GPU_SURF

The application of image processing technology is widely used in aerial disaster image registration.At present,the need for timeliness and accuracy of image processing with high resolution and large amount of data is more and more urgent.Through the image registration technology,the super clear disaster scene image obtained by aerial photography is registered and spliced,and the high-definition and high-quality global image information obtained is of scientific significance for emergency rescue and disaster analysis.According to the characteristics of the acquired image data,such as uneven illumination,excessive or insufficient exposure,rotation of view angle,large amount of data,etc.,this paper starts from the fast image registration and splicing to reduce the influence of these factors,and analyzes and selects the appropriate image preprocessing method,parallel calculation optimization of feature point extraction algorithm,efficient feature point matching and error elimination Finally,the image registration and splicing system based on GPU_SURF algorithm is designed and implemented in combination with the above research.The specific work of this paper is as follows:(1)In order to reduce the redundant and invalid data in the image information,the input image is grayed out by weighted average filtering,and the denoised image enhancement operation is performed on the input image using improved median filtering.Preprocess the input image,remove noise,improve the contrast of the image,enhance the detectability of the relevant information,and improve the image quality.(2)In order to improve the speed of feature detection and description algorithm,this paper improves and optimizes the GPU parallel computing based on the traditional surf algorithm,makes full use of computer resources,builds the integral image on the GPU through the subsection prefix addition,and then uses the GPU parallel core to build the scale space and calculate the feature descriptor.This improved algorithm uses GPU to accelerate the algorithm,and the speed is nearly 10 times faster than the traditional surf algorithm.(3)In order to improve the accuracy of image registration,this paper uses FLANN's fast nearest neighbor algorithm k to fine screen the feature pairs,and then uses the progressive consistent sampling method to remove the false matching of feature pairs,and obtains a largenumber of matching feature pairs with high accuracy.The image registration rate is nearly30% higher than the traditional method.The optimization algorithm is used in the registration experiment of the aerial images of the flood area,and the algorithm speed and the registration results have achieved the expected experimental results.(4)In order to maximize the convenience of users,this paper integrates the previous work to further the image registration operation,and designs and realizes the image registration system based on the optimization algorithm of feature matching.