Research And Implementation Of Aerial Image Mosaic Algorithm

Recently,with the development of UAV(unmanned aerial vehicle)aerial image technology,the research for aerial image stitching has received increasing attention.Since there is the limitation for the sensor size of the image capturing apparatus,it is hard for a single captured image to cover the entire area,which means the aerial image stitching is necessary.In terms of the characteristic of aerial image technology,this paper aims to improve the conventional image stitching algorithm,reduce the consumption of resources,and enhance the efficiency of image stitching.First of all,this paper summarizes various ways to obtain aerial image as well as its applications.The detailed process of image stitching and its different application scenarios are introduced;The state-of-the-art of the feature points detection is presented,which is the most essential part in image stitching;The possible problems in image stitching are illustrated,and the state-of-the-art of the aerial image stitching area is displayed.Moreover,the image preprocessing part of the image stitching is introduced.The model of aerial image acquisition is constructed,and the image transformation matrix in the process of image mosaic is optimized according to the imaging model of aeria l image.The occupancy for computing resources and the time-consuming of each phase of image stitching is tested,and the stitching work of actual aerial image is accomplished.In addition,The feature extraction process of the feature points extraction method in the process of image mosaic is simulated by experiment.According to the experimental results,the feature points detection and the feature descriptor extraction part with the most computational resources are calculated.In the feature point detection part,because the splicing object itself is not large scale transformation,so the differential Gaussian pyramid to reduce the layer,greatly reducing the construction of differential Gaussian pyramid need to consume the computing resources.In the feature description sub-extraction part,the sub-region of the sampling point is re-divided,and the concentric circle is used instead of the rectangular partitioning method in the original algorithm,which further highlights the influence of the sampling point neighborhood on the feature descriptor extractionFinally,according to the above for the splicing algorithm optimization design ideas,completed the aviation image of the splicing program.In the feature point detection stage,according to the feature point matching rate and splicing effect to select the appropriate Gaussian pyramid structure parameters.Using the F LANN algorithm to match the extracted features,and use the RANSAC method to screen out the wrong match pairs,and finally through the use o f transitional conversion matrix to complete the aerial image of the multi-map stitching.