Research On UAV Remote Sensing Image Registration Algorithm

Along with the development of socio-economic, unmanned aerial vehicle remotesensing technology has been widely used in the field of geological environment, disasterinvestigation, land use dynamic monitoring and updating of topographic maps relyingon its flexible and efficient characteristics. Because of changes in surface morphology,climate change, the effects of high altitude air currents, the images got by UAV havesome problems that frame size is small, amount is too many and forward overlap andside overlap are irregular. Realizing the efficient registration is the difficulty ofapplication. Effect of image registration will directly affect the quality of the subsequentimage processing, so to research a algorithm suitable for unmanned aerial vehiclesremote sensing image registration has important theoretical and practical significance.This paper proposes an improved algorithm based on the research andimplementation of traditional SURF algorithm. The algorithm of this paper is improvedin two ways, first of all, for their deficiencies in feature extraction stage: a small numberof feature points and the feature points extracted does not reflect the structuralcharacteristics of the image. Proposing a method of Harris corner detection inmulti-scale space, and this way can keep the scale invariance and stable number offeature points. Then, because more time is wasted in the feature point matching stageusing the exhaustive search method, the paper uses the KD-Tree search strategy insteadof the original strategy to improve the real-time of algorithm. The specific content of thealgorithm proposed by this paper is: first, build the multi-scale space by integral Imageand box filters of SURF algorithm, and detect the Harris corner points in multi-scalespace for getting features points; Then determine the main direction by computing theHaar Wavelet response value of the feature point, and generated64dimension SURFdescriptor; Finally, use KD-Tree search strategy to find the nearest and next nearest neighbor points, and compute ratio of the distance between the nearest neighbor pointsand features points and the distance between the next nearest neighbor points andfeatures points. If it is less than a given threshold, then regard the nearest neighbor pointas the matching point.Experimental results show that, compared with the traditional SURF algorithm, theimproved algorithm improves matching accuracy at the time of guarantying real-timeperformance. And this paper ensures the accuracy and timeliness of unmanned aerialvehicle remote sensing images registration.