Unmanned aerial vehicle image rapid registration method based on GPU + feature recognition

Abstract

The invention provides an unmanned aerial vehicle image rapid registration method based on GPU + feature recognition. The method comprises the following steps: 1, receiving image data acquired by aerial photography of an unmanned aerial vehicle in real time; 2, using a GPU accelerated KAZE algorithm adjacent images, and completing feature point extraction, feature point descriptor calculation andfeature matching in a GPU; 3, matching the feature points extracted from the adjacent images by GPU acceleration feature points, and carrying out feature matching through a matching method based on the ratio of the nearest neighbor distance to the next nearest neighbor distance; 4, selecting the first n points with higher matching quality as high-quality matching point pairs by using a PROSAC algorithm; 5, accelerating feature point matching by using the GPU according to the high-quality matching points stored in two images; 6, estimating homography matrixes of the images by adopting a PROSACalgorithm, and 7, realizing rapid registration of the images according to the homography matrixes among the images.

Description

technical field [0001] The invention relates to the technical field of image processing, in particular to a method for fast registration of UAV images based on GPU+feature recognition. Background technique [0002] Natural disasters occur frequently in our country, and the use of conventional aerospace photogrammetry means has a long period and high cost, and cannot meet the requirements of emergency surveying and mapping and small-area high-resolution geographic information data update in a timely and effective manner. A powerful supplement to the measurement method, the aerial photography image obtained by the camera carried by the UAV can not only view the acquired data in real time, obtain the real-time situation of the research area, but also perform rapid image registration on the received image synchronously. It is used in environmental monitoring, supervision and law enforcement, border security control, farmland crop monitoring, daily standard detection, constructio...

AI Technical Summary

Problems solved by technology

This method is robust to image rotation, scaling, scale transformation, and illumination differences, but it also has disadvantages such as complex principles, large calculations, and long time consumption.

The existing UAV image registration method has low computational efficiency, consumes a lot of time, cannot meet the real-time requirements, or loses a large amount of image detail information, which is not conducive to subsequent practical applications

Registration methods in the prior art more or less have the disadvantages of unsatisfactory registration effect and low registration efficiency

Images