Research On Fast Stitching Method For Unmanned Aerial Vehicle Images

UAV image stitching aims to combine multiple images containing local information taken from multiple viewpoints on the ground by drones into a single panoramic image with a large field of view and high resolution.It has important applications in both military defense and civilian fields.Currently,feature-based traditional image stitching methods are mainly used for aerial image stitching,and many methods focus on improving the quality of stitching in order to achieve natural stitching effects.However,these methods usually take a long time and are difficult to meet the special requirements of emergency relief and other scenarios.Based on the background,this thesis studies the time-consuming image registration module in the feature-based image stitching framework,aiming to achieve a good balance between stitching quality and efficiency.The specific research contents and work are as follows:(1)Research on fast UAV image stitching method based on improved Oriented FAST and Rotated BRIEF(ORB)algorithm.In response to UAV images lacking prior information such as geographical location,this thesis proposes a fast drone image stitching method based on ORB features.Firstly,the ORB algorithm,which has good real-time performance,is used in the feature extraction stage,and its problems of few feature points and concentrated distribution are improved by using a variable threshold and introducing a quadtree structure.Secondly,the Grid-based Motion Statistics(GMS)algorithm is used in the feature matching stage to achieve fast and efficient filtering of mis-matches,providing better quality feature matching pairs for estimating the transformation parameters.Next,to solve the problem of error accumulation in multi-image stitching,a global registration strategy is adopted,and alignment constraints are constructed to optimize the transformation parameters.Finally,multi-threading technology is used to accelerate the feature extraction process.Experimental results show that compared with the Auto Stitch method,the method proposed in this chapter has certain improvements in both stitching efficiency and registration accuracy.(2)Research on fast stitching method for UAV images based on Position and Orientation System(POS)information.For scenes where POS data is available for UAVs,this thesis proposes a fast stitching method for UAV images based on POS information.Firstly,in order to address the issue of invalid calculation during the matching process of multiple images,GPS information is used to estimate the overlap relationship between images,so as to quickly determine the necessity of matching.Secondly,considering that the pose-based registration method does not have the problem of error accumulation,a transformation model is established based on the imaging process of UAV images.The camera intrinsic and extrinsic parameters are initialized using POS information,providing a good initial value for global optimization.Finally,in order to improve the accuracy of registration,the camera parameters are optimized by minimizing the projection error.The experiments show that this method can effectively reduce error accumulation and improve running efficiency.It can stitch multiple high-resolution aerial images covering an area of approximately 1.5 square kilometers in about 5 minutes.In summary,this thesis proposes two fast UAV image stitching methods for different scenarios.While ensuring the stitching effect,they effectively improve the efficiency of stitching and can basically meet the demands of emergency situations.