Design And Implementation Of Registration Algorithm For Multimodal Video Streams

In order to make the UAV overcome the misjudgment caused by such bad weather at night,rain and fog,the target camouflage and so on,and achieve the purpose of 24/7 monitoring,realizing the integration goal of "registration,integration and tracking identification",this paper combines two kinds of visible light and infrared modal image or video processing and analysis,and completed the more crucial multimodal video stream registration process.The main contents of this paper include the realization of algorithms for multimodal image registration and multimodal video stream registration and the verification results on UAS.The main innovations and main contributions of this paper include the following three points:(1)Have proposed and implemented "DCE + LSS" high-precision multi-modal image registration algorithm.Researched and analyzed the DCE evolution algorithm of extracting the target contour on the static image,and improved the termination condition of the evolution process to achieve the effect of automatically terminating the evolution.Using the contour of DCE evolution,a new LSS feature descriptor is used to describe the 20?4-dimensional vector of each contour point,and the LSS descriptor is transformed into logarithmic polar coordinates to represent the image to enhance the tolerance of local images' deformation.Finally,multi-modal image registration is completed.(2)Have proposed and implemented video foreground extraction algorithm combining time and space information and so on.Based on the multi-modal image registration algorithm "DCE + LSS",the inter-frame correlation of multi-modal video stream has deeply analyzed,and the target(marked as foreground in this paper)based on the target motion has extracted from the multi-modal video stream.This algorithm is marked as the Background Features-method(BFs-method)in this paper.This algorithm not only has improved the defect of foregrounds "holes" caused by the LBSP background modeling algorithm but also has solved the "ghosting" phenomenon to a certain extent in the process of target foreground extraction.(3)Have implemented the registration and optimization of multimodal video stream on UAV.Based on the foreground obtained by the BFs-method algorithm,multi-modal video stream registration was carried out.Firstly,the foreground contour points were described based on the pixel-level shape context.Then the chi-square detection and the Hungarian algorithm were introduced to obtain a temporary set of global pairs.In this paper,we improved the update strategy of FIFO and seted up a new scoring mechanism,and scored matching point pairs in the buffer and determined the point-to-point update in the buffer.At the same time,the homology change matrixes were optimized step by step by using the minimum error coverage.Finally the global optimum homography change matrix was obtained,and registration of the multimodal video stream was completed.