Research On Frame Infrared Image Stitching Technology

Stitching technology of infrared image is usually applied to target search and tracking,infrared aerial remote sensing,online monitoring and security alert.In the practical application,in addition to pursue higher stitching accuracy,we should also improve the stitching speed as much as possible in order to meet the demand for real-time infrared imaging system.On the basis of study each key step of stitching technology of frame infrared image,an image stitching algorithm which is suitable for aerial frame infrared imaging system is proposed.The main contents of this paper is:1)Focus on reducing the number of redundant feature points,holding stable feature points and reducing the complexity of the descriptor,a modified SURF-BRISK feature extraction is proposed by applying the Canny edge and BRISK descriptor to SURF feature extraction.Compared with the SURF algorithm,the average number of feature points extracted by the improved edge SURF-BRISK algorithm is reduced by 49.3%.And reduction of feature points has no effect on the success rate of registration.2)Because the infrared image has fewer feature points,we improve the building and searching algorithm of hierarchical clustering tree algorithm.The improved hierarchical clustering tree algorithm can adapt to different infrared image with different number of feature points.Compared with the conventional hierarchical clustering tree algorithm,the correct matching rate of improved hierarchical clustering tree algorithm is increased by 5.3%,and the time consuming of improved hierarchical clustering tree algorithm is reduced by 16.1%3)In the aspect of image global registration,a simplified bundle adjustment scheme is adopted based on the characteristics of motion mode of infrared imaging system.In each scan line all images are adjusted.And in the adjacent scan line,only the first image,middle image and tail image of the scan line are adjusted.Compared with the conventional bundle adjustment,the time consuming of simplified bundle adjustment scheme is averagely reduced by 12.3%.4)The blending quality of result image is improved by using the blending method based on the optimum matching point.The first scan line images are fused in the span direction,and the other scan line images are fused in span and flight direction.Under the condition that the image has complex background,the method of frame infrared image stitching proposed in this paper can meet the needs of image stitching in different scenes,and the stitching accuracy can meet the requirements of engineering application.Compared with the stitching method based on SURF,the stitching method of frame infrared image proposed in this paper improves the stitching accuracy,and the average time consuming is only 1/3 of that of SURF stitching method.