| Image stitching technology can effectively stitch together two images with overlapping areas.For the two images,the first thing need to do is to find the key points in the two images.This paper used the SIFT algorithm with good stability to complete this process,and proposed an efficient key point extraction algorithm.Based on the characteristics of FPGA,the phases of establishing the Gaussian pyramid,establishing the difference of Gaussian pyramid,key point detection and key point description were implemented and optimized.And in order to reduce the resource usage and increase the computing speed,the way that the key point extraction algorithm processes the image was optimized.The key point matching and the calculation of the transformation matrix are required before image fusion.According to the experiments,this paper used the nearest neighbor matching algorithm to complete the key point matching,and proposed various optimization ways.In order to obtain the better result and avoid the influence of mismatch,the random sample consensus algorithm was used to calculate the transformation matrix.During the image fusion,the bilinear interpolation algorithm and Alpha fusion algorithm were used to complete this process.Based on the above modules,this paper proposed the image stitching algorithm based on dynamic region.The algorithm can make full use of the inter-frame relationship and dynamically evaluate the overlapping part of the two images,thereby reducing the amount of calculation and increasing the speed of image stitching.In addition,in order to improve the speed of key point matching,the improved key point matching algorithm was proposed.Finally,after implementing the above modules and corresponding peripheral circuits,this paper implemented the full HD camera real-time stitching system.According to the experiments and comparisons,it was found that the indicators of the system had met the requirements and the processing performance was excellent. |
PDF Full Text Request