Research Of Real-time Video Mosaic Based On GPU And CPU Collaborative Computing

Video stitching technology refers to a number of small angle video image mosaic camera equipment to collect a wide road with video.Video mosaic technology is used in military defense,security monitoring and other fields,is of great significance to reduce the risk and improve the safety.The video mosaic contains video image acquisition,image preprocessing steps,image registration,image fusion and panoramic video generating,video stitching system of real-time is higher,and the process of image registration and image fusion algorithm of the highest complexity,more time-consuming.To solve this problem,this paper focuses on the parallel implementation of image registration and image stitching process of fusion algorithm based on the related research of CPU(Central Processing Unit)and GPU(Graphics Processing Unit).First of all,for fast image registration in video mosaics,this paper proposes a method based on extended SURF(Speeded Up Robust Features)scale,the image registration method based on.SURF operator of image changes,illumination invariance.This algorithm uses the rapid detection of Fast-Hessian feature points;and then uses a local ring color extension information fusion to describe the characteristics of SURF descriptor extension;finally the automatic image feature matching algorithm of block similarity measure based on k-d tree.The experimental results show that this algorithm has strong stability,real-time.Finally,this paper designed and implemented a video mosaic system CPU and GPU cooperative computing.The system realizes the video image acquisition based on CPU and GPU collaboration,image preprocessing,image registration based on SURF,multi resolution image fusion process based on a series of video mosaic,and details of image registration and image fusion at the end of the implementation of the GPU algorithm.Experiments show that the system has good real-time video mosaic effect and high.