| With the rapid development of mobile shooting platforms such as smart phones and drones,the proportion of video captured by mobile shooting devices is getting larger than before.Therefore,video image stabilization become a hot research issue in many fields.Due to the flexibility of the mobile shooting platform,the captured video is often accompanied by large-scale random jitter,perspective transformation,and foreground occlusion.However,the existing video image stabilization technology has problems such as inaccurate global motion vector estimation in the motion estimation part,and poor image repair effect in the motion compensation part.In order to solve these problems,this thesis puts forward the improvement of motion estimation and motion compensation repair methods based on the in-depth study of video stabilization repair technology.At the same time,a prototype system of video stabilization image restoration for mobile shooting platform was designed and implemented.The main research work of this thesis includes the following:(1)A motion estimation algorithm based on the optimization of reference frames and separation of foreground and background feature points is proposed to improve the accuracy of global motion vector estimation.First,the reference frame selection method is optimized based on the principle of adjacent frame priority,and then the feature points in the video frame are clustered and initially screened through the fusion of grid clustering and density clustering algorithms.Finally,the RANSAC algorithm is used to obtain the optimal homography matrix,which is further purified by combining perspective projection transformation and distance criterion to obtain accurate background feature point pairs for calculating global motion vectors.Experiments show that the algorithm can effectively improve the accuracy of global motion vectors and the effect of image stabilization.(2)A motion compensation repair algorithm based on time series network prediction and pyramid fusion is proposed to improve the situation of high image quality and unnatural splicing at the part of defect repair.First,adaptively determine whether the current frame needs to be filled and repaired.Then,for the frame to be repaired,the fusion model of CNN and GRU is sent to predict the complete image.Then,the image fusion was reconstructed by the Gaussian Laplacian pyramid model with fused weighted optimal seam.Finally,according to the adaptive cropping strategy,the effective area and size of the video are retained.Experimental results show that the algorithm effectively improves the repair effect of the defective part,and can effectively deal with the image stabilization repair task of the video taken by the mobile shooting platform.(3)A video stabilization prototype system for mobile shooting platform is designed and implemented.The system uses a functional modular design,including video input module,motion estimation module,trajectory smoothing module,video image repair module and storage module.Experiments show that the system has good ease of use and effectiveness,is suitable for a variety of devices including mobile shooting platforms,and has high application value and prospects. |
PDF Full Text Request