Application Of Modified FAST Method In Video Mosaicing

Video mosaicing is a technology that uses a series of algorithms to collect video image features from overlap of images and uses these features to stitch video image as a wide scene image. Video mosaicing is a hot topic in field of image processing and computer vision. It made outstanding contributions in the field of car video, remote sensing, monitoring and resource exploration. Video mosaicing is based on image mosaicing. On the one hand, pros and cons of this technology may immediately determine the quality of image mosaicing, include image matching and fusion. On the other hand, video mosaicing is different from image mosaicing, real time performance is one of the main issues. And ghosting phenomenon is generated easily from the stitched images. For this problem, a high real time, high stability algorithm named as modified FAST method is provided to decrease ghosting phenomenon. Several main works of this paper show as follow.1) The paper used Visual Studio, OpenCV, DirectShow and camera to set up system environment. The step of image mosaicing and the system flow chart is designed before the start of the paper’s study. DirectShow is used to get datas from dual cameras. Advantages and disadvantages of the existing feature points extraction algorithm are compared, then a most suitable feature points extraction algorithm is selected.2) As to poor noise immunity of FAST algorithm, enhanced threshold method and local noise suppression method are used to decrease noises. In order to avoid manual inputting FAST algorithm threshold, a adaptive threshold method is proposed through calculating image’s contrast. For the purpose of eliminated feature points accumulation phenomena, local maximum suppression method is available. In order to obtain accurate positions of feature points and eliminate false feature points, the curve fitting method is adopted in the paper. Because of FAST algorithm having strong response to scene’s border in the image, The paper use Hassian matrix to decrease this responses.3) Center of gravity method is used to establish the direction of FAST feature points, and Brief method is used to describe FAST feature points and produce descriptor. Hamming method is used to calculate distance of descriptors in the different pictures, and find the optimal matching points pair. The matching points pair is filtered by RANSAC method then it is used to calculate the transformation matrix. Finally, a improved boundary fusion method is adopted to fuse video.4) Matching and fusion experiments in different algorithms is used to test real time performance and stability of modified FAST method. Matching and fusion experiments in different illuminations is used to test illumination adaptability of modified FAST method. Matching and fusion experiments in different strength of noise is used to test noise immunity performance of modified FAST method.