ln order to enhance real-time video stitching, this paper proposes the improved FAST method which is used to stitch video image. The process of this method is as fol ows. self-adapting threshold is calculated to detect and picks up enough feature points. those points are extracted through suppressing noise of single pixel, removing border and unstable feature points. For the purpose of increasing speed of match feature points, BRlEF method is used to decribe feature points. Here hamming method operates to match the feature points of two relevant images. The outline feature points is removed by RANSAC method. The match feature point pairs are used to compute homography matrix, then each frame of video is stitched. Background trembling coused by dynamic video stitc-hing method can be reduced through combining dynamic and static stitching method. The experiments show that improved-based FAST method is of higher positional accuracy and satisfied with real-time requirements. lt can be used to improve the speed of stitc-hing, decrease ghosted image created by static stitching of different deep view.%为提高视频拼接的实时性，提出一种改进型FAST快速视频拼接算法。该算法的过程是先计算图像的自适应阈值，根据该阈值对角点粗提取，接着对角点精提取，包括抑制单点噪声、剔除边缘角点、剔除不稳定的角点。为提高角点匹配速度，利用BRIEF算法对角点进行描述，通过Hamming算法匹配两幅图像的角点，接着使用RANSAC算法剔除外点。根据匹配点对计算变换矩阵，拼接每一帧视频图像。由于动态视频拼接产生的背景抖动现象可以通过动静态结合的拼接方法改善。实验表明，该拼接算法的速度显著提高，定位精度较高，能够满足实时性要求，而且能够改善静态拼接中景深不同而产生的鬼影现象。
展开▼
