计算机视觉12-6-基于特征的图像对准.pptVIP

* * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * * 2021/3/26 * RANSAC for Homography 2021/3/26 * RANSAC for Homography 2021/3/26 * Probabilistic model for verification 2021/3/26 * Finding the panoramas 2021/3/26 * Finding the panoramas 2021/3/26 * Finding the panoramas 2021/3/26 * Finding the panoramas 2021/3/26 * Parameterise each camera by rotation and focal length This gives pairwise homographies Homography for Rotation 2021/3/26 * Bundle Adjustment New images initialised with rotation, focal length of best matching image 2021/3/26 * Bundle Adjustment New images initialised with rotation, focal length of best matching image 2021/3/26 * Multi-band Blending Burt Adelson 1983 Blend frequency bands over range ? l 2021/3/26 * Results 2021/3/26 * 增强现实（Augmented Reality） Demo1 Demo2 Demo3 2021/3/26 * 增强现实 2021/3/26 * 位姿估计----增强现实 2021/3/26 * 第一次作业: 图像拼接(任选) 编写自己的图像拼接程序 将程序放在个人信箱的网络硬盘中 工具: Matlab or OpenCV 时间:课程结束之前 2021/3/26 * 第一次作业: 增强现实（任选） 编写自己的增强现实程序 将程序放在个人信箱的网络硬盘中 工具: ARTOOLKIT 时间:课程结束之前 * * * * * * * * * * * * * * * 2021/3/26 * 特征匹配 ? 生成特征点对: 对于图像中每个子块,在其它图像中找到和它相似的一系列候选匹配块 2021/3/26 * 特征空间内外点的排除 怎样判断哪些特征点对是可靠的? 启发式: 比较特征空间内最近邻的距离和次近邻的距离 对于那些不太明确的特征队而言,最近邻的距离和次近邻的距离的比值通常会比较大 David G. Lowe. Distinctive image features from scale-invariant keypoints.” IJCV 60 (2), pp. 91-110, 2004. Threshold of 0.8 provides good separation 2021/3/26 * 外点的处理 几何拟合方法: RANSAC Hough transform 2021/3/26 * RANSAC RANSAC loop: Randomly select a seed group of matches Compute transformation from seed group Find inliers to this transformation If the number of inliers is sufficiently large, re-compute least-squares estimate of transformation on all of the inliers Keep the transformation with the largest number of inliers 2021/3/26 * RANSAC example: Translation Putative matches 2021/3/26 * Select one match, count inliers RANSAC example: Translation 2021/3/26 * Select one match, count inliers RANSAC example: Translation 2021/3/26 * Select translation wit