%使用| imageDatastore |获得一个目录中的所有图像文件名称列表。imageDir = fullfile (toolboxdir (“愿景”),“visiondata”,…“structureFromMotion”);imd = imageDatastore (imageDir);%显示图像。图蒙太奇(imds.Files,“大小”,3,2);%将图像转换为灰度。图像=细胞(1,元素个数(imds.Files));为i = 1:元素个数(imds.Files)我= readimage (imd,我);{我}= im2gray图像(i);结束标题(输入图像序列的);

data =加载(fullfile (imageDir“cameraParams.mat”));cameraParams = data.cameraParams;

%得到相机的内在参数intrinsic = cameraParams.Intrinsics;% Undistort第一形象。I = undistortImage(图片{1},intrinsic);%检测功能。“NumOctaves”有助于检测大规模增加%在高分辨率图像特征。使用一个ROI来消除虚假的%图像的边缘特征。边境= 50;投资回报率=[边境,边境,大小(我,2)- 2 *边境,边界大小(我,1)- 2 *);prevPoints = detectSURFFeatures (NumOctaves = 8,我投资回报率(ROI);%提取特征。使用“正直”特性提高了匹配,只要%摄像机运动涉及很少或没有平面旋转。prevFeatures = extractFeatures(我prevPoints直立= true);%创建一个空imageviewset对象来管理相关的数据%的观点。vSet = imageviewset;%添加第一个视图。与第一个视图相关联的摄像机%和原点,面向沿z轴。viewId = 1;vSet = addView (vSet viewId rigidtform3d,分= prevPoints);

为我= 2:元素个数(图片)% Undistort当前图像。I = undistortImage(图片{我},intrinsic);%检测、提取和匹配特性。currPoints = detectSURFFeatures (NumOctaves = 8,我投资回报率(ROI);currFeatures = extractFeatures(我currPoints直立= true);indexPairs = matchFeatures (prevFeatures currFeatures,…独特MaxRatio = 0.7, = true);%选择匹配点。matchedPoints1 = prevPoints (indexPairs (: 1));matchedPoints2 = currPoints (indexPairs (:, 2));%估计当前视图的相机构成相对于前面的视图。%的计算规模,这意味着之间的距离%的相机在前面的视图和当前视图设置为1。%这将被捆绑调整修正。[relPose, inlierIdx] = helperEstimateRelativePose (…matchedPoints1、matchedPoints2 intrinsic);%得到包含前面的摄像机构成的表。.AbsolutePose prevPose =姿势(vSet张);%计算当前相机构成全球坐标系统%相对于第一个视图。currPose = rigidtform3d (prevPose.A * relPose.A);%设置当前视图添加到视图。vSet = addView (vSet我currPose点= currPoints);%之前存储之间的匹配点和当前视图。vSet = addConnection (vSet,张,我、relPose匹配= indexPairs (inlierIdx:));%找到跟踪所有观点。跟踪= findTracks (vSet);%获得所有视图包含相机带来的表。坎波斯=姿势(vSet);%满足初始点位置的3 d世界。xyzPoints = triangulateMultiview(追踪,坎波斯,intrinsic);%改进3 d世界分和相机的姿势。[xyzPoints,坎波斯,reprojectionErrors] = bundleAdjustment (xyzPoints,…痕迹,坎波斯,intrinsic FixedViewId = 1,…PointsUndistorted = true);%存储精制相机姿势。vSet = updateView (vSet campos);prevFeatures = currFeatures;prevPoints = currPoints;结束

%显示相机的姿势。坎波斯=姿势(vSet);图;plotCamera(坎波斯,大小= 0.2);持有在%排除嘈杂的3 d点。goodIdx = (reprojectionErrors < 5);xyzPoints = xyzPoints (goodIdx:);%显示3 d点。pcshow (xyzPoints VerticalAxis =“y”VerticalAxisDir =“下来”MarkerSize = 45);网格在持有从%指定查看卷。loc1 = camPoses.AbsolutePose (1) .Translation;xlim ([loc1 (1) 5, loc1 (1) + 4]);ylim ([loc1 (2) 5, loc1 (2) + 4]);zlim ([loc1 (3) 1, loc1 (3) + 20));camorbit (0, -30);标题(“精制相机姿势”);

%阅读和undistort第一形象I = undistortImage(图片{1},intrinsic);%检测角落的形象。prevPoints = detectMinEigenFeatures(我MinQuality = 0.001);%创建重点跟踪对象跟踪点在视图。追踪=愿景。PointTracker (MaxBidirectionalError = 1, NumPyramidLevels = 6);%初始化跟踪点。prevPoints = prevPoints.Location;初始化(prevPoints追踪,我);%存储在视图中密集点集。vSet = updateConnection (vSet, 1, 2,匹配= 0 (0,2));vSet = updateView (vSet 1分= prevPoints);%跟踪点在所有视图。为我= 2:元素个数(图片)%阅读和undistort当前图像。I = undistortImage(图片{我},intrinsic);%跟踪点。[currPoints, validIdx] =步骤(追踪,我);%清楚之间的匹配点。如果我<元素个数(图像)vSet = updateConnection (vSet,我,我+ 1,匹配= 0 (0,2));结束vSet = updateView (vSet,分= currPoints);%点匹配存储在视图集。匹配= repmat((1:尺寸(prevPoints, 1))”, [1, 2]);匹配=匹配(validIdx:);vSet = updateConnection (vSet,张,我匹配=匹配);结束%找到跟踪所有观点。跟踪= findTracks (vSet);%找到跟踪所有观点。坎波斯=姿势(vSet);%满足初始点位置的3 d世界。xyzPoints = triangulateMultiview(追踪,坎波斯,…intrinsic);%改进3 d世界分和相机的姿势。[xyzPoints,坎波斯,reprojectionErrors] = bundleAdjustment (…xyzPoints,追踪,坎波斯,intrinsic FixedViewId = 1,…PointsUndistorted = true);

%显示精制相机的姿势。图;plotCamera(坎波斯,大小= 0.2);持有在%点排除嘈杂的3 d世界。goodIdx = (reprojectionErrors < 5);%显示点密集的3 d世界。pcshow (xyzPoints (goodIdx:), VerticalAxis =“y”VerticalAxisDir =“下来”MarkerSize = 45);网格在持有从%指定查看卷。loc1 = camPoses.AbsolutePose (1) .Translation;xlim ([loc1 (1) 5, loc1 (1) + 4]);ylim ([loc1 (2) 5, loc1 (2) + 4]);zlim ([loc1 (3) 1, loc1 (3) + 20));camorbit (0, -30);标题(“密集的重建”);