使用|imageDatastore|获取目录中所有镜像文件名的列表。imageDir = fullfile(toolboxdir(“愿景”)，“visiondata”，.．.“structureFromMotion”）;imds = imageDatastore(imageDir);显示图片。图蒙太奇(imd)。文件,“大小”， [3,2]);将图像转换为灰度。images = cell(1, numel(imds.Files));为i = 1:numel(imds. files) i = readimage(imds, i);images{i} = im2gray(i);结束标题(“输入图像序列”）;

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

获取相机的内在参数intrinsic = cameraparams . intrinsic;还原第一个图像。I = undistortion timage(图像{1}，intrinsic);检测特征。增加“NumOctaves”有助于检测大规模%特征在高分辨率图像。使用ROI来消除虚假%的特征围绕图像的边缘。Border = 50;roi = [border, border, size(I, 2)- 2*border, size(I, 1)- 2*border];prevPoints = detectsurffatures (I, numocaves =8, ROI= ROI);提取特征。使用“直立”功能可以改善匹配，只要相机运动涉及很少或没有平面内旋转。。prevFeatures = extractFeatures(I, prevPoints，直立=true);创建一个空的imageviewset对象来管理与每个对象相关的数据%的观点。vSet = imageviewset;添加第一个视图。将相机与第一个视图相关联%和原点，沿z轴方向。viewId = 1;vSet = addView(vSet, viewId, rigidtform3d, Points=prevPoints);

为I = 2:数字(图像)还原当前图像。I = undistortion timage (images{I}， intrinsic);检测，提取和匹配特征。currPoints = detectsurffatures (I, numocaves =8, ROI= ROI);currFeatures = extractFeatures(I, currPoints，直立=true);indexPairs = matchFeatures(prevFeatures, currFeatures，.．.独特MaxRatio = 0.7, = true);选择匹配的点。matchedPoints1 = prevPoints(indexPairs(:， 1));matchedPoints2 = currPoints(indexPairs(:， 2));估计当前视图相对于前一个视图的相机姿态。%姿势是按比例计算的，这意味着之间的距离%前一个视图和当前视图中的摄像机的数量设置为1。这将在bundle调整中得到修正。[relPose, inlierIdx] = helperEstimateRelativePose(.．.matchedPoints1, matchedPoints2, intrinsic);获取包含前一个相机姿势的表。prevPose = pose (vSet, i-1).AbsolutePose;在全局坐标系中计算当前相机姿态%相对于第一个视图。currPose = rigidtform3d(prevpose . * relpose .);将当前视图添加到视图集中。vSet = addView(vSet, i, currPose, Points=currPoints);存储前一个视图和当前视图之间的匹配点。vSet = addConnection(vSet, i-1, i, relPose, Matches=indexPairs(inlierIdx，:));在所有视图中查找点轨迹。tracks = findTracks(vSet);获取包含所有视图的相机姿势的表。camposed = pose (vSet);三角化三维世界点的初始位置。。xyzPoints = triangulateMultiview(轨迹，campose, intrinsic);精炼3-D世界点和相机姿势。[xyzPoints, camposed, reprojectionErrors] = bundleadadjustment (xyzPoints, camposed, reprojectionErrors).．.轨道，campose, intrinsic, FixedViewId=1，.．.PointsUndistorted = true);存储精致的相机姿势。vSet = updateView(vSet, camposed);prevFeatures = currFeatures;prevPoints = currPoints;结束

显示相机姿势。camposed = pose (vSet);图;plotCamera(坎波斯,大小= 0.2);持有在排除有噪声的3-D点。goodIdx = (reprojectionErrors < 5);xyzPoints = xyzPoints(goodIdx，:);显示3d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(“精致的相机姿势”）;

读取并还原第一个图像I = undistortion timage(图像{1}，intrinsic);检测第一个图像中的角落。prevPoints = detectMinEigenFeatures(I, MinQuality=0.001);创建点跟踪对象来跟踪视图中的点。追踪者=视觉。PointTracker(MaxBidirectionalError=1, NumPyramidLevels=6);初始化点跟踪器。prevPoints = prevPoints. location;初始化(跟踪器，prevPoints, I);在视图集中存储密集点。vSet = updateConnection(vSet, 1,2，匹配= 0 (0,2));vSet = updateView(vSet, 1, Points=prevPoints);跟踪所有视图中的点。为I = 2:数字(图像)读取并还原当前图像。I = undistortion timage (images{I}， intrinsic);跟踪分数。[currPoints, validIdx] = step(跟踪器，I);清除点之间的旧匹配。如果vSet = updateConnection(vSet, i, i+1，匹配= 0 (0,2));结束(vSet, i, Points=currPoints);将匹配的点存储在视图集中。matches = repmat((1:size(prevPoints, 1))'， [1,2]);matches = matches(validIdx，:);vSet = updateConnection(vSet, i-1, i, Matches= Matches);结束在所有视图中查找点轨迹。tracks = findTracks(vSet);在所有视图中查找点轨迹。camposed = pose (vSet);三角化三维世界点的初始位置。。xyzPoints = triangulateMultiview(跟踪，campositions，.．.intrinsic);精炼3-D世界点和相机姿势。[xyzPoints, camposed, reprojectionErrors] = bundleadadjustment (.．.xyzPoints, tracks, campose, intrinsic, FixedViewId=1，.．.PointsUndistorted = true);

%显示精致的相机姿势。图;plotCamera(坎波斯,大小= 0.2);持有在排除有噪声的3-D世界点。goodIdx = (reprojectionErrors < 5);显示密集的三维世界点。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”）;