OpenCV 图像拼接-Stitcher类-Stitching detailed使用与参数介绍

关于OpenCV图像拼接的方法，如果不熟悉的话，可以先看看我整理的如下四篇博客：

• OpenCV常用图像拼接方法(一)：直接拼接(硬拼)

• OpenCV常用图像拼接方法(二)：基于模板匹配拼接

• OpenCV常用图像拼接方法(三)：基于特征匹配拼接

• OpenCV常用图像拼接方法(四)：基于Stitcher类拼接

本篇博客是Stitcher类的扩展介绍，通过例程stitching_detailed.cpp的使用和参数介绍，帮助大家了解Stitcher类拼接的具体步骤和方法，先看看其内部的流程结构图(如下)：

stitching_detailed.cpp目录如下，可以在自己安装的OpenCV目录下找到，笔者这里使用的OpenCV4.4版本 

stitching_detailed.cpp具体源码如下： 

// 05_Image_Stitch_Stitching_Detailed.cpp : 此文件包含 "main" 函数。程序执行将在此处开始并结束。  2 //  3 #include "pch.h"
#include <iostream>
#include <fstream>
#include <string>
#include "opencv2/opencv_modules.hpp"
#include <opencv2/core/utility.hpp>
#include "opencv2/imgcodecs.hpp"
#include "opencv2/highgui.hpp"
#include "opencv2/stitching/detail/autocalib.hpp"
#include "opencv2/stitching/detail/blenders.hpp"
#include "opencv2/stitching/detail/timelapsers.hpp"
#include "opencv2/stitching/detail/camera.hpp"
#include "opencv2/stitching/detail/exposure_compensate.hpp"
#include "opencv2/stitching/detail/matchers.hpp"
#include "opencv2/stitching/detail/motion_estimators.hpp"
#include "opencv2/stitching/detail/seam_finders.hpp"
#include "opencv2/stitching/detail/warpers.hpp"
#include "opencv2/stitching/warpers.hpp"
 
#ifdef HAVE_OPENCV_XFEATURES2D  23 #include "opencv2/xfeatures2d.hpp"
#include "opencv2/xfeatures2d/nonfree.hpp"
#endif
 
#define ENABLE_LOG 1
#define LOG(msg) std::cout << msg
#define LOGLN(msg) std::cout << msg << std::endl
 
using namespace std;  32 using namespace cv;  33 using namespace cv::detail;  34  
static void printUsage(char** argv)  36 {  37     cout <<
        "Rotation model images stitcher.\n\n"
        << argv[0] << " img1 img2 [...imgN] [flags]\n\n"
        "Flags:\n"
        " --preview\n"
        " Run stitching in the preview mode. Works faster than usual mode,\n"
        " but output image will have lower resolution.\n"
        " --try_cuda (yes|no)\n"
        " Try to use CUDA. The default value is 'no'. All default values\n"
        " are for CPU mode.\n"
        "\nMotion Estimation Flags:\n"
        " --work_megapix <float>\n"
        " Resolution for image registration step. The default is 0.6 Mpx.\n"
        " --features (surf|orb|sift|akaze)\n"
        " Type of features used for images matching.\n"
        " The default is surf if available, orb otherwise.\n"
        " --matcher (homography|affine)\n"
        " Matcher used for pairwise image matching.\n"
        " --estimator (homography|affine)\n"
        " Type of estimator used for transformation estimation.\n"
        " --match_conf <float>\n"
        " Confidence for feature matching step. The default is 0.65 for surf and 0.3 for orb.\n"
        " --conf_thresh <float>\n"
        " Threshold for two images are from the same panorama confidence.\n"
        " The default is 1.0.\n"
        " --ba (no|reproj|ray|affine)\n"
        " Bundle adjustment cost function. The default is ray.\n"
        " --ba_refine_mask (mask)\n"
        " Set refinement mask for bundle adjustment. It looks like 'x_xxx',\n"
        " where 'x' means refine respective parameter and '_' means don't\n"
        " refine one, and has the following format:\n"
        " <fx><skew><ppx><aspect><ppy>. The default mask is 'xxxxx'. If bundle\n"
        " adjustment doesn't support estimation of selected parameter then\n"
        " the respective flag is ignored.\n"
        " --wave_correct (no|horiz|vert)\n"
        " Perform wave effect correction. The default is 'horiz'.\n"
        " --save_graph <file_name>\n"
        " Save matches graph represented in DOT language to <file_name> file.\n"
        " Labels description: Nm is number of matches, Ni is number of inliers,\n"
        " C is confidence.\n"
        "\nCompositing Flags:\n"
        " --warp (affine|plane|cylindrical|spherical|fisheye|stereographic|compressedPlaneA2B1|compressedPlaneA1.5B1|compressedPlanePortraitA2B1|compressedPlanePortraitA1.5B1|paniniA2B1|paniniA1.5B1|paniniPortraitA2B1|paniniPortraitA1.5B1|mercator|transverseMercator)\n"
        " Warp surface type. The default is 'spherical'.\n"
        " --seam_megapix <float>\n"
        " Resolution for seam estimation step. The default is 0.1 Mpx.\n"
        " --seam (no|voronoi|gc_color|gc_colorgrad)\n"
        " Seam estimation method. The default is 'gc_color'.\n"
        " --compose_megapix <float>\n"
        " Resolution for compositing step. Use -1 for original resolution.\n"
        " The default is -1.\n"
        " --expos_comp (no|gain|gain_blocks|channels|channels_blocks)\n"
        " Exposure compensation method. The default is 'gain_blocks'.\n"
        " --expos_comp_nr_feeds <int>\n"
        " Number of exposure compensation feed. The default is 1.\n"
        " --expos_comp_nr_filtering <int>\n"
        " Number of filtering iterations of the exposure compensation gains.\n"
        " Only used when using a block exposure compensation method.\n"
        " The default is 2.\n"
        " --expos_comp_block_size <int>\n"
        " BLock size in pixels used by the exposure compensator.\n"
        " Only used when using a block exposure compensation method.\n"
        " The default is 32.\n"
        " --blend (no|feather|multiband)\n"
        " Blending method. The default is 'multiband'.\n"
        " --blend_strength <float>\n"
        " Blending strength from [0,100] range. The default is 5.\n"
        " --output <result_img>\n"
        " The default is 'result.jpg'.\n"
        " --timelapse (as_is|crop) \n"
        " Output warped images separately as frames of a time lapse movie, with 'fixed_' prepended to input file names.\n"
        " --rangewidth <int>\n"
        " uses range_width to limit number of images to match with.\n"; 109 } 110  
 
// Default command line args
vector<String> img_names; 114 bool preview = false; 115 bool try_cuda = false; 116 double work_megapix = 0.6; 117 double seam_megapix = 0.1; 118 double compose_megapix = -1; 119 float conf_thresh = 1.f; 120 #ifdef HAVE_OPENCV_XFEATURES2D 121 string features_type = "surf"; 122 float match_conf = 0.65f; 123 #else
string features_type = "orb"; 125 float match_conf = 0.3f; 126 #endif
string matcher_type = "homography"; 128 string estimator_type = "homography"; 129 string ba_cost_func = "ray"; 130 string ba_refine_mask = "xxxxx"; 131 bool do_wave_correct = true; 132 WaveCorrectKind wave_correct = detail::WAVE_CORRECT_HORIZ; 133 bool save_graph = false; 134 std::string save_graph_to; 135 string warp_type = "spherical"; 136 int expos_comp_type = ExposureCompensator::GAIN_BLOCKS; 137 int expos_comp_nr_feeds = 1; 138 int expos_comp_nr_filtering = 2; 139 int expos_comp_block_size = 32; 140 string seam_find_type = "gc_color"; 141 int blend_type = Blender::MULTI_BAND; 142 int timelapse_type = Timelapser::AS_IS; 143 float blend_strength = 5; 144 string result_name = "result.jpg"; 145 bool timelapse = false; 146 int range_width = -1; 147  
 
static int parseCmdArgs(int argc, char** argv) 150 { 151     if (argc == 1) 152  { 153  printUsage(argv); 154         return -1; 155  } 156     for (int i = 1; i < argc; ++i) 157  { 158         if (string(argv[i]) == "--help" || string(argv[i]) == "/?") 159  { 160  printUsage(argv); 161             return -1; 162  } 163         else if (string(argv[i]) == "--preview") 164  { 165             preview = true; 166  } 167         else if (string(argv[i]) == "--try_cuda") 168  { 169             if (string(argv[i + 1]) == "no") 170                 try_cuda = false; 171             else if (string(argv[i + 1]) == "yes") 172                 try_cuda = true; 173             else
 { 175                 cout << "Bad --try_cuda flag value\n"; 176                 return -1; 177  } 178             i++; 179  } 180         else if (string(argv[i]) == "--work_megapix") 181  { 182             work_megapix = atof(argv[i + 1]); 183             i++; 184  } 185         else if (string(argv[i]) == "--seam_megapix") 186  { 187             seam_megapix = atof(argv[i + 1]); 188             i++; 189  } 190         else if (string(argv[i]) == "--compose_megapix") 191  { 192             compose_megapix = atof(argv[i + 1]); 193             i++; 194  } 195         else if (string(argv[i]) == "--result") 196  { 197             result_name = argv[i + 1]; 198             i++; 199  } 200         else if (string(argv[i]) == "--features") 201  { 202             features_type = argv[i + 1]; 203             if (string(features_type) == "orb") 204                 match_conf = 0.3f; 205             i++; 206  } 207         else if (string(argv[i]) == "--matcher") 208  { 209             if (string(argv[i + 1]) == "homography" || string(argv[i + 1]) == "affine") 210                 matcher_type = argv[i + 1]; 211             else
 { 213                 cout << "Bad --matcher flag value\n"; 214                 return -1; 215  } 216             i++; 217  } 218         else if (string(argv[i]) == "--estimator") 219  { 220             if (string(argv[i + 1]) == "homography" || string(argv[i + 1]) == "affine") 221                 estimator_type = argv[i + 1]; 222             else
 { 224                 cout << "Bad --estimator flag value\n"; 225                 return -1; 226  } 227             i++; 228  } 229         else if (string(argv[i]) == "--match_conf") 230  { 231             match_conf = static_cast<float>(atof(argv[i + 1])); 232             i++; 233  } 234         else if (string(argv[i]) == "--conf_thresh") 235  { 236             conf_thresh = static_cast<float>(atof(argv[i + 1])); 237             i++; 238  } 239         else if (string(argv[i]) == "--ba") 240  { 241             ba_cost_func = argv[i + 1]; 242             i++; 243  } 244         else if (string(argv[i]) == "--ba_refine_mask") 245  { 246             ba_refine_mask = argv[i + 1]; 247             if (ba_refine_mask.size() != 5) 248  { 249                 cout << "Incorrect refinement mask length.\n"; 250                 return -1; 251  } 252             i++; 253  } 254         else if (string(argv[i]) == "--wave_correct") 255  { 256             if (string(argv[i + 1]) == "no") 257                 do_wave_correct = false; 258             else if (string(argv[i + 1]) == "horiz") 259  { 260                 do_wave_correct = true; 261                 wave_correct = detail::WAVE_CORRECT_HORIZ; 262  } 263             else if (string(argv[i + 1]) == "vert") 264  { 265                 do_wave_correct = true; 266                 wave_correct = detail::WAVE_CORRECT_VERT; 267  } 268             else
 { 270                 cout << "Bad --wave_correct flag value\n"; 271                 return -1; 272  } 273             i++; 274  } 275         else if (string(argv[i]) == "--save_graph") 276  { 277             save_graph = true; 278             save_graph_to = argv[i + 1]; 279             i++; 280  } 281         else if (string(argv[i]) == "--warp") 282  { 283             warp_type = string(argv[i + 1]); 284             i++; 285  } 286         else if (string(argv[i]) == "--expos_comp") 287  { 288             if (string(argv[i + 1]) == "no") 289                 expos_comp_type = ExposureCompensator::NO; 290             else if (string(argv[i + 1]) == "gain") 291                 expos_comp_type = ExposureCompensator::GAIN; 292             else if (string(argv[i + 1]) == "gain_blocks") 293                 expos_comp_type = ExposureCompensator::GAIN_BLOCKS; 294             else if (string(argv[i + 1]) == "channels") 295                 expos_comp_type = ExposureCompensator::CHANNELS; 296             else if (string(argv[i + 1]) == "channels_blocks") 297                 expos_comp_type = ExposureCompensator::CHANNELS_BLOCKS; 298             else
 { 300                 cout << "Bad exposure compensation method\n"; 301                 return -1; 302  } 303             i++; 304  } 305         else if (string(argv[i]) == "--expos_comp_nr_feeds") 306  { 307             expos_comp_nr_feeds = atoi(argv[i + 1]); 308             i++; 309  } 310         else if (string(argv[i]) == "--expos_comp_nr_filtering") 311  { 312             expos_comp_nr_filtering = atoi(argv[i + 1]); 313             i++; 314  } 315         else if (string(argv[i]) == "--expos_comp_block_size") 316  { 317             expos_comp_block_size = atoi(argv[i + 1]); 318             i++; 319  } 320         else if (string(argv[i]) == "--seam") 321  { 322             if (string(argv[i + 1]) == "no" ||
                string(argv[i + 1]) == "voronoi" ||
                string(argv[i + 1]) == "gc_color" ||
                string(argv[i + 1]) == "gc_colorgrad" ||
                string(argv[i + 1]) == "dp_color" ||
                string(argv[i + 1]) == "dp_colorgrad") 328                 seam_find_type = argv[i + 1]; 329             else
 { 331                 cout << "Bad seam finding method\n"; 332                 return -1; 333  } 334             i++; 335  } 336         else if (string(argv[i]) == "--blend") 337  { 338             if (string(argv[i + 1]) == "no") 339                 blend_type = Blender::NO; 340             else if (string(argv[i + 1]) == "feather") 341                 blend_type = Blender::FEATHER; 342             else if (string(argv[i + 1]) == "multiband") 343                 blend_type = Blender::MULTI_BAND; 344             else
 { 346                 cout << "Bad blending method\n"; 347                 return -1; 348  } 349             i++; 350  } 351         else if (string(argv[i]) == "--timelapse") 352  { 353             timelapse = true; 354  
            if (string(argv[i + 1]) == "as_is") 356                 timelapse_type = Timelapser::AS_IS; 357             else if (string(argv[i + 1]) == "crop") 358                 timelapse_type = Timelapser::CROP; 359             else
 { 361                 cout << "Bad timelapse method\n"; 362                 return -1; 363  } 364             i++; 365  } 366         else if (string(argv[i]) == "--rangewidth") 367  { 368             range_width = atoi(argv[i + 1]); 369             i++; 370  } 371         else if (string(argv[i]) == "--blend_strength") 372  { 373             blend_strength = static_cast<float>(atof(argv[i + 1])); 374             i++; 375  } 376         else if (string(argv[i]) == "--output") 377  { 378             result_name = argv[i + 1]; 379             i++; 380  } 381         else
 img_names.push_back(argv[i]); 383  } 384     if (preview) 385  { 386         compose_megapix = 0.6; 387  } 388     return 0; 389 } 390  
 
int main(int argc, char* argv[]) 393 { 394 #if ENABLE_LOG
    int64 app_start_time = getTickCount(); 396 #endif
 
#if 0
    cv::setBreakOnError(true); 400 #endif
 
    int retval = parseCmdArgs(argc, argv); 403     if (retval) 404         return retval; 405  
    // Check if have enough images
    int num_images = static_cast<int>(img_names.size()); 408     if (num_images < 2) 409  { 410         LOGLN("Need more images"); 411         return -1; 412  } 413  
    double work_scale = 1, seam_scale = 1, compose_scale = 1; 415     bool is_work_scale_set = false, is_seam_scale_set = false, is_compose_scale_set = false; 416  
    LOGLN("Finding features..."); 418 #if ENABLE_LOG
    int64 t = getTickCount(); 420 #endif
 
    Ptr<Feature2D> finder; 423     if (features_type == "orb") 424  { 425         finder = ORB::create(); 426  } 427     else if (features_type == "akaze") 428  { 429         finder = AKAZE::create(); 430  } 431 #ifdef HAVE_OPENCV_XFEATURES2D 432     else if (features_type == "surf") 433  { 434         finder = xfeatures2d::SURF::create(); 435  } 436 #endif
    else if (features_type == "sift") 438  { 439         finder = SIFT::create(); 440  } 441     else
 { 443         cout << "Unknown 2D features type: '" << features_type << "'.\n"; 444         return -1; 445  } 446  
 Mat full_img, img; 448     vector<ImageFeatures> features(num_images); 449     vector<Mat> images(num_images); 450     vector<Size> full_img_sizes(num_images); 451     double seam_work_aspect = 1; 452  
    for (int i = 0; i < num_images; ++i) 454  { 455         full_img = imread(samples::findFile(img_names[i])); 456         full_img_sizes[i] = full_img.size(); 457  
        if (full_img.empty()) 459  { 460             LOGLN("Can't open image " << img_names[i]); 461             return -1; 462  } 463         if (work_megapix < 0) 464  { 465             img = full_img; 466             work_scale = 1; 467             is_work_scale_set = true; 468  } 469         else
 { 471             if (!is_work_scale_set) 472  { 473                 work_scale = min(1.0, sqrt(work_megapix * 1e6 / full_img.size().area())); 474                 is_work_scale_set = true; 475  } 476  resize(full_img, img, Size(), work_scale, work_scale, INTER_LINEAR_EXACT); 477  } 478         if (!is_seam_scale_set) 479  { 480             seam_scale = min(1.0, sqrt(seam_megapix * 1e6 / full_img.size().area())); 481             seam_work_aspect = seam_scale / work_scale; 482             is_seam_scale_set = true; 483  } 484  
 computeImageFeatures(finder, img, features[i]); 486         features[i].img_idx = i; 487         LOGLN("Features in image #" << i + 1 << ": " << features[i].keypoints.size()); 488  
 resize(full_img, img, Size(), seam_scale, seam_scale, INTER_LINEAR_EXACT); 490         images[i] = img.clone(); 491  } 492  
 full_img.release(); 494  img.release(); 495  
    LOGLN("Finding features, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); 497  
    LOG("Pairwise matching"); 499 #if ENABLE_LOG
    t = getTickCount(); 501 #endif
    vector<MatchesInfo> pairwise_matches; 503     Ptr<FeaturesMatcher> matcher; 504     if (matcher_type == "affine") 505         matcher = makePtr<AffineBestOf2NearestMatcher>(false, try_cuda, match_conf); 506     else if (range_width == -1) 507         matcher = makePtr<BestOf2NearestMatcher>(try_cuda, match_conf); 508     else
        matcher = makePtr<BestOf2NearestRangeMatcher>(range_width, try_cuda, match_conf); 510  
    (*matcher)(features, pairwise_matches); 512     matcher->collectGarbage(); 513  
    LOGLN("Pairwise matching, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); 515  
    // Check if we should save matches graph
    if (save_graph) 518  { 519         LOGLN("Saving matches graph..."); 520  ofstream f(save_graph_to.c_str()); 521         f << matchesGraphAsString(img_names, pairwise_matches, conf_thresh); 522  } 523  
    // Leave only images we are sure are from the same panorama
    vector<int> indices = leaveBiggestComponent(features, pairwise_matches, conf_thresh); 526     vector<Mat> img_subset; 527     vector<String> img_names_subset; 528     vector<Size> full_img_sizes_subset; 529     for (size_t i = 0; i < indices.size(); ++i) 530  { 531  img_names_subset.push_back(img_names[indices[i]]); 532  img_subset.push_back(images[indices[i]]); 533  full_img_sizes_subset.push_back(full_img_sizes[indices[i]]); 534  } 535  
    images = img_subset; 537     img_names = img_names_subset; 538     full_img_sizes = full_img_sizes_subset; 539  
    // Check if we still have enough images
    num_images = static_cast<int>(img_names.size()); 542     if (num_images < 2) 543  { 544         LOGLN("Need more images"); 545         return -1; 546  } 547  
    Ptr<Estimator> estimator; 549     if (estimator_type == "affine") 550         estimator = makePtr<AffineBasedEstimator>(); 551     else
        estimator = makePtr<HomographyBasedEstimator>(); 553  
    vector<CameraParams> cameras; 555     if (!(*estimator)(features, pairwise_matches, cameras)) 556  { 557         cout << "Homography estimation failed.\n"; 558         return -1; 559  } 560  
    for (size_t i = 0; i < cameras.size(); ++i) 562  { 563  Mat R; 564  cameras[i].R.convertTo(R, CV_32F); 565         cameras[i].R = R; 566         LOGLN("Initial camera intrinsics #" << indices[i] + 1 << ":\nK:\n" << cameras[i].K() << "\nR:\n" << cameras[i].R); 567  } 568  
    Ptr<detail::BundleAdjusterBase> adjuster; 570     if (ba_cost_func == "reproj") adjuster = makePtr<detail::BundleAdjusterReproj>(); 571     else if (ba_cost_func == "ray") adjuster = makePtr<detail::BundleAdjusterRay>(); 572     else if (ba_cost_func == "affine") adjuster = makePtr<detail::BundleAdjusterAffinePartial>(); 573     else if (ba_cost_func == "no") adjuster = makePtr<NoBundleAdjuster>(); 574     else
 { 576         cout << "Unknown bundle adjustment cost function: '" << ba_cost_func << "'.\n"; 577         return -1; 578  } 579     adjuster->setConfThresh(conf_thresh); 580     Mat_<uchar> refine_mask = Mat::zeros(3, 3, CV_8U); 581     if (ba_refine_mask[0] == 'x') refine_mask(0, 0) = 1; 582     if (ba_refine_mask[1] == 'x') refine_mask(0, 1) = 1; 583     if (ba_refine_mask[2] == 'x') refine_mask(0, 2) = 1; 584     if (ba_refine_mask[3] == 'x') refine_mask(1, 1) = 1; 585     if (ba_refine_mask[4] == 'x') refine_mask(1, 2) = 1; 586     adjuster->setRefinementMask(refine_mask); 587     if (!(*adjuster)(features, pairwise_matches, cameras)) 588  { 589         cout << "Camera parameters adjusting failed.\n"; 590         return -1; 591  } 592  
    // Find median focal length
 
    vector<double> focals; 596     for (size_t i = 0; i < cameras.size(); ++i) 597  { 598         LOGLN("Camera #" << indices[i] + 1 << ":\nK:\n" << cameras[i].K() << "\nR:\n" << cameras[i].R); 599  focals.push_back(cameras[i].focal); 600  } 601  
 sort(focals.begin(), focals.end()); 603     float warped_image_scale; 604     if (focals.size() % 2 == 1) 605         warped_image_scale = static_cast<float>(focals[focals.size() / 2]); 606     else
        warped_image_scale = static_cast<float>(focals[focals.size() / 2 - 1] + focals[focals.size() / 2]) * 0.5f; 608  
    if (do_wave_correct) 610  { 611         vector<Mat> rmats; 612         for (size_t i = 0; i < cameras.size(); ++i) 613  rmats.push_back(cameras[i].R.clone()); 614  waveCorrect(rmats, wave_correct); 615         for (size_t i = 0; i < cameras.size(); ++i) 616             cameras[i].R = rmats[i]; 617  } 618  
    LOGLN("Warping images (auxiliary)... "); 620 #if ENABLE_LOG
    t = getTickCount(); 622 #endif
 
    vector<Point> corners(num_images); 625     vector<UMat> masks_warped(num_images); 626     vector<UMat> images_warped(num_images); 627     vector<Size> sizes(num_images); 628     vector<UMat> masks(num_images); 629  
    // Prepare images masks
    for (int i = 0; i < num_images; ++i) 632  { 633  masks[i].create(images[i].size(), CV_8U); 634         masks[i].setTo(Scalar::all(255)); 635  } 636  
    // Warp images and their masks
 
    Ptr<WarperCreator> warper_creator; 640 #ifdef HAVE_OPENCV_CUDAWARPING 641     if (try_cuda && cuda::getCudaEnabledDeviceCount() > 0) 642  { 643         if (warp_type == "plane") 644             warper_creator = makePtr<cv::PlaneWarperGpu>(); 645         else if (warp_type == "cylindrical") 646             warper_creator = makePtr<cv::CylindricalWarperGpu>(); 647         else if (warp_type == "spherical") 648             warper_creator = makePtr<cv::SphericalWarperGpu>(); 649  } 650     else
#endif
 { 653         if (warp_type == "plane") 654             warper_creator = makePtr<cv::PlaneWarper>(); 655         else if (warp_type == "affine") 656             warper_creator = makePtr<cv::AffineWarper>(); 657         else if (warp_type == "cylindrical") 658             warper_creator = makePtr<cv::CylindricalWarper>(); 659         else if (warp_type == "spherical") 660             warper_creator = makePtr<cv::SphericalWarper>(); 661         else if (warp_type == "fisheye") 662             warper_creator = makePtr<cv::FisheyeWarper>(); 663         else if (warp_type == "stereographic") 664             warper_creator = makePtr<cv::StereographicWarper>(); 665         else if (warp_type == "compressedPlaneA2B1") 666             warper_creator = makePtr<cv::CompressedRectilinearWarper>(2.0f, 1.0f); 667         else if (warp_type == "compressedPlaneA1.5B1") 668             warper_creator = makePtr<cv::CompressedRectilinearWarper>(1.5f, 1.0f); 669         else if (warp_type == "compressedPlanePortraitA2B1") 670             warper_creator = makePtr<cv::CompressedRectilinearPortraitWarper>(2.0f, 1.0f); 671         else if (warp_type == "compressedPlanePortraitA1.5B1") 672             warper_creator = makePtr<cv::CompressedRectilinearPortraitWarper>(1.5f, 1.0f); 673         else if (warp_type == "paniniA2B1") 674             warper_creator = makePtr<cv::PaniniWarper>(2.0f, 1.0f); 675         else if (warp_type == "paniniA1.5B1") 676             warper_creator = makePtr<cv::PaniniWarper>(1.5f, 1.0f); 677         else if (warp_type == "paniniPortraitA2B1") 678             warper_creator = makePtr<cv::PaniniPortraitWarper>(2.0f, 1.0f); 679         else if (warp_type == "paniniPortraitA1.5B1") 680             warper_creator = makePtr<cv::PaniniPortraitWarper>(1.5f, 1.0f); 681         else if (warp_type == "mercator") 682             warper_creator = makePtr<cv::MercatorWarper>(); 683         else if (warp_type == "transverseMercator") 684             warper_creator = makePtr<cv::TransverseMercatorWarper>(); 685  } 686  
    if (!warper_creator) 688  { 689         cout << "Can't create the following warper '" << warp_type << "'\n"; 690         return 1; 691  } 692  
    Ptr<RotationWarper> warper = warper_creator->create(static_cast<float>(warped_image_scale * seam_work_aspect)); 694  
    for (int i = 0; i < num_images; ++i) 696  { 697         Mat_<float> K; 698  cameras[i].K().convertTo(K, CV_32F); 699         float swa = (float)seam_work_aspect; 700         K(0, 0) *= swa; K(0, 2) *= swa; 701         K(1, 1) *= swa; K(1, 2) *= swa; 702  
        corners[i] = warper->warp(images[i], K, cameras[i].R, INTER_LINEAR, BORDER_REFLECT, images_warped[i]); 704         sizes[i] = images_warped[i].size(); 705  
        warper->warp(masks[i], K, cameras[i].R, INTER_NEAREST, BORDER_CONSTANT, masks_warped[i]); 707  } 708  
    vector<UMat> images_warped_f(num_images); 710     for (int i = 0; i < num_images; ++i) 711  images_warped[i].convertTo(images_warped_f[i], CV_32F); 712  
    LOGLN("Warping images, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); 714  
    LOGLN("Compensating exposure..."); 716 #if ENABLE_LOG
    t = getTickCount(); 718 #endif
 
    Ptr<ExposureCompensator> compensator = ExposureCompensator::createDefault(expos_comp_type); 721     if (dynamic_cast<GainCompensator*>(compensator.get())) 722  { 723         GainCompensator* gcompensator = dynamic_cast<GainCompensator*>(compensator.get()); 724         gcompensator->setNrFeeds(expos_comp_nr_feeds); 725  } 726  
    if (dynamic_cast<ChannelsCompensator*>(compensator.get())) 728  { 729         ChannelsCompensator* ccompensator = dynamic_cast<ChannelsCompensator*>(compensator.get()); 730         ccompensator->setNrFeeds(expos_comp_nr_feeds); 731  } 732  
    if (dynamic_cast<BlocksCompensator*>(compensator.get())) 734  { 735         BlocksCompensator* bcompensator = dynamic_cast<BlocksCompensator*>(compensator.get()); 736         bcompensator->setNrFeeds(expos_comp_nr_feeds); 737         bcompensator->setNrGainsFilteringIterations(expos_comp_nr_filtering); 738         bcompensator->setBlockSize(expos_comp_block_size, expos_comp_block_size); 739  } 740  
    compensator->feed(corners, images_warped, masks_warped); 742  
    LOGLN("Compensating exposure, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); 744  
    LOGLN("Finding seams..."); 746 #if ENABLE_LOG
    t = getTickCount(); 748 #endif
 
    Ptr<SeamFinder> seam_finder; 751     if (seam_find_type == "no") 752         seam_finder = makePtr<detail::NoSeamFinder>(); 753     else if (seam_find_type == "voronoi") 754         seam_finder = makePtr<detail::VoronoiSeamFinder>(); 755     else if (seam_find_type == "gc_color") 756  { 757 #ifdef HAVE_OPENCV_CUDALEGACY 758         if (try_cuda && cuda::getCudaEnabledDeviceCount() > 0) 759             seam_finder = makePtr<detail::GraphCutSeamFinderGpu>(GraphCutSeamFinderBase::COST_COLOR); 760         else
#endif
            seam_finder = makePtr<detail::GraphCutSeamFinder>(GraphCutSeamFinderBase::COST_COLOR); 763  } 764     else if (seam_find_type == "gc_colorgrad") 765  { 766 #ifdef HAVE_OPENCV_CUDALEGACY 767         if (try_cuda && cuda::getCudaEnabledDeviceCount() > 0) 768             seam_finder = makePtr<detail::GraphCutSeamFinderGpu>(GraphCutSeamFinderBase::COST_COLOR_GRAD); 769         else
#endif
            seam_finder = makePtr<detail::GraphCutSeamFinder>(GraphCutSeamFinderBase::COST_COLOR_GRAD); 772  } 773     else if (seam_find_type == "dp_color") 774         seam_finder = makePtr<detail::DpSeamFinder>(DpSeamFinder::COLOR); 775     else if (seam_find_type == "dp_colorgrad") 776         seam_finder = makePtr<detail::DpSeamFinder>(DpSeamFinder::COLOR_GRAD); 777     if (!seam_finder) 778  { 779         cout << "Can't create the following seam finder '" << seam_find_type << "'\n"; 780         return 1; 781  } 782  
    seam_finder->find(images_warped_f, corners, masks_warped); 784  
    LOGLN("Finding seams, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); 786  
    // Release unused memory
 images.clear(); 789  images_warped.clear(); 790  images_warped_f.clear(); 791  masks.clear(); 792  
    LOGLN("Compositing..."); 794 #if ENABLE_LOG
    t = getTickCount(); 796 #endif
 
 Mat img_warped, img_warped_s; 799  Mat dilated_mask, seam_mask, mask, mask_warped; 800     Ptr<Blender> blender; 801     Ptr<Timelapser> timelapser; 802     //double compose_seam_aspect = 1;
    double compose_work_aspect = 1; 804  
    for (int img_idx = 0; img_idx < num_images; ++img_idx) 806  { 807         LOGLN("Compositing image #" << indices[img_idx] + 1); 808  
        // Read image and resize it if necessary
        full_img = imread(samples::findFile(img_names[img_idx])); 811         if (!is_compose_scale_set) 812  { 813             if (compose_megapix > 0) 814                 compose_scale = min(1.0, sqrt(compose_megapix * 1e6 / full_img.size().area())); 815             is_compose_scale_set = true; 816  
            // Compute relative scales 818             //compose_seam_aspect = compose_scale / seam_scale;
            compose_work_aspect = compose_scale / work_scale; 820  
            // Update warped image scale
            warped_image_scale *= static_cast<float>(compose_work_aspect); 823             warper = warper_creator->create(warped_image_scale); 824  
            // Update corners and sizes
            for (int i = 0; i < num_images; ++i) 827  { 828                 // Update intrinsics
                cameras[i].focal *= compose_work_aspect; 830                 cameras[i].ppx *= compose_work_aspect; 831                 cameras[i].ppy *= compose_work_aspect; 832  
                // Update corner and size
                Size sz = full_img_sizes[i]; 835                 if (std::abs(compose_scale - 1) > 1e-1) 836  { 837                     sz.width = cvRound(full_img_sizes[i].width * compose_scale); 838                     sz.height = cvRound(full_img_sizes[i].height * compose_scale); 839  } 840  
 Mat K; 842  cameras[i].K().convertTo(K, CV_32F); 843                 Rect roi = warper->warpRoi(sz, K, cameras[i].R); 844                 corners[i] = roi.tl(); 845                 sizes[i] = roi.size(); 846  } 847  } 848         if (abs(compose_scale - 1) > 1e-1) 849  resize(full_img, img, Size(), compose_scale, compose_scale, INTER_LINEAR_EXACT); 850         else
            img = full_img; 852  full_img.release(); 853         Size img_size = img.size(); 854  
 Mat K; 856  cameras[img_idx].K().convertTo(K, CV_32F); 857  
        // Warp the current image
        warper->warp(img, K, cameras[img_idx].R, INTER_LINEAR, BORDER_REFLECT, img_warped); 860  
        // Warp the current image mask
 mask.create(img_size, CV_8U); 863         mask.setTo(Scalar::all(255)); 864         warper->warp(mask, K, cameras[img_idx].R, INTER_NEAREST, BORDER_CONSTANT, mask_warped); 865  
        // Compensate exposure
        compensator->apply(img_idx, corners[img_idx], img_warped, mask_warped); 868  
 img_warped.convertTo(img_warped_s, CV_16S); 870  img_warped.release(); 871  img.release(); 872  mask.release(); 873  
 dilate(masks_warped[img_idx], dilated_mask, Mat()); 875         resize(dilated_mask, seam_mask, mask_warped.size(), 0, 0, INTER_LINEAR_EXACT); 876         mask_warped = seam_mask & mask_warped; 877  
        if (!blender && !timelapse) 879  { 880             blender = Blender::createDefault(blend_type, try_cuda); 881             Size dst_sz = resultRoi(corners, sizes).size(); 882             float blend_width = sqrt(static_cast<float>(dst_sz.area())) * blend_strength / 100.f; 883             if (blend_width < 1.f) 884                 blender = Blender::createDefault(Blender::NO, try_cuda); 885             else if (blend_type == Blender::MULTI_BAND) 886  { 887                 MultiBandBlender* mb = dynamic_cast<MultiBandBlender*>(blender.get()); 888                 mb->setNumBands(static_cast<int>(ceil(log(blend_width) / log(2.)) - 1.)); 889                 LOGLN("Multi-band blender, number of bands: " << mb->numBands()); 890  } 891             else if (blend_type == Blender::FEATHER) 892  { 893                 FeatherBlender* fb = dynamic_cast<FeatherBlender*>(blender.get()); 894                 fb->setSharpness(1.f / blend_width); 895                 LOGLN("Feather blender, sharpness: " << fb->sharpness()); 896  } 897             blender->prepare(corners, sizes); 898  } 899         else if (!timelapser && timelapse) 900  { 901             timelapser = Timelapser::createDefault(timelapse_type); 902             timelapser->initialize(corners, sizes); 903  } 904  
        // Blend the current image
        if (timelapse) 907  { 908             timelapser->process(img_warped_s, Mat::ones(img_warped_s.size(), CV_8UC1), corners[img_idx]); 909  String fixedFileName; 910             size_t pos_s = String(img_names[img_idx]).find_last_of("/\\"); 911             if (pos_s == String::npos) 912  { 913                 fixedFileName = "fixed_" + img_names[img_idx]; 914  } 915             else
 { 917                 fixedFileName = "fixed_" + String(img_names[img_idx]).substr(pos_s + 1, String(img_names[img_idx]).length() - pos_s); 918  } 919             imwrite(fixedFileName, timelapser->getDst()); 920  } 921         else
 { 923             blender->feed(img_warped_s, mask_warped, corners[img_idx]); 924  } 925  } 926  
    if (!timelapse) 928  { 929  Mat result, result_mask; 930         blender->blend(result, result_mask); 931  
        LOGLN("Compositing, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); 933  
 imwrite(result_name, result); 935  } 936  
    LOGLN("Finished, total time: " << ((getTickCount() - app_start_time) / getTickFrequency()) << " sec"); 938     return 0; 939 }

stitching_detail 程序运行流程

• 命令行调用程序，输入源图像以及程序的参数      
• 特征点检测，判断是使用 surf 还是 orb，默认是 surf
• 对图像的特征点进行匹配，使用最近邻和次近邻方法，将两个最优的匹配的置信度 保存下来
• 对图像进行排序以及将置信度高的图像保存到同一个集合中，删除置信度比较低的图像间的匹配，得到能正确匹配的图像序列。这样将置信度高于门限的所有匹配合并到一个集合中 
• 对所有图像进行相机参数粗略估计，然后求出旋转矩阵
• 使用光束平均法进一步精准的估计出旋转矩阵
• 波形校正，水平或者垂直
• 拼接      
• 融合，多频段融合，光照补偿

stitching_detail 程序接口介绍 

• img1 img2 img3 输入图像      
• --preview  以预览模式运行程序，比正常模式要快，但输出图像分辨率低，拼接的分辨 率 compose_megapix 设置为 0.6
• --try_gpu  (yes|no)  是否使用 CUDA加速，默认为 no，使用CPU模式
• /* 运动估计参数 */    
• --work_megapix <--work_megapix <float>> 图像匹配时的分辨率大小，默认为 0.6    
• --features (surf | orb | sift | akaze) 选择 surf 或者 orb 算法进行特征点匹配，默认为 surf  
• --matcher (homography | affine) 用于成对图像匹配的匹配器  
• --estimator (homography | affine) 用于转换估计的估计器类型
• --match_conf <float> 特征点匹配步骤的匹配置信度，最近邻匹配距离与次近邻匹配距离的比值，surf 默认为 0.65，orb 默认为 0.3    
• --conf_thresh <float> 两幅图来自同一全景图的置信度，默认为 1.0    
• --ba (no | reproj | ray | affine) 光束平均法的误差函数选择，默认是 ray 方法    
• --ba_refine_mask (mask) 光束平均法设置优化掩码
• --wave_correct (no|horiz|vert) 波形校验水平，垂直或者没有 默认是 horiz(水平)
• --save_graph <file_name> 将匹配的图形以点的形式保存到文件中， Nm 代表匹配的数量，NI代表正确匹配的数量，C 表示置信度
• /*图像融合参数:*/ 
• --warp (plane|cylindrical|spherical|fisheye|stereographic|compressedPlaneA2B1|compressedPla  neA1.5B1|compressedPlanePortraitA2B1|compressedPlanePortraitA1.5B1|paniniA2B1|paniniA1.5B1|paniniPortraitA2B1|paniniPor traitA1.5B1|mercator|transverseMercator)     选择融合的平面，默认是球形    
• --seam_megapix <float> 拼接缝像素的大小 默认是 0.1
• --seam (no|voronoi|gc_color|gc_colorgrad) 拼接缝隙估计方法 默认是 gc_color    
• --compose_megapix <float> 拼接分辨率，默认为-1    
• --expos_comp (no|gain|gain_blocks) 光照补偿方法，默认是 gain_blocks    
• --blend (no|feather|multiband) 融合方法，默认是多频段融合    
• --blend_strength <float> 融合强度，0-100.默认是 5.    
• --output <result_img> 输出图像的文件名，默认是 result,jpg     命令使用实例，以及程序运行时的提示： 

上面使用默认参数，详细输出信息如下：

E:\Practice\OpenCV\Algorithm_Summary\Image_Stitching\x64\Debug>05_Image_Stitch_Stitching_Detailed.exe ./imgs/boat1.jpg ./imgs/boat2.jpg ./imgs/boat3.jpg ./imgs/boat4.jpg ./imgs/boat5.jpg ./imgs/boat6.jpg  2 Finding features...  3 [ INFO:0] global C:\build\master_winpack-build-win64-vc15\opencv\modules\core\src\ocl.cpp (891) cv::ocl::haveOpenCL Initialize OpenCL runtime...  4 Features in image #1: 500
[ INFO:0] global C:\build\master_winpack-build-win64-vc15\opencv\modules\core\src\ocl.cpp (433) cv::ocl::OpenCLBinaryCacheConfigurator::OpenCLBinaryCacheConfigurator Successfully initialized OpenCL cache directory: C:\Users\A4080599\AppData\Local\Temp\opencv\4.4\opencl_cache\  6 [ INFO:0] global C:\build\master_winpack-build-win64-vc15\opencv\modules\core\src\ocl.cpp (457) cv::ocl::OpenCLBinaryCacheConfigurator::prepareCacheDirectoryForContext Preparing OpenCL cache configuration for context: NVIDIA_Corporation--GeForce_GTX_1070--411_31  7 Features in image #2: 500
Features in image #3: 500
Features in image #4: 500
Features in image #5: 500
Features in image #6: 500
Finding features, time: 5.46377 sec  13 Pairwise matchingPairwise matching, time: 3.24159 sec  14 Initial camera intrinsics #1:  15 K:  16 [534.6674906996568, 0, 474.5;  17  0, 534.6674906996568, 316;  18  0, 0, 1]  19 R:  20 [0.91843718, -0.09762425, -1.1678253;  21  0.0034433089, 1.0835428, -0.025021957;  22  0.28152198, 0.16100603, 0.91920781]  23 Initial camera intrinsics #2:  24 K:  25 [534.6674906996568, 0, 474.5;  26  0, 534.6674906996568, 316;  27  0, 0, 1]  28 R:  29 [1.001171, -0.085758291, -0.64530683;  30  0.010103324, 1.0520245, -0.030576767;  31  0.15743911, 0.12035993, 1]  32 Initial camera intrinsics #3:  33 K:  34 [534.6674906996568, 0, 474.5;  35  0, 534.6674906996568, 316;  36  0, 0, 1]  37 R:  38 [1, 0, 0;  39  0, 1, 0;  40  0, 0, 1]  41 Initial camera intrinsics #4:  42 K:  43 [534.6674906996568, 0, 474.5;  44  0, 534.6674906996568, 316;  45  0, 0, 1]  46 R:  47 [0.8474561, 0.028589081, 0.75133896;  48  -0.0014587968, 0.92028928, 0.033205934;  49  -0.17483309, 0.018777205, 0.84592116]  50 Initial camera intrinsics #5:  51 K:  52 [534.6674906996568, 0, 474.5;  53  0, 534.6674906996568, 316;  54  0, 0, 1]  55 R:  56 [0.60283858, 0.069275051, 1.2121853;  57  -0.014153662, 0.85474133, 0.014057174;  58  -0.29529575, 0.053770453, 0.61932623]  59 Initial camera intrinsics #6:  60 K:  61 [534.6674906996568, 0, 474.5;  62  0, 534.6674906996568, 316;  63  0, 0, 1]  64 R:  65 [0.41477469, 0.075901195, 1.4396564;  66  -0.015423983, 0.82344943, 0.0061162044;  67  -0.35168326, 0.055747174, 0.42653102]  68 Camera #1:  69 K:  70 [1068.953598931666, 0, 474.5;  71  0, 1068.953598931666, 316;  72  0, 0, 1]  73 R:  74 [0.84266716, -0.010490002, -0.53833258;  75  0.004485324, 0.99991232, -0.01246338;  76  0.53841609, 0.0080878884, 0.84264034]  77 Camera #2:  78 K:  79 [1064.878323247434, 0, 474.5;  80  0, 1064.878323247434, 316;  81  0, 0, 1]  82 R:  83 [0.95117813, -0.015436338, -0.3082563;  84  0.01137107, 0.99982315, -0.014980057;  85  0.308433, 0.010743499, 0.95118535]  86 Camera #3:  87 K:  88 [1065.382193682081, 0, 474.5;  89  0, 1065.382193682081, 316;  90  0, 0, 1]  91 R:  92 [1, -1.6298145e-09, 0;  93  -1.5716068e-09, 1, 0;  94  0, 0, 1]  95 Camera #4:  96 K:  97 [1067.611537959627, 0, 474.5;  98  0, 1067.611537959627, 316;  99  0, 0, 1] 100 R: 101 [0.91316396, -7.9067249e-06, 0.40759254; 102  -0.0075879274, 0.99982637, 0.017019274; 103  -0.4075219, -0.018634165, 0.91300529] 104 Camera #5: 105 K: 106 [1080.708135180496, 0, 474.5; 107  0, 1080.708135180496, 316; 108  0, 0, 1] 109 R: 110 [0.70923853, 0.0025724203, 0.70496398; 111  -0.0098195076, 0.99993235, 0.0062302947; 112  -0.70490021, -0.01134116, 0.70921582] 113 Camera #6: 114 K: 115 [1080.90412660159, 0, 474.5; 116  0, 1080.90412660159, 316; 117  0, 0, 1] 118 R: 119 [0.49985889, 3.5938341e-05, 0.86610687; 120  -0.00682831, 0.99996907, 0.0038993564; 121  -0.86607999, -0.0078631733, 0.49984369] 122 Warping images (auxiliary)... 123 Warping images, time: 0.0791121 sec 124 Compensating exposure... 125 Compensating exposure, time: 0.72288 sec 126 Finding seams... 127 Finding seams, time: 3.09237 sec 128 Compositing... 129 Compositing image #1
Multi-band blender, number of bands: 8
Compositing image #2
Compositing image #3
Compositing image #4
Compositing image #5
Compositing image #6
Compositing, time: 13.7766 sec 137 Finished, total time: 29.4535 sec

输入图像boat1.jpg、boat2.jpg、boat3.jpg、boat4.jpg、boat5.jpg、boat6.jpg如下(可以在OpenCV安装目录下找到D:\OpenCV4.4\opencv_extra-master\testdata\stitching)

 

 

 

结果图：

参数warp_type 设置为"plane"，效果图如下：

参数warp_type 设置为"fisheye"，效果图如下(旋转90°后)：

其他的参数可以根据自己需要修改，如果要自己完成还需要详细了解拼接步骤再优化。