前面已经对串联匹配有了一定的了解,现在用它来改进 Opencv 的stitching ,
先找来三个博文为模板,分别是:
1。《任意n张图像拼接_效果很好_计算机视觉大作业1终版》
2。《 Opencv2.4.9源码分析——Stitching(八)》
3。《图像拼接(十):OPenCV stitching和stitching_detailed》中的“stitching_detailed使用示例”
把他们中的一些Mat 转化为opencv 3.0 用到的 UMat 。
为什么不直接用3.0的示例呢?主要是示例不太友好方便,修改地方太多,自己的e文也太差。
通过测试:
1文只有一种长宽比,改变长宽比就出错。
2文速度较慢,注解不错。
3文没有中文注解,但速度较快,所以就以3文为模板修改匹配。
针对3.0修改后为:
//stitching_detailed使用 3.0 //用串联匹配代替原匹配 // #define ENABLE_LOG 1 #include <iostream> #include <fstream> #include <string> #include "opencv2/opencv_modules.hpp" #include "opencv2/highgui/highgui.hpp" #include "opencv2/stitching/detail/autocalib.hpp" #include "opencv2/stitching/detail/blenders.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/util.hpp" #include "opencv2/stitching/detail/warpers.hpp" #include "opencv2/stitching/warpers.hpp" using namespace std; using namespace cv; using namespace cv::detail; // // 默认命令行参数 vector<string> img_names; bool preview = false; bool try_gpu = true; double work_megapix = 0.6; double seam_megapix = 0.1; double compose_megapix = -1; float conf_thresh = 1.f; string features_type = "orb";//"surf"; string ba_cost_func = "reproj";//重映射误差方法 "ray";//射线发散误差方法 string ba_refine_mask = "xxxxx"; bool do_wave_correct = true; WaveCorrectKind wave_correct = detail::WAVE_CORRECT_HORIZ;// 波形校验,水平 // 波校正垂直 WAVE_CORRECT_VERT bool save_graph = false;//是否保存匹配图 std::string save_graph_to; string warp_type = "spherical";//球面投影 int expos_comp_type = ExposureCompensator::GAIN_BLOCKS; float match_conf = 0.3f; string seam_find_type = "gc_color"; int blend_type = Blender::MULTI_BAND; float blend_strength = 5; string result_name = "result.jpg"; void LoadImageNamesFromFile(char* name,vector<string>& image_names);//从列表中载入图像名 void i_matcher(vector<ImageFeatures> &features, vector<MatchesInfo> &pairwise_matches); int main(int argc, char* argv[]) { //读入图像 double ttt = getTickCount(); cout<<"读出文件名..."<<endl; LoadImageNamesFromFile("list.txt",img_names);//从list.txt文件装载图像文件名 #if ENABLE_LOG int64 app_start_time = getTickCount(); #endif cv::setBreakOnError(true); /*int retval = parseCmdArgs(argc, argv); if (retval) return retval;*/ // Check if have enough images int num_images = static_cast<int>(img_names.size()); cout<<"有 "<<num_images<<" 个图"<<endl; if (num_images < 2) { LOGLN("Need more images"); return -1; } double work_scale = 1, seam_scale = 1, compose_scale = 1; bool is_work_scale_set = false, is_seam_scale_set = false, is_compose_scale_set = false; //LOGLN("Finding features..."); cout<<"正在寻找图像特征..."<<endl; #if ENABLE_LOG int64 t = getTickCount(); #endif Ptr<FeaturesFinder> finder; if (features_type == "surf") { #if defined(HAVE_OPENCV_NONFREE) && defined(HAVE_OPENCV_GPU) if (try_gpu && gpu::getCudaEnabledDeviceCount() > 0) finder = new SurfFeaturesFinderGpu(); else #endif finder = new SurfFeaturesFinder(); } else if (features_type == "orb") { finder = new OrbFeaturesFinder(); } else { cout << "Unknown 2D features type: '" << features_type << "'.\n"; return -1; } Mat full_img, img; vector<ImageFeatures> features(num_images); vector<Mat> images(num_images); vector<Size> full_img_sizes(num_images); double seam_work_aspect = 1; for (int i = 0; i < num_images; ++i) { full_img = imread(img_names[i]); full_img_sizes[i] = full_img.size(); if (full_img.empty()) { LOGLN("Can't open image " << img_names[i]); return -1; } if (work_megapix < 0) { img = full_img; work_scale = 1; is_work_scale_set = true; } else { if (!is_work_scale_set) { work_scale = min(1.0, sqrt(work_megapix * 1e6 / full_img.size().area())); is_work_scale_set = true; } resize(full_img, img, Size(), work_scale, work_scale); } if (!is_seam_scale_set) { seam_scale = min(1.0, sqrt(seam_megapix * 1e6 / full_img.size().area())); seam_work_aspect = seam_scale / work_scale; is_seam_scale_set = true; } (*finder)(img, features[i]); features[i].img_idx = i; //LOGLN("Features in image #" << i+1 << ": " << features[i].keypoints.size()); cout<<"图像 #" << i+1 << "特征数: " << features[i].keypoints.size()<<endl; resize(full_img, img, Size(), seam_scale, seam_scale); images[i] = img.clone(); } finder->collectGarbage(); full_img.release(); img.release(); //LOGLN("Finding features, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); cout<<"寻找特征用时:"<< ((getTickCount() - t) / getTickFrequency()) << " 秒"<<endl; //LOG("Pairwise matching"); cout<<"正在成对匹配..."<<endl; #if ENABLE_LOG t = getTickCount(); #endif vector<MatchesInfo> pairwise_matches; BestOf2NearestMatcher matcher(try_gpu, match_conf); //matcher(features, pairwise_matches); //matcher.collectGarbage(); //这里用我们自己的匹配代替 i_matcher(features, pairwise_matches); //LOGLN("Pairwise matching, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); cout<<"成对匹配用时:"<< ((getTickCount() - t) / getTickFrequency()) << " 秒"<<endl; //检查是否应保存匹配图 if (save_graph) { //LOGLN("Saving matches graph..."); cout << "正在保存匹配图..." << endl; ofstream f(save_graph_to.c_str()); vector<String> is_img_names;//string转化为Strig。 for(size_t i=0;i<img_names.size ();i++) is_img_names.push_back(img_names[i]); f << matchesGraphAsString(is_img_names, pairwise_matches, conf_thresh); } // 只留下我们确信来自同一全景的图像 vector<int> indices = leaveBiggestComponent(features, pairwise_matches, conf_thresh); vector<Mat> img_subset; vector<string> img_names_subset; vector<Size> full_img_sizes_subset; for (size_t i = 0; i < indices.size(); ++i) { img_names_subset.push_back(img_names[indices[i]]); img_subset.push_back(images[indices[i]]); full_img_sizes_subset.push_back(full_img_sizes[indices[i]]); } images = img_subset; img_names = img_names_subset; full_img_sizes = full_img_sizes_subset; // 检查我们是否还有足够的图像 num_images = static_cast<int>(img_names.size()); cout << "要拼接的图像数:" <<num_images<< endl; if (num_images < 2) { LOGLN("Need more images"); return -1; } cout << "相机参数(粗)..."<< endl; HomographyBasedEstimator estimator; vector<CameraParams> cameras;//表示相机参数矢量队列 estimator(features, pairwise_matches, cameras);//相机参数评估 for (size_t i = 0; i < cameras.size(); ++i) { Mat R; cameras[i].R.convertTo(R, CV_32F); cameras[i].R = R; //LOGLN("Initial intrinsics #" << indices[i]+1 << ":\n" << cameras[i].K()); //cout<<"初始内参 #" << indices[i]+1 << ":\n" << cameras[i].K()<<endl; } //cout << "相机参数(细)..."<< endl;//光束平差法图像太多会卡在这里,先去掉 // Ptr<detail::BundleAdjusterBase> adjuster; // if (ba_cost_func == "reproj") adjuster = new detail::BundleAdjusterReproj(); // else if (ba_cost_func == "ray") adjuster = new detail::BundleAdjusterRay(); // else // { // cout << "Unknown bundle adjustment cost function: '" << ba_cost_func << "'.\n"; // return -1; // } // adjuster->setConfThresh(conf_thresh); // Mat_<uchar> refine_mask = Mat::zeros(3, 3, CV_8U); // if (ba_refine_mask[0] == 'x') refine_mask(0,0) = 1; // if (ba_refine_mask[1] == 'x') refine_mask(0,1) = 1; // if (ba_refine_mask[2] == 'x') refine_mask(0,2) = 1; // if (ba_refine_mask[3] == 'x') refine_mask(1,1) = 1; // if (ba_refine_mask[4] == 'x') refine_mask(1,2) = 1; // adjuster->setRefinementMask(refine_mask); // (*adjuster)(features, pairwise_matches, cameras); // 求出的焦距取中值 vector<double> focals; for (size_t i = 0; i < cameras.size(); ++i) { //LOGLN("Camera #" << indices[i]+1 << ":\n" << cameras[i].K()); focals.push_back(cameras[i].focal); } sort(focals.begin(), focals.end()); float warped_image_scale; if (focals.size() % 2 == 1) warped_image_scale = static_cast<float>(focals[focals.size() / 2]); else warped_image_scale = static_cast<float>(focals[focals.size() / 2 - 1] + focals[focals.size() / 2]) * 0.5f; if (do_wave_correct) { vector<Mat> rmats; for (size_t i = 0; i < cameras.size(); ++i) rmats.push_back(cameras[i].R.clone()); waveCorrect(rmats, wave_correct); for (size_t i = 0; i < cameras.size(); ++i) cameras[i].R = rmats[i]; } //LOGLN("Warping images (auxiliary)... "); cout<<"正在扭曲图像(辅助)..."<<endl; #if ENABLE_LOG t = getTickCount(); #endif vector<Point> corners(num_images); vector<Mat> masks_warped(num_images); vector<Mat> images_warped(num_images); vector<Size> sizes(num_images); vector<Mat> masks(num_images); // Preapre images masks for (int i = 0; i < num_images; ++i) { masks[i].create(images[i].size(), CV_8U); masks[i].setTo(Scalar::all(255)); } // Warp images and their masks Ptr<WarperCreator> warper_creator; #if defined(HAVE_OPENCV_GPU) if (try_gpu && gpu::getCudaEnabledDeviceCount() > 0) { if (warp_type == "plane") warper_creator = new cv::PlaneWarperGpu(); else if (warp_type == "cylindrical") warper_creator = new cv::CylindricalWarperGpu(); else if (warp_type == "spherical") warper_creator = new cv::SphericalWarperGpu(); } else #endif { if (warp_type == "plane") warper_creator = new cv::PlaneWarper(); else if (warp_type == "cylindrical") warper_creator = new cv::CylindricalWarper(); else if (warp_type == "spherical") warper_creator = new cv::SphericalWarper(); else if (warp_type == "fisheye") warper_creator = new cv::FisheyeWarper(); else if (warp_type == "stereographic") warper_creator = new cv::StereographicWarper(); else if (warp_type == "compressedPlaneA2B1") warper_creator = new cv::CompressedRectilinearWarper(2, 1); else if (warp_type == "compressedPlaneA1.5B1") warper_creator = new cv::CompressedRectilinearWarper(1.5, 1); else if (warp_type == "compressedPlanePortraitA2B1") warper_creator = new cv::CompressedRectilinearPortraitWarper(2, 1); else if (warp_type == "compressedPlanePortraitA1.5B1") warper_creator = new cv::CompressedRectilinearPortraitWarper(1.5, 1); else if (warp_type == "paniniA2B1") warper_creator = new cv::PaniniWarper(2, 1); else if (warp_type == "paniniA1.5B1") warper_creator = new cv::PaniniWarper(1.5, 1); else if (warp_type == "paniniPortraitA2B1") warper_creator = new cv::PaniniPortraitWarper(2, 1); else if (warp_type == "paniniPortraitA1.5B1") warper_creator = new cv::PaniniPortraitWarper(1.5, 1); else if (warp_type == "mercator") warper_creator = new cv::MercatorWarper(); else if (warp_type == "transverseMercator") warper_creator = new cv::TransverseMercatorWarper(); } if (warper_creator.empty()) { cout << "Can't create the following warper '" << warp_type << "'\n"; return 1; } Ptr<RotationWarper> warper = warper_creator->create(static_cast<float>(warped_image_scale * seam_work_aspect)); for (int i = 0; i < num_images; ++i) { Mat_<float> K; cameras[i].K().convertTo(K, CV_32F); float swa = (float)seam_work_aspect; K(0,0) *= swa; K(0,2) *= swa; K(1,1) *= swa; K(1,2) *= swa; corners[i] = warper->warp(images[i], K, cameras[i].R, INTER_LINEAR, BORDER_REFLECT, images_warped[i]); sizes[i] = images_warped[i].size(); warper->warp(masks[i], K, cameras[i].R, INTER_NEAREST, BORDER_CONSTANT, masks_warped[i]); } vector<Mat> images_warped_f(num_images); for (int i = 0; i < num_images; ++i) images_warped[i].convertTo(images_warped_f[i], CV_32F); //LOGLN("Warping images, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); cout<<"扭曲图像用时:"<< ((getTickCount() - t) / getTickFrequency()) << " 秒"<<endl; cout<<"正在曝光补偿..."<<endl; Ptr<ExposureCompensator> compensator = ExposureCompensator::createDefault(expos_comp_type); //Mat转变为UMat----------------------------------------------------------开始 vector<UMat> u_images_warped; //复制一个 for (unsigned int i = 0; i < images_warped.size(); ++i) u_images_warped.push_back(images_warped[i].getUMat(cv::ACCESS_READ)); vector<UMat> u_masks_warped; //复制一个 for (unsigned int i = 0; i < masks_warped.size(); ++i) u_masks_warped.push_back(masks_warped[i].getUMat(cv::ACCESS_READ)); //Mat转变为UMat----------------------------------------------------------结束 compensator->feed(corners, u_images_warped, u_masks_warped); //compensator->feed(corners, images_warped, masks_warped); cout<<"正在寻找接缝..."<<endl; Ptr<SeamFinder> seam_finder; if (seam_find_type == "no") seam_finder = new detail::NoSeamFinder(); else if (seam_find_type == "voronoi") seam_finder = new detail::VoronoiSeamFinder(); else if (seam_find_type == "gc_color") { #if defined(HAVE_OPENCV_GPU) if (try_gpu && gpu::getCudaEnabledDeviceCount() > 0) seam_finder = new detail::GraphCutSeamFinderGpu(GraphCutSeamFinderBase::COST_COLOR); else #endif seam_finder = new detail::GraphCutSeamFinder(GraphCutSeamFinderBase::COST_COLOR); } else if (seam_find_type == "gc_colorgrad") { #if defined(HAVE_OPENCV_GPU) if (try_gpu && gpu::getCudaEnabledDeviceCount() > 0) seam_finder = new detail::GraphCutSeamFinderGpu(GraphCutSeamFinderBase::COST_COLOR_GRAD); else #endif seam_finder = new detail::GraphCutSeamFinder(GraphCutSeamFinderBase::COST_COLOR_GRAD); } else if (seam_find_type == "dp_color") seam_finder = new detail::DpSeamFinder(DpSeamFinder::COLOR); else if (seam_find_type == "dp_colorgrad") seam_finder = new detail::DpSeamFinder(DpSeamFinder::COLOR_GRAD); if (seam_finder.empty()) { cout << "Can't create the following seam finder '" << seam_find_type << "'\n"; return 1; } //Mat转变为UMat----------------------------------------------------------开始 vector<UMat> u_images_warped_f; //images_warped_f 转化为 u_images_warped_f for (unsigned int i = 0; i < images_warped_f.size(); ++i) u_images_warped_f.push_back(images_warped_f[i].getUMat(cv::ACCESS_READ)); //Mat转变为UMat----------------------------------------------------------结束 //seam_finder->find(images_warped_f, corners, masks_warped); seam_finder->find(u_images_warped_f, corners, u_masks_warped); // // 释放不再使用的内存 images.clear(); images_warped.clear(); images_warped_f.clear(); masks.clear(); //LOGLN("Compositing..."); cout<<"正在合成图像..." <<endl; #if ENABLE_LOG t = getTickCount(); #endif Mat img_warped, img_warped_s; Mat dilated_mask, seam_mask, mask, mask_warped; Ptr<Blender> blender; //double compose_seam_aspect = 1; double compose_work_aspect = 1; for (int img_idx = 0; img_idx < num_images; ++img_idx) { //LOGLN("Compositing image #" << indices[img_idx]+1); cout<<"合成图像 #" << indices[img_idx]+1<<endl; // Read image and resize it if necessary full_img = imread(img_names[img_idx]); if (!is_compose_scale_set) { if (compose_megapix > 0) compose_scale = min(1.0, sqrt(compose_megapix * 1e6 / full_img.size().area())); is_compose_scale_set = true; // 计算相对比例 //compose_seam_aspect = compose_scale / seam_scale; compose_work_aspect = compose_scale / work_scale; // 更新扭曲图像比例 warped_image_scale *= static_cast<float>(compose_work_aspect); warper = warper_creator->create(warped_image_scale); // 更新角和尺寸 for (int i = 0; i < num_images; ++i) { // 更新本质 cameras[i].focal *= compose_work_aspect; cameras[i].ppx *= compose_work_aspect; cameras[i].ppy *= compose_work_aspect; // Update corner and size Size sz = full_img_sizes[i]; if (std::abs(compose_scale - 1) > 1e-1) { sz.width = cvRound(full_img_sizes[i].width * compose_scale); sz.height = cvRound(full_img_sizes[i].height * compose_scale); } Mat K; cameras[i].K().convertTo(K, CV_32F); Rect roi = warper->warpRoi(sz, K, cameras[i].R); corners[i] = roi.tl(); sizes[i] = roi.size(); } } if (abs(compose_scale - 1) > 1e-1) resize(full_img, img, Size(), compose_scale, compose_scale); else img = full_img; full_img.release(); Size img_size = img.size(); Mat K; cameras[img_idx].K().convertTo(K, CV_32F); // 扭曲当前图像 warper->warp(img, K, cameras[img_idx].R, INTER_LINEAR, BORDER_REFLECT, img_warped); // 扭曲当前图像掩码 mask.create(img_size, CV_8U); mask.setTo(Scalar::all(255)); warper->warp(mask, K, cameras[img_idx].R, INTER_NEAREST, BORDER_CONSTANT, mask_warped); // 曝光补偿 compensator->apply(img_idx, corners[img_idx], img_warped, mask_warped); img_warped.convertTo(img_warped_s, CV_16S); img_warped.release(); img.release(); mask.release(); dilate(masks_warped[img_idx], dilated_mask, Mat()); resize(dilated_mask, seam_mask, mask_warped.size()); mask_warped = seam_mask & mask_warped; if (blender.empty()) { blender = Blender::createDefault(blend_type, try_gpu); Size dst_sz = resultRoi(corners, sizes).size(); float blend_width = sqrt(static_cast<float>(dst_sz.area())) * blend_strength / 100.f; if (blend_width < 1.f) blender = Blender::createDefault(Blender::NO, try_gpu); else if (blend_type == Blender::MULTI_BAND) { MultiBandBlender* mb = dynamic_cast<MultiBandBlender*>(static_cast<Blender*>(blender)); mb->setNumBands(static_cast<int>(ceil(log(blend_width)/log(2.)) - 1.)); //LOGLN("Multi-band blender, number of bands: " << mb->numBands()); cout<<" 多频段图像融合, 分段数: " << mb->numBands()<<endl; } else if (blend_type == Blender::FEATHER) { FeatherBlender* fb = dynamic_cast<FeatherBlender*>(static_cast<Blender*>(blender)); fb->setSharpness(1.f/blend_width); //LOGLN("Feather blender, sharpness: " << fb->sharpness()); cout<<"羽化融合,清晰度:" << fb->sharpness() <<endl; } blender->prepare(corners, sizes); } // 混合当前图像 blender->feed(img_warped_s, mask_warped, corners[img_idx]); } Mat result, result_mask; blender->blend(result, result_mask); //LOGLN("Compositing, time: " << ((getTickCount() - t) / getTickFrequency()) << " sec"); cout<<"图像混合用时:"<< ((getTickCount() - t) / getTickFrequency()) << " 秒"<<endl; imwrite(result_name, result); result.convertTo(result,CV_8UC1); imshow("stitch",result); ttt = ((double)getTickCount() - ttt) / getTickFrequency(); cout << "总的拼接用时:" << ttt << " 秒"<< endl; waitKey(0); //LOGLN("Finished, total time: " << ((getTickCount() - app_start_time) / getTickFrequency()) << " sec"); return 0; }
由于光束平差太卡,也去掉了
串联匹配以博文《Opencv2.4.9源码分析——Stitching(二)》为模板作为一个单独cpp,
比便简单,直接看吧:
//串联匹配 #define ENABLE_LOG 1 #include <opencv2/opencv.hpp> #include "opencv2/stitching/detail/autocalib.hpp" #include "opencv2/stitching/detail/blenders.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/util.hpp" #include "opencv2/stitching/detail/warpers.hpp" #include "opencv2/stitching/warpers.hpp" #include <iostream> #include <fstream> #include <string> #include <iomanip> using namespace cv; using namespace std; using namespace detail; void f2_matcher(vector<ImageFeatures> &features,vector<MatchesInfo> &f2_matches) { //vector<MatchesInfo> f2_matches; //特征匹配 BestOf2NearestMatcher matcher(false, 0.3f, 6, 6); //定义特征匹配器,2NN方法 matcher(features, f2_matches); //进行特征匹配 } void i_matcher(vector<ImageFeatures> &features, vector<MatchesInfo> &pairwise_matches) { int num_images=features.size (); //1。串联匹配 vector<vector<MatchesInfo>> f2_2;//f2_2[i] 表示 i 和 i+1 的匹配关系(0 开头,比图像数小 1) for (int i = 1; i < num_images; ++i) { vector<ImageFeatures> f2; vector<MatchesInfo> m2; f2.push_back (features[i-1]); f2.push_back (features[i]); f2_matcher(f2,m2); f2_2.push_back(m2); } //2。把串联匹配 ----按opencv stitching 拼接的匹配关系组在一起 MatchesInfo f;//大小: n x n (n个图) for (int i = 0; i < num_images; ++i) { for (int j = 0; j < num_images; ++j) { //cout<<"i,j:"<<i<<","<<j<<endl; if(i==j)//自身不用匹配 { f.src_img_idx = -1; f.dst_img_idx = -1; f.num_inliers = 0; f.confidence = 0; pairwise_matches.push_back (f); }else if(i+1==j)//相连(顺) { pairwise_matches.push_back (f2_2[i][1]); //修改匹配关系 pairwise_matches[pairwise_matches.size ()-1].src_img_idx =i; pairwise_matches[pairwise_matches.size ()-1].dst_img_idx =j; }else if(j+1==i)//相连(倒) { pairwise_matches.push_back (f2_2[j][2]); //修改匹配关系 pairwise_matches[pairwise_matches.size ()-1].src_img_idx =i; pairwise_matches[pairwise_matches.size ()-1].dst_img_idx =j; }else//其它略过 { f.src_img_idx = -1; f.dst_img_idx = -1; f.num_inliers = 0; f.confidence = 0; pairwise_matches.push_back (f); } //cout<<"size:"<<pairwise_matches.size ()<<endl; } } }
用前面的83个图来测试下,到曝光补偿这步就内存溢出而出错了,当减少到40个图时就出来了
用时 408.58 秒,6分多,7分钟不到