随笔- 1525 文章- 2 评论- 66 阅读- 128万

G2O(2) 基本例子 3D-3D位姿求解 -（一元点多边 3D点对位姿求解）求解3D点1到3D点2的变换关系而不是优化3D点1的位姿

残差

1通常2D像素对3D点位姿和点

2 但是这个里面没有2D像素，是单纯的3D点对3D点位姿求解

CMakeLists.txt

cmake_minimum_required(VERSION 2.8)
project(vo1)
 
set(CMAKE_BUILD_TYPE "Release")
add_definitions("-DENABLE_SSE")
set(CMAKE_CXX_FLAGS "-std=c++11 -O2 ${SSE_FLAGS} -msse4")
list(APPEND CMAKE_MODULE_PATH ${PROJECT_SOURCE_DIR}/cmake)
 
find_package(OpenCV 3 REQUIRED)
find_package(G2O REQUIRED)
find_package(Sophus REQUIRED)
 
include_directories(
        ${OpenCV_INCLUDE_DIRS}
        ${G2O_INCLUDE_DIRS}
        ${Sophus_INCLUDE_DIRS}
        "/usr/include/eigen3/"
)
 
add_executable(orb_cv orb_cv.cpp)
target_link_libraries(orb_cv ${OpenCV_LIBS})
 
add_executable(orb_self orb_self.cpp)
target_link_libraries(orb_self ${OpenCV_LIBS})
 
# add_executable( pose_estimation_2d2d pose_estimation_2d2d.cpp extra.cpp ) # use this if in OpenCV2 
add_executable(pose_estimation_2d2d pose_estimation_2d2d.cpp)
target_link_libraries(pose_estimation_2d2d ${OpenCV_LIBS})
 
# # add_executable( triangulation triangulation.cpp extra.cpp) # use this if in opencv2
add_executable(triangulation triangulation.cpp)
target_link_libraries(triangulation ${OpenCV_LIBS})
 
add_executable(pose_estimation_3d2d pose_estimation_3d2d.cpp)
target_link_libraries(pose_estimation_3d2d
        g2o_core g2o_stuff
        ${OpenCV_LIBS})
 
add_executable(pose_estimation_3d3d pose_estimation_3d3d.cpp)
target_link_libraries(pose_estimation_3d3d
        g2o_core g2o_stuff
        ${OpenCV_LIBS})

mian

100

101

102

103

104

105

106

107

108

109

110

111

112

113

114

115

116

117

118

119

120

121

122

123

124

125

126

127

128

129

130

131

132

133

134

135

136

137

138

139

140

141

142

143

144

145

146

147

148

149

150

151

152

153

154

155

156

157

158

159

160

161

162

163

164

165

166

167

168

169

170

171

172

173

174

175

176

177

178

179

180

181

182

183

184

185

186

187

188

189

190

191

192

193

194

195

196

197

198

199

200

201

202

203

204

205

206

207

208

209

210

211

212

213

214

215

216

217

218

219

220

221

222

223

224

225

226

227

228

229

230

231

232

233

234

235

236

237

238

239

240

241

242

243

244

245

246

247

248

249

250

251

252

253

254

255

256

257

258

259

260

261

262

263

264

265

266

267

268

269

270

271

272

273

274

275

276

277

278

279

280

281

282

283

284

285

286

287

288

289

290

291

292

293

294

295

296

297

298

#include <iostream>
#include <opencv2/core/core.hpp>
#include <opencv2/features2d/features2d.hpp>
#include <opencv2/highgui/highgui.hpp>
#include <opencv2/calib3d/calib3d.hpp>
#include <Eigen/Core>
#include <Eigen/Dense>
#include <Eigen/Geometry>
#include <Eigen/SVD>
#include <g2o/core/base_vertex.h>
#include <g2o/core/base_unary_edge.h>
#include <g2o/core/block_solver.h>
#include <g2o/core/optimization_algorithm_gauss_newton.h>
#include <g2o/core/optimization_algorithm_levenberg.h>
#include <g2o/solvers/dense/linear_solver_dense.h>
#include <chrono>
#include <sophus/se3.hpp>
 
using namespace std;
using namespace cv;
 
void find_feature_matches(
  const Mat &img_1, const Mat &img_2,
  std::vector<KeyPoint> &keypoints_1,
  std::vector<KeyPoint> &keypoints_2,
  std::vector<DMatch> &matches);
 
// 像素坐标转相机归一化坐标
Point2d pixel2cam(const Point2d &p, const Mat &K);
 
void pose_estimation_3d3d(
  const vector<Point3f> &pts1,
  const vector<Point3f> &pts2,
  Mat &R, Mat &t
);
 
void bundleAdjustment(
  const vector<Point3f> &points_3d,
  const vector<Point3f> &points_2d,
  Mat &R, Mat &t
);
 
/// vertex and edges used in g2o ba
class VertexPose : public g2o::BaseVertex<6, Sophus::SE3d> {
public:
  EIGEN_MAKE_ALIGNED_OPERATOR_NEW;
 
  virtual void setToOriginImpl() override {
    _estimate = Sophus::SE3d();
  }
 
  /// left multiplication on SE3
  virtual void oplusImpl(const double *update) override {
    Eigen::Matrix<double, 6, 1> update_eigen;
    update_eigen << update[0], update[1], update[2], update[3], update[4], update[5];
    _estimate = Sophus::SE3d::exp(update_eigen) * _estimate;
  }
 
  virtual bool read(istream &in) override {}
 
  virtual bool write(ostream &out) const override {}
};
 
/// g2o edge
class EdgeProjectXYZRGBDPoseOnly : public g2o::BaseUnaryEdge<3, Eigen::Vector3d, VertexPose> {
public:
  EIGEN_MAKE_ALIGNED_OPERATOR_NEW;
 
  EdgeProjectXYZRGBDPoseOnly(const Eigen::Vector3d &point) : _point(point) {}
 
  virtual void computeError() override {
    const VertexPose *pose = static_cast<const VertexPose *> ( _vertices[0] );
    _error = _measurement - pose->estimate() * _point;
  }
 
  virtual void linearizeOplus() override {
    VertexPose *pose = static_cast<VertexPose *>(_vertices[0]);
    Sophus::SE3d T = pose->estimate();
    Eigen::Vector3d xyz_trans = T * _point;
    _jacobianOplusXi.block<3, 3>(0, 0) = -Eigen::Matrix3d::Identity();
    _jacobianOplusXi.block<3, 3>(0, 3) = Sophus::SO3d::hat(xyz_trans);
  }
 
  bool read(istream &in) {}
 
  bool write(ostream &out) const {}
 
protected:
  Eigen::Vector3d _point;
};
 
int main(int argc, char **argv) {
  if (argc != 5) {
    cout << "usage: pose_estimation_3d3d img1 img2 depth1 depth2" << endl;
    return 1;
  }
  //-- 读取图像
  Mat img_1 = imread(argv[1], CV_LOAD_IMAGE_COLOR);
  Mat img_2 = imread(argv[2], CV_LOAD_IMAGE_COLOR);
 
  vector<KeyPoint> keypoints_1, keypoints_2;
  vector<DMatch> matches;
  find_feature_matches(img_1, img_2, keypoints_1, keypoints_2, matches);
  cout << "一共找到了" << matches.size() << "组匹配点" << endl;
 
  // 建立3D点
  Mat depth1 = imread(argv[3], CV_LOAD_IMAGE_UNCHANGED);       // 深度图为16位无符号数，单通道图像
  Mat depth2 = imread(argv[4], CV_LOAD_IMAGE_UNCHANGED);       // 深度图为16位无符号数，单通道图像
  Mat K = (Mat_<double>(3, 3) << 520.9, 0, 325.1, 0, 521.0, 249.7, 0, 0, 1);
  vector<Point3f> pts1, pts2;
 
  for (DMatch m:matches) {
    ushort d1 = depth1.ptr<unsigned short>(int(keypoints_1[m.queryIdx].pt.y))[int(keypoints_1[m.queryIdx].pt.x)];
    ushort d2 = depth2.ptr<unsigned short>(int(keypoints_2[m.trainIdx].pt.y))[int(keypoints_2[m.trainIdx].pt.x)];
    if (d1 == 0 || d2 == 0)   // bad depth
      continue;
    Point2d p1 = pixel2cam(keypoints_1[m.queryIdx].pt, K);
    Point2d p2 = pixel2cam(keypoints_2[m.trainIdx].pt, K);
    float dd1 = float(d1) / 5000.0;
    float dd2 = float(d2) / 5000.0;
    pts1.push_back(Point3f(p1.x * dd1, p1.y * dd1, dd1));
    pts2.push_back(Point3f(p2.x * dd2, p2.y * dd2, dd2));
  }
 
  cout << "3d-3d pairs: " << pts1.size() << endl;
  Mat R, t;
  pose_estimation_3d3d(pts1, pts2, R, t);
  cout << "ICP via SVD results: " << endl;
  cout << "R = " << R << endl;
  cout << "t = " << t << endl;
  cout << "R_inv = " << R.t() << endl;
  cout << "t_inv = " << -R.t() * t << endl;
 
  cout << "calling bundle adjustment" << endl;
 
  bundleAdjustment(pts1, pts2, R, t);
 
  // verify p1 = R * p2 + t
  for (int i = 0; i < 5; i++) {
    cout << "p1 = " << pts1[i] << endl;
    cout << "p2 = " << pts2[i] << endl;
    cout << "(R*p2+t) = " <<
         R * (Mat_<double>(3, 1) << pts2[i].x, pts2[i].y, pts2[i].z) + t
         << endl;
    cout << endl;
  }
}
 
void find_feature_matches(const Mat &img_1, const Mat &img_2,
                          std::vector<KeyPoint> &keypoints_1,
                          std::vector<KeyPoint> &keypoints_2,
                          std::vector<DMatch> &matches) {
  //-- 初始化
  Mat descriptors_1, descriptors_2;
  // used in OpenCV3
  Ptr<FeatureDetector> detector = ORB::create();
  Ptr<DescriptorExtractor> descriptor = ORB::create();
  // use this if you are in OpenCV2
  // Ptr<FeatureDetector> detector = FeatureDetector::create ( "ORB" );
  // Ptr<DescriptorExtractor> descriptor = DescriptorExtractor::create ( "ORB" );
  Ptr<DescriptorMatcher> matcher = DescriptorMatcher::create("BruteForce-Hamming");
  //-- 第一步:检测 Oriented FAST 角点位置
  detector->detect(img_1, keypoints_1);
  detector->detect(img_2, keypoints_2);
 
  //-- 第二步:根据角点位置计算 BRIEF 描述子
  descriptor->compute(img_1, keypoints_1, descriptors_1);
  descriptor->compute(img_2, keypoints_2, descriptors_2);
 
  //-- 第三步:对两幅图像中的BRIEF描述子进行匹配，使用 Hamming 距离
  vector<DMatch> match;
  // BFMatcher matcher ( NORM_HAMMING );
  matcher->match(descriptors_1, descriptors_2, match);
 
  //-- 第四步:匹配点对筛选
  double min_dist = 10000, max_dist = 0;
 
  //找出所有匹配之间的最小距离和最大距离, 即是最相似的和最不相似的两组点之间的距离
  for (int i = 0; i < descriptors_1.rows; i++) {
    double dist = match[i].distance;
    if (dist < min_dist) min_dist = dist;
    if (dist > max_dist) max_dist = dist;
  }
 
  printf("-- Max dist : %f \n", max_dist);
  printf("-- Min dist : %f \n", min_dist);
 
  //当描述子之间的距离大于两倍的最小距离时,即认为匹配有误.但有时候最小距离会非常小,设置一个经验值30作为下限.
  for (int i = 0; i < descriptors_1.rows; i++) {
    if (match[i].distance <= max(2 * min_dist, 30.0)) {
      matches.push_back(match[i]);
    }
  }
}
 
Point2d pixel2cam(const Point2d &p, const Mat &K) {
  return Point2d(
    (p.x - K.at<double>(0, 2)) / K.at<double>(0, 0),
    (p.y - K.at<double>(1, 2)) / K.at<double>(1, 1)
  );
}
 
void pose_estimation_3d3d(const vector<Point3f> &pts1,
                          const vector<Point3f> &pts2,
                          Mat &R, Mat &t) {
  Point3f p1, p2;     // center of mass
  int N = pts1.size();
  for (int i = 0; i < N; i++) {
    p1 += pts1[i];
    p2 += pts2[i];
  }
  p1 = Point3f(Vec3f(p1) / N);
  p2 = Point3f(Vec3f(p2) / N);
  vector<Point3f> q1(N), q2(N); // remove the center
  for (int i = 0; i < N; i++) {
    q1[i] = pts1[i] - p1;
    q2[i] = pts2[i] - p2;
  }
 
  // compute q1*q2^T
  Eigen::Matrix3d W = Eigen::Matrix3d::Zero();
  for (int i = 0; i < N; i++) {
    W += Eigen::Vector3d(q1[i].x, q1[i].y, q1[i].z) * Eigen::Vector3d(q2[i].x, q2[i].y, q2[i].z).transpose();
  }
  cout << "W=" << W << endl;
 
  // SVD on W
  Eigen::JacobiSVD<Eigen::Matrix3d> svd(W, Eigen::ComputeFullU | Eigen::ComputeFullV);
  Eigen::Matrix3d U = svd.matrixU();
  Eigen::Matrix3d V = svd.matrixV();
 
  cout << "U=" << U << endl;
  cout << "V=" << V << endl;
 
  Eigen::Matrix3d R_ = U * (V.transpose());
  if (R_.determinant() < 0) {
    R_ = -R_;
  }
  Eigen::Vector3d t_ = Eigen::Vector3d(p1.x, p1.y, p1.z) - R_ * Eigen::Vector3d(p2.x, p2.y, p2.z);
 
  // convert to cv::Mat
  R = (Mat_<double>(3, 3) <<
    R_(0, 0), R_(0, 1), R_(0, 2),
    R_(1, 0), R_(1, 1), R_(1, 2),
    R_(2, 0), R_(2, 1), R_(2, 2)
  );
  t = (Mat_<double>(3, 1) << t_(0, 0), t_(1, 0), t_(2, 0));
}
 
void bundleAdjustment(
  const vector<Point3f> &pts1,
  const vector<Point3f> &pts2,
  Mat &R, Mat &t) {
  // 构建图优化，先设定g2o
  typedef g2o::BlockSolverX BlockSolverType;
  typedef g2o::LinearSolverDense<BlockSolverType::PoseMatrixType> LinearSolverType; // 线性求解器类型
  // 梯度下降方法，可以从GN, LM, DogLeg 中选
  auto solver = new g2o::OptimizationAlgorithmLevenberg(
  g2o::make_unique<BlockSolverType>(g2o::make_unique<LinearSolverType>()));
  g2o::SparseOptimizer optimizer;     // 图模型
  optimizer.setAlgorithm(solver);   // 设置求解器
  optimizer.setVerbose(true);       // 打开调试输出
 
  // vertex
  VertexPose *pose = new VertexPose(); // camera pose R t 
  pose->setId(0);
  pose->setEstimate(Sophus::SE3d());
  optimizer.addVertex(pose);
 
  // edges
  for (size_t i = 0; i < pts1.size(); i++) {
    EdgeProjectXYZRGBDPoseOnly *edge = new EdgeProjectXYZRGBDPoseOnly(Eigen::Vector3d(pts2[i].x, pts2[i].y, pts2[i].z));
    edge->setVertex(0, pose);
    edge->setMeasurement(Eigen::Vector3d(pts1[i].x, pts1[i].y, pts1[i].z));
    edge->setInformation(Eigen::Matrix3d::Identity());
    optimizer.addEdge(edge);
  }
 
  chrono::steady_clock::time_point t1 = chrono::steady_clock::now();
  optimizer.initializeOptimization();
  optimizer.optimize(10);
  chrono::steady_clock::time_point t2 = chrono::steady_clock::now();
  chrono::duration<double> time_used = chrono::duration_cast<chrono::duration<double>>(t2 - t1);
  cout << "optimization costs time: " << time_used.count() << " seconds." << endl;
 
  cout << endl << "after optimization:" << endl;
  cout << "T=\n" << pose->estimate().matrix() << endl;
=
  // convert to cv::Mat
  Eigen::Matrix3d R_ = pose->estimate().rotationMatrix();
  Eigen::Vector3d t_ = pose->estimate().translation();
  R = (Mat_<double>(3, 3) <<
    R_(0, 0), R_(0, 1), R_(0, 2),
    R_(1, 0), R_(1, 1), R_(1, 2),
    R_(2, 0), R_(2, 1), R_(2, 2)
  );
  t = (Mat_<double>(3, 1) << t_(0, 0), t_(1, 0), t_(2, 0));
}