initial LOST implementation tested

release/4.3a0
Akshay Krishnan 2022-06-17 10:57:51 -07:00
parent 784f16fe75
commit c49ad326d1
3 changed files with 109 additions and 0 deletions

View File

@ -96,6 +96,24 @@ TEST(triangulation, twoPoses) {
EXPECT(assert_equal(Point3(4.995, 0.499167, 1.19814), *actual4, 1e-4)); EXPECT(assert_equal(Point3(4.995, 0.499167, 1.19814), *actual4, 1e-4));
} }
TEST(triangulation, twoCamerasLOST) {
std::vector<PinholeCamera<Cal3_S2>> cameras = {camera1, camera2};
std::vector<Point2> measurements = {z1, z2};
// 1. Test simple triangulation, perfect in no noise situation
Point3 actual1 = //
triangulateLOSTPoint3<Cal3_S2>(cameras, measurements);
EXPECT(assert_equal(landmark, actual1, 1e-12));
// 3. Add some noise and try again: result should be ~ (4.995,
// 0.499167, 1.19814)
measurements[0] += Point2(0.1, 0.5);
measurements[1] += Point2(-0.2, 0.3);
Point3 actual2 = //
triangulateLOSTPoint3<Cal3_S2>(cameras, measurements);
EXPECT(assert_equal(Point3(4.995, 0.499167, 1.19814), actual2, 1e-4));
}
//****************************************************************************** //******************************************************************************
// Simple test with a well-behaved two camera situation with Cal3DS2 calibration. // Simple test with a well-behaved two camera situation with Cal3DS2 calibration.
TEST(triangulation, twoPosesCal3DS2) { TEST(triangulation, twoPosesCal3DS2) {

View File

@ -39,6 +39,11 @@ Vector4 triangulateHomogeneousDLT(
const Point2& p = measurements.at(i); const Point2& p = measurements.at(i);
// build system of equations // build system of equations
// [A_1; A_2; A_3] x = [b_1; b_2; b_3]
// [b_3 * A_1 - b_1 * A_3] x = 0
// [b_3 * A_2 - b_2 * A_3] x = 0
// A' x = 0
// A' 2x4 = [b_3 * A_1 - b_1 * A_3; b_3 * A_2 - b_2 * A_3]
A.row(row) = p.x() * projection.row(2) - projection.row(0); A.row(row) = p.x() * projection.row(2) - projection.row(0);
A.row(row + 1) = p.y() * projection.row(2) - projection.row(1); A.row(row + 1) = p.y() * projection.row(2) - projection.row(1);
} }
@ -53,6 +58,47 @@ Vector4 triangulateHomogeneousDLT(
return v; return v;
} }
Vector3 triangulateLOSTHomogeneous(
const std::vector<Pose3>& poses,
const std::vector<Point3>& calibrated_measurements) {
// TODO(akshay-krishnan): make this an argument.
const double sigma_x = 1e-3;
size_t m = calibrated_measurements.size();
assert(m == poses.size());
// Construct the system matrices.
Matrix A = Matrix::Zero(m * 2, 3);
Matrix b = Matrix::Zero(m * 2, 1);
for (size_t i = 0; i < m; i++) {
const Pose3& wTi = poses[i];
// TODO(akshay-krishnan): are there better ways to select j?
const int j = (i + 1) % m;
const Pose3& wTj = poses[j];
Point3 d_ij = wTj.translation() - wTi.translation();
Point3 w_measurement_i = wTi.rotation().rotate(calibrated_measurements[i]);
Point3 w_measurement_j = wTj.rotation().rotate(calibrated_measurements[j]);
double numerator = w_measurement_i.cross(w_measurement_j).norm();
double denominator = d_ij.cross(w_measurement_j).norm();
double q_i = numerator / (sigma_x * denominator);
Matrix23 coefficient_mat = q_i * skewSymmetric(calibrated_measurements[i]).topLeftCorner(2, 3) * wTi.rotation().matrix().transpose();
A.row(2 * i) = coefficient_mat.row(0);
A.row(2 * i + 1) = coefficient_mat.row(1);
Point2 p = coefficient_mat * wTi.translation();
b(2 * i) = p.x();
b(2 * i + 1) = p.y();
}
// Solve Ax = b, using QR decomposition
return A.colPivHouseholderQr().solve(b);
}
Vector4 triangulateHomogeneousDLT( Vector4 triangulateHomogeneousDLT(
const std::vector<Matrix34, Eigen::aligned_allocator<Matrix34>>& projection_matrices, const std::vector<Matrix34, Eigen::aligned_allocator<Matrix34>>& projection_matrices,
const std::vector<Unit3>& measurements, double rank_tol) { const std::vector<Unit3>& measurements, double rank_tol) {

View File

@ -62,6 +62,18 @@ GTSAM_EXPORT Vector4 triangulateHomogeneousDLT(
const std::vector<Matrix34, Eigen::aligned_allocator<Matrix34>>& projection_matrices, const std::vector<Matrix34, Eigen::aligned_allocator<Matrix34>>& projection_matrices,
const Point2Vector& measurements, double rank_tol = 1e-9); const Point2Vector& measurements, double rank_tol = 1e-9);
/**
* @brief
*
* @param projection_matrices
* @param measurements
* @param rank_tol
* @return GTSAM_EXPORT
*/
GTSAM_EXPORT Vector3
triangulateLOSTHomogeneous(const std::vector<Pose3>& poses,
const std::vector<Point3>& calibrated_measurements);
/** /**
* Same math as Hartley and Zisserman, 2nd Ed., page 312, but with unit-norm bearing vectors * Same math as Hartley and Zisserman, 2nd Ed., page 312, but with unit-norm bearing vectors
* (contrarily to pinhole projection, the z entry is not assumed to be 1 as in Hartley and Zisserman) * (contrarily to pinhole projection, the z entry is not assumed to be 1 as in Hartley and Zisserman)
@ -382,6 +394,39 @@ Point3 triangulatePoint3(const std::vector<Pose3>& poses,
return point; return point;
} }
template <class CALIBRATION>
Point3 triangulateLOSTPoint3(const std::vector<PinholeCamera<CALIBRATION>>& cameras,
const std::vector<Point2>& measurements) {
const size_t num_cameras = cameras.size();
assert(measurements.size() == num_cameras);
if (num_cameras < 2) throw(TriangulationUnderconstrainedException());
// Convert measurements to image plane coordinates.
std::vector<Point3> calibrated_measurements;
calibrated_measurements.reserve(measurements.size());
for (int i = 0; i < measurements.size(); ++i) {
Point2 p = cameras[i].calibration().calibrate(measurements[i]);
calibrated_measurements.emplace_back(p.x(), p.y(), 1.0);
}
std::vector<Pose3> poses;
poses.reserve(cameras.size());
for (const auto& camera : cameras) poses.push_back(camera.pose());
Point3 point = triangulateLOSTHomogeneous(poses, calibrated_measurements);
#ifdef GTSAM_THROW_CHEIRALITY_EXCEPTION
// verify that the triangulated point lies in front of all cameras
for (const auto& camera : cameras) {
const Point3& p_local = camera.pose().transformTo(point);
if (p_local.z() <= 0) throw(TriangulationCheiralityException());
}
#endif
return point;
}
/** /**
* Function to triangulate 3D landmark point from an arbitrary number * Function to triangulate 3D landmark point from an arbitrary number
* of poses (at least 2) using the DLT. This function is similar to the one * of poses (at least 2) using the DLT. This function is similar to the one