diff --git a/examples/VisualISAM2Example.cpp b/examples/VisualISAM2Example.cpp index 157768be7..751b776f6 100644 --- a/examples/VisualISAM2Example.cpp +++ b/examples/VisualISAM2Example.cpp @@ -11,8 +11,8 @@ /** * @file VisualISAM2Example.cpp - * @brief A visualSLAM example for the structure-from-motion problem on a simulated dataset - * This version uses iSAM2 to solve the problem incrementally + * @brief A visualSLAM example for the structure-from-motion problem on a + * simulated dataset This version uses iSAM2 to solve the problem incrementally * @author Duy-Nguyen Ta */ @@ -25,27 +25,28 @@ // For loading the data #include "SFMdata.h" -// Camera observations of landmarks (i.e. pixel coordinates) will be stored as Point2 (x, y). +// Camera observations of landmarks will be stored as Point2 (x, y). #include -// Each variable in the system (poses and landmarks) must be identified with a unique key. -// We can either use simple integer keys (1, 2, 3, ...) or symbols (X1, X2, L1). -// Here we will use Symbols +// Each variable in the system (poses and landmarks) must be identified with a +// unique key. We can either use simple integer keys (1, 2, 3, ...) or symbols +// (X1, X2, L1). Here we will use Symbols #include -// We want to use iSAM2 to solve the structure-from-motion problem incrementally, so -// include iSAM2 here +// We want to use iSAM2 to solve the structure-from-motion problem +// incrementally, so include iSAM2 here #include -// iSAM2 requires as input a set set of new factors to be added stored in a factor graph, -// and initial guesses for any new variables used in the added factors +// iSAM2 requires as input a set of new factors to be added stored in a factor +// graph, and initial guesses for any new variables used in the added factors #include #include -// In GTSAM, measurement functions are represented as 'factors'. Several common factors -// have been provided with the library for solving robotics/SLAM/Bundle Adjustment problems. -// Here we will use Projection factors to model the camera's landmark observations. -// Also, we will initialize the robot at some location using a Prior factor. +// In GTSAM, measurement functions are represented as 'factors'. Several common +// factors have been provided with the library for solving robotics/SLAM/Bundle +// Adjustment problems. Here we will use Projection factors to model the +// camera's landmark observations. Also, we will initialize the robot at some +// location using a Prior factor. #include #include @@ -56,12 +57,11 @@ using namespace gtsam; /* ************************************************************************* */ int main(int argc, char* argv[]) { - // Define the camera calibration parameters Cal3_S2::shared_ptr K(new Cal3_S2(50.0, 50.0, 0.0, 50.0, 50.0)); - // Define the camera observation noise model - noiseModel::Isotropic::shared_ptr measurementNoise = noiseModel::Isotropic::Sigma(2, 1.0); // one pixel in u and v + // Define the camera observation noise model, 1 pixel stddev + auto measurementNoise = noiseModel::Isotropic::Sigma(2, 1.0); // Create the set of ground-truth landmarks vector points = createPoints(); @@ -69,10 +69,12 @@ int main(int argc, char* argv[]) { // Create the set of ground-truth poses vector poses = createPoses(); - // Create an iSAM2 object. Unlike iSAM1, which performs periodic batch steps to maintain proper linearization - // and efficient variable ordering, iSAM2 performs partial relinearization/reordering at each step. A parameter - // structure is available that allows the user to set various properties, such as the relinearization threshold - // and type of linear solver. For this example, we we set the relinearization threshold small so the iSAM2 result + // Create an iSAM2 object. Unlike iSAM1, which performs periodic batch steps + // to maintain proper linearization and efficient variable ordering, iSAM2 + // performs partial relinearization/reordering at each step. A parameter + // structure is available that allows the user to set various properties, such + // as the relinearization threshold and type of linear solver. For this + // example, we we set the relinearization threshold small so the iSAM2 result // will approach the batch result. ISAM2Params parameters; parameters.relinearizeThreshold = 0.01; @@ -83,44 +85,52 @@ int main(int argc, char* argv[]) { NonlinearFactorGraph graph; Values initialEstimate; - // Loop over the different poses, adding the observations to iSAM incrementally + // Loop over the poses, adding the observations to iSAM incrementally for (size_t i = 0; i < poses.size(); ++i) { - // Add factors for each landmark observation for (size_t j = 0; j < points.size(); ++j) { SimpleCamera camera(poses[i], *K); Point2 measurement = camera.project(points[j]); - graph.emplace_shared >(measurement, measurementNoise, Symbol('x', i), Symbol('l', j), K); + graph.emplace_shared >( + measurement, measurementNoise, Symbol('x', i), Symbol('l', j), K); } // Add an initial guess for the current pose // Intentionally initialize the variables off from the ground truth - initialEstimate.insert(Symbol('x', i), poses[i].compose(Pose3(Rot3::Rodrigues(-0.1, 0.2, 0.25), Point3(0.05, -0.10, 0.20)))); + static Pose3 kDeltaPose(Rot3::Rodrigues(-0.1, 0.2, 0.25), + Point3(0.05, -0.10, 0.20)); + initialEstimate.insert(Symbol('x', i), poses[i] * kDeltaPose); - // If this is the first iteration, add a prior on the first pose to set the coordinate frame - // and a prior on the first landmark to set the scale - // Also, as iSAM solves incrementally, we must wait until each is observed at least twice before - // adding it to iSAM. - if( i == 0) { - // Add a prior on pose x0 - noiseModel::Diagonal::shared_ptr poseNoise = noiseModel::Diagonal::Sigmas((Vector(6) << Vector3::Constant(0.3),Vector3::Constant(0.1)).finished()); // 30cm std on x,y,z 0.1 rad on roll,pitch,yaw - graph.emplace_shared >(Symbol('x', 0), poses[0], poseNoise); + // If this is the first iteration, add a prior on the first pose to set the + // coordinate frame and a prior on the first landmark to set the scale Also, + // as iSAM solves incrementally, we must wait until each is observed at + // least twice before adding it to iSAM. + if (i == 0) { + // Add a prior on pose x0, 30cm std on x,y,z and 0.1 rad on roll,pitch,yaw + static auto kPosePrior = noiseModel::Diagonal::Sigmas( + (Vector(6) << Vector3::Constant(0.3), Vector3::Constant(0.1)) + .finished()); + graph.emplace_shared >(Symbol('x', 0), poses[0], + kPosePrior); // Add a prior on landmark l0 - noiseModel::Isotropic::shared_ptr pointNoise = noiseModel::Isotropic::Sigma(3, 0.1); - graph.emplace_shared >(Symbol('l', 0), points[0], pointNoise); // add directly to graph + static auto kPointPrior = noiseModel::Isotropic::Sigma(3, 0.1); + graph.emplace_shared >(Symbol('l', 0), points[0], + kPointPrior); // Add initial guesses to all observed landmarks // Intentionally initialize the variables off from the ground truth + static Point3 kDeltaPoint(-0.25, 0.20, 0.15); for (size_t j = 0; j < points.size(); ++j) - initialEstimate.insert(Symbol('l', j), points[j] +Point3(-0.25, 0.20, 0.15)); + initialEstimate.insert(Symbol('l', j), points[j] + kDeltaPoint); } else { // Update iSAM with the new factors isam.update(graph, initialEstimate); - // Each call to iSAM2 update(*) performs one iteration of the iterative nonlinear solver. - // If accuracy is desired at the expense of time, update(*) can be called additional times - // to perform multiple optimizer iterations every step. + // Each call to iSAM2 update(*) performs one iteration of the iterative + // nonlinear solver. If accuracy is desired at the expense of time, + // update(*) can be called additional times to perform multiple optimizer + // iterations every step. isam.update(); Values currentEstimate = isam.calculateEstimate(); cout << "****************************************************" << endl;