| 1 | /* ---------------------------------------------------------------------------- |
| 2 | |
| 3 | * GTSAM Copyright 2010, Georgia Tech Research Corporation, |
| 4 | * Atlanta, Georgia 30332-0415 |
| 5 | * All Rights Reserved |
| 6 | * Authors: Frank Dellaert, et al. (see THANKS for the full author list) |
| 7 | |
| 8 | * See LICENSE for the license information |
| 9 | |
| 10 | * -------------------------------------------------------------------------- */ |
| 11 | |
| 12 | /** |
| 13 | * @file VisualISAM2Example.cpp |
| 14 | * @brief A visualSLAM example for the structure-from-motion problem on a |
| 15 | * simulated dataset This version uses iSAM2 to solve the problem incrementally |
| 16 | * @author Duy-Nguyen Ta |
| 17 | */ |
| 18 | |
| 19 | /** |
| 20 | * A structure-from-motion example with landmarks |
| 21 | * - The landmarks form a 10 meter cube |
| 22 | * - The robot rotates around the landmarks, always facing towards the cube |
| 23 | */ |
| 24 | |
| 25 | // For loading the data |
| 26 | #include "SFMdata.h" |
| 27 | |
| 28 | // Camera observations of landmarks will be stored as Point2 (x, y). |
| 29 | #include <gtsam/geometry/Point2.h> |
| 30 | |
| 31 | // Each variable in the system (poses and landmarks) must be identified with a |
| 32 | // unique key. We can either use simple integer keys (1, 2, 3, ...) or symbols |
| 33 | // (X1, X2, L1). Here we will use Symbols |
| 34 | #include <gtsam/inference/Symbol.h> |
| 35 | |
| 36 | // We want to use iSAM2 to solve the structure-from-motion problem |
| 37 | // incrementally, so include iSAM2 here |
| 38 | #include <gtsam/nonlinear/ISAM2.h> |
| 39 | |
| 40 | // iSAM2 requires as input a set of new factors to be added stored in a factor |
| 41 | // graph, and initial guesses for any new variables used in the added factors |
| 42 | #include <gtsam/nonlinear/NonlinearFactorGraph.h> |
| 43 | #include <gtsam/nonlinear/Values.h> |
| 44 | |
| 45 | // In GTSAM, measurement functions are represented as 'factors'. Several common |
| 46 | // factors have been provided with the library for solving robotics/SLAM/Bundle |
| 47 | // Adjustment problems. Here we will use Projection factors to model the |
| 48 | // camera's landmark observations. Also, we will initialize the robot at some |
| 49 | // location using a Prior factor. |
| 50 | #include <gtsam/slam/ProjectionFactor.h> |
| 51 | |
| 52 | #include <vector> |
| 53 | |
| 54 | using namespace std; |
| 55 | using namespace gtsam; |
| 56 | |
| 57 | /* ************************************************************************* */ |
| 58 | int main(int argc, char* argv[]) { |
| 59 | // Define the camera calibration parameters |
| 60 | Cal3_S2::shared_ptr K(new Cal3_S2(50.0, 50.0, 0.0, 50.0, 50.0)); |
| 61 | |
| 62 | // Define the camera observation noise model, 1 pixel stddev |
| 63 | auto measurementNoise = noiseModel::Isotropic::Sigma(dim: 2, sigma: 1.0); |
| 64 | |
| 65 | // Create the set of ground-truth landmarks |
| 66 | vector<Point3> points = createPoints(); |
| 67 | |
| 68 | // Create the set of ground-truth poses |
| 69 | vector<Pose3> poses = createPoses(); |
| 70 | |
| 71 | // Create an iSAM2 object. Unlike iSAM1, which performs periodic batch steps |
| 72 | // to maintain proper linearization and efficient variable ordering, iSAM2 |
| 73 | // performs partial relinearization/reordering at each step. A parameter |
| 74 | // structure is available that allows the user to set various properties, such |
| 75 | // as the relinearization threshold and type of linear solver. For this |
| 76 | // example, we we set the relinearization threshold small so the iSAM2 result |
| 77 | // will approach the batch result. |
| 78 | ISAM2Params parameters; |
| 79 | parameters.relinearizeThreshold = 0.01; |
| 80 | parameters.relinearizeSkip = 1; |
| 81 | ISAM2 isam(parameters); |
| 82 | |
| 83 | // Create a Factor Graph and Values to hold the new data |
| 84 | NonlinearFactorGraph graph; |
| 85 | Values initialEstimate; |
| 86 | |
| 87 | // Loop over the poses, adding the observations to iSAM incrementally |
| 88 | for (size_t i = 0; i < poses.size(); ++i) { |
| 89 | // Add factors for each landmark observation |
| 90 | for (size_t j = 0; j < points.size(); ++j) { |
| 91 | PinholeCamera<Cal3_S2> camera(poses[i], *K); |
| 92 | Point2 measurement = camera.project(pw: points[j]); |
| 93 | graph.emplace_shared<GenericProjectionFactor<Pose3, Point3, Cal3_S2> >( |
| 94 | args&: measurement, args&: measurementNoise, args: Symbol('x', i), args: Symbol('l', j), args&: K); |
| 95 | } |
| 96 | |
| 97 | // Add an initial guess for the current pose |
| 98 | // Intentionally initialize the variables off from the ground truth |
| 99 | static Pose3 kDeltaPose(Rot3::Rodrigues(wx: -0.1, wy: 0.2, wz: 0.25), |
| 100 | Point3(0.05, -0.10, 0.20)); |
| 101 | initialEstimate.insert(j: Symbol('x', i), val: poses[i] * kDeltaPose); |
| 102 | |
| 103 | // If this is the first iteration, add a prior on the first pose to set the |
| 104 | // coordinate frame and a prior on the first landmark to set the scale Also, |
| 105 | // as iSAM solves incrementally, we must wait until each is observed at |
| 106 | // least twice before adding it to iSAM. |
| 107 | if (i == 0) { |
| 108 | // Add a prior on pose x0, 30cm std on x,y,z and 0.1 rad on roll,pitch,yaw |
| 109 | static auto kPosePrior = noiseModel::Diagonal::Sigmas( |
| 110 | sigmas: (Vector(6) << Vector3::Constant(value: 0.1), Vector3::Constant(value: 0.3)) |
| 111 | .finished()); |
| 112 | graph.addPrior(key: Symbol('x', 0), prior: poses[0], model: kPosePrior); |
| 113 | |
| 114 | // Add a prior on landmark l0 |
| 115 | static auto kPointPrior = noiseModel::Isotropic::Sigma(dim: 3, sigma: 0.1); |
| 116 | graph.addPrior(key: Symbol('l', 0), prior: points[0], model: kPointPrior); |
| 117 | |
| 118 | // Add initial guesses to all observed landmarks |
| 119 | // Intentionally initialize the variables off from the ground truth |
| 120 | static Point3 kDeltaPoint(-0.25, 0.20, 0.15); |
| 121 | for (size_t j = 0; j < points.size(); ++j) |
| 122 | initialEstimate.insert<Point3>(j: Symbol('l', j), val: points[j] + kDeltaPoint); |
| 123 | |
| 124 | } else { |
| 125 | // Update iSAM with the new factors |
| 126 | isam.update(newFactors: graph, newTheta: initialEstimate); |
| 127 | // Each call to iSAM2 update(*) performs one iteration of the iterative |
| 128 | // nonlinear solver. If accuracy is desired at the expense of time, |
| 129 | // update(*) can be called additional times to perform multiple optimizer |
| 130 | // iterations every step. |
| 131 | isam.update(); |
| 132 | Values currentEstimate = isam.calculateEstimate(); |
| 133 | cout << "****************************************************" << endl; |
| 134 | cout << "Frame " << i << ": " << endl; |
| 135 | currentEstimate.print(str: "Current estimate: " ); |
| 136 | |
| 137 | // Clear the factor graph and values for the next iteration |
| 138 | graph.resize(size: 0); |
| 139 | initialEstimate.clear(); |
| 140 | } |
| 141 | } |
| 142 | |
| 143 | return 0; |
| 144 | } |
| 145 | /* ************************************************************************* */ |
| 146 | |