| 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 SmartProjectionFactorExample.cpp |
| 14 | * @brief A stereo visual odometry example |
| 15 | * @date May 30, 2014 |
| 16 | * @author Stephen Camp |
| 17 | * @author Chris Beall |
| 18 | */ |
| 19 | |
| 20 | |
| 21 | /** |
| 22 | * A smart projection factor example based on stereo data, throwing away the |
| 23 | * measurement from the right camera |
| 24 | * -robot starts at origin |
| 25 | * -moves forward, taking periodic stereo measurements |
| 26 | * -makes monocular observations of many landmarks |
| 27 | */ |
| 28 | |
| 29 | #include <gtsam/slam/SmartProjectionPoseFactor.h> |
| 30 | #include <gtsam/slam/dataset.h> |
| 31 | #include <gtsam/geometry/Cal3_S2Stereo.h> |
| 32 | #include <gtsam/nonlinear/Values.h> |
| 33 | #include <gtsam/nonlinear/utilities.h> |
| 34 | #include <gtsam/nonlinear/NonlinearEquality.h> |
| 35 | #include <gtsam/nonlinear/NonlinearFactorGraph.h> |
| 36 | #include <gtsam/nonlinear/LevenbergMarquardtOptimizer.h> |
| 37 | #include <gtsam/inference/Symbol.h> |
| 38 | |
| 39 | #include <string> |
| 40 | #include <fstream> |
| 41 | #include <iostream> |
| 42 | |
| 43 | using namespace std; |
| 44 | using namespace gtsam; |
| 45 | |
| 46 | int main(int argc, char** argv){ |
| 47 | typedef SmartProjectionPoseFactor<Cal3_S2> SmartFactor; |
| 48 | |
| 49 | Values initial_estimate; |
| 50 | NonlinearFactorGraph graph; |
| 51 | const noiseModel::Isotropic::shared_ptr model = noiseModel::Isotropic::Sigma(dim: 2,sigma: 1); |
| 52 | |
| 53 | string calibration_loc = findExampleDataFile(name: "VO_calibration.txt" ); |
| 54 | string pose_loc = findExampleDataFile(name: "VO_camera_poses_large.txt" ); |
| 55 | string factor_loc = findExampleDataFile(name: "VO_stereo_factors_large.txt" ); |
| 56 | |
| 57 | //read camera calibration info from file |
| 58 | // focal lengths fx, fy, skew s, principal point u0, v0, baseline b |
| 59 | cout << "Reading calibration info" << endl; |
| 60 | ifstream calibration_file(calibration_loc.c_str()); |
| 61 | |
| 62 | double fx, fy, s, u0, v0, b; |
| 63 | calibration_file >> fx >> fy >> s >> u0 >> v0 >> b; |
| 64 | const Cal3_S2::shared_ptr K(new Cal3_S2(fx, fy, s, u0, v0)); |
| 65 | |
| 66 | cout << "Reading camera poses" << endl; |
| 67 | ifstream pose_file(pose_loc.c_str()); |
| 68 | |
| 69 | int pose_index; |
| 70 | MatrixRowMajor m(4,4); |
| 71 | //read camera pose parameters and use to make initial estimates of camera poses |
| 72 | while (pose_file >> pose_index) { |
| 73 | for (int i = 0; i < 16; i++) |
| 74 | pose_file >> m.data()[i]; |
| 75 | initial_estimate.insert(j: pose_index, val: Pose3(m)); |
| 76 | } |
| 77 | |
| 78 | // landmark keys |
| 79 | size_t landmark_key; |
| 80 | |
| 81 | // pixel coordinates uL, uR, v (same for left/right images due to rectification) |
| 82 | // landmark coordinates X, Y, Z in camera frame, resulting from triangulation |
| 83 | double uL, uR, v, X, Y, Z; |
| 84 | ifstream factor_file(factor_loc.c_str()); |
| 85 | cout << "Reading stereo factors" << endl; |
| 86 | |
| 87 | //read stereo measurements and construct smart factors |
| 88 | |
| 89 | SmartFactor::shared_ptr factor(new SmartFactor(model, K)); |
| 90 | size_t current_l = 3; // hardcoded landmark ID from first measurement |
| 91 | |
| 92 | while (factor_file >> pose_index >> landmark_key >> uL >> uR >> v >> X >> Y >> Z) { |
| 93 | |
| 94 | if(current_l != landmark_key) { |
| 95 | graph.push_back(factor); |
| 96 | factor = SmartFactor::shared_ptr(new SmartFactor(model, K)); |
| 97 | current_l = landmark_key; |
| 98 | } |
| 99 | factor->add(measured: Point2(uL,v), key: pose_index); |
| 100 | } |
| 101 | |
| 102 | Pose3 firstPose = initial_estimate.at<Pose3>(j: 1); |
| 103 | //constrain the first pose such that it cannot change from its original value during optimization |
| 104 | // NOTE: NonlinearEquality forces the optimizer to use QR rather than Cholesky |
| 105 | // QR is much slower than Cholesky, but numerically more stable |
| 106 | graph.emplace_shared<NonlinearEquality<Pose3> >(args: 1,args&: firstPose); |
| 107 | |
| 108 | LevenbergMarquardtParams params; |
| 109 | params.verbosityLM = LevenbergMarquardtParams::TRYLAMBDA; |
| 110 | params.verbosity = NonlinearOptimizerParams::ERROR; |
| 111 | |
| 112 | cout << "Optimizing" << endl; |
| 113 | //create Levenberg-Marquardt optimizer to optimize the factor graph |
| 114 | LevenbergMarquardtOptimizer optimizer(graph, initial_estimate, params); |
| 115 | Values result = optimizer.optimize(); |
| 116 | |
| 117 | cout << "Final result sample:" << endl; |
| 118 | Values pose_values = utilities::allPose3s(values: result); |
| 119 | pose_values.print(str: "Final camera poses:\n" ); |
| 120 | |
| 121 | return 0; |
| 122 | } |
| 123 | |