Go to the documentation of this file.00001
00002
00003
00004
00005
00006
00007
00008
00009
00010
00011
00012
00013
00014
00015
00016
00017
00018
00019
00020
00021
00022
00023
00024
00025
00026
00027
00028
00029
00030
00031
00032
00033
00034
00035
00036
00037
00038
00039
00040 #include <pcl/console/parse.h>
00041 #include <pcl/io/pcd_io.h>
00042 #include <pcl/point_types.h>
00043 #include <pcl/registration/icp.h>
00044 #include <pcl/registration/icp_nl.h>
00045
00046 #include <string>
00047 #include <iostream>
00048 #include <fstream>
00049 #include <vector>
00050
00051 typedef pcl::PointXYZ PointType;
00052 typedef pcl::PointCloud<PointType> Cloud;
00053 typedef Cloud::ConstPtr CloudConstPtr;
00054 typedef Cloud::Ptr CloudPtr;
00055
00056 int
00057 main (int argc, char **argv)
00058 {
00059 double dist = 0.05;
00060 pcl::console::parse_argument (argc, argv, "-d", dist);
00061
00062 double rans = 0.05;
00063 pcl::console::parse_argument (argc, argv, "-r", rans);
00064
00065 int iter = 50;
00066 pcl::console::parse_argument (argc, argv, "-i", iter);
00067
00068 bool nonLinear = false;
00069 pcl::console::parse_argument (argc, argv, "-n", nonLinear);
00070
00071 std::vector<int> pcd_indices;
00072 pcd_indices = pcl::console::parse_file_extension_argument (argc, argv, ".pcd");
00073
00074 CloudPtr model (new Cloud);
00075 if (pcl::io::loadPCDFile (argv[pcd_indices[0]], *model) == -1)
00076 {
00077 std::cout << "Could not read file" << std::endl;
00078 return -1;
00079 }
00080 std::cout << argv[pcd_indices[0]] << " width: " << model->width << " height: " << model->height << std::endl;
00081
00082 std::string result_filename (argv[pcd_indices[0]]);
00083 result_filename = result_filename.substr (result_filename.rfind ("/") + 1);
00084 pcl::io::savePCDFile (result_filename.c_str (), *model);
00085 std::cout << "saving first model to " << result_filename << std::endl;
00086
00087 Eigen::Matrix4f t (Eigen::Matrix4f::Identity ());
00088
00089 for (size_t i = 1; i < pcd_indices.size (); i++)
00090 {
00091 CloudPtr data (new Cloud);
00092 if (pcl::io::loadPCDFile (argv[pcd_indices[i]], *data) == -1)
00093 {
00094 std::cout << "Could not read file" << std::endl;
00095 return -1;
00096 }
00097 std::cout << argv[pcd_indices[i]] << " width: " << data->width << " height: " << data->height << std::endl;
00098
00099 pcl::IterativeClosestPoint<PointType, PointType> *icp;
00100
00101 if (nonLinear)
00102 {
00103 std::cout << "Using IterativeClosestPointNonLinear" << std::endl;
00104 icp = new pcl::IterativeClosestPointNonLinear<PointType, PointType>();
00105 }
00106 else
00107 {
00108 std::cout << "Using IterativeClosestPoint" << std::endl;
00109 icp = new pcl::IterativeClosestPoint<PointType, PointType>();
00110 }
00111
00112 icp->setMaximumIterations (iter);
00113 icp->setMaxCorrespondenceDistance (dist);
00114 icp->setRANSACOutlierRejectionThreshold (rans);
00115
00116 icp->setInputTarget (model);
00117
00118 icp->setInputSource (data);
00119
00120 CloudPtr tmp (new Cloud);
00121 icp->align (*tmp);
00122
00123 t = t * icp->getFinalTransformation ();
00124
00125 pcl::transformPointCloud (*data, *tmp, t);
00126
00127 std::cout << icp->getFinalTransformation () << std::endl;
00128
00129 *model = *data;
00130
00131 std::string result_filename (argv[pcd_indices[i]]);
00132 result_filename = result_filename.substr (result_filename.rfind ("/") + 1);
00133 pcl::io::savePCDFileBinary (result_filename.c_str (), *tmp);
00134 std::cout << "saving result to " << result_filename << std::endl;
00135 }
00136
00137 return 0;
00138 }