本文主要是介绍ROS实现可视化点云关键点(iss),希望对大家解决编程问题提供一定的参考价值,需要的开发者们随着小编来一起学习吧!
目录
- 可视化关键点
- 完整代码
可视化关键点
keypoint_core.h
//创建一了类 进行欧式聚类#ifndef __KEYPOINT_CORE__
#define __KEYPOINT_CORE__#include <iostream>
#include <vector>
#include <math.h>#include <ros/ros.h>
#include <sensor_msgs/PointCloud2.h>
#include <pcl_conversions/pcl_conversions.h>
#include <pcl_ros/point_cloud.h>
#include <pcl_ros/transforms.h>
#include <pcl/point_cloud.h> // make_Shared()
#include <pcl/point_types.h>
#include <pcl/conversions.h>
#include <pcl/kdtree/kdtree.h>//kd树搜索算法
#include <pcl/search/organized.h>
#include <pcl/search/kdtree.h>
#include <time.h>
#include <pcl/keypoints/iss_3d.h> // 关键点
#include <pcl/filters/voxel_grid.h>
#include <pcl/search/kdtree.h>#include <pcl/filters/voxel_grid.h> // 下采样#include <std_msgs/Header.h> using pcl::NormalEstimation;
using pcl::search::KdTree;
typedef pcl::PointXYZ PointT;
typedef pcl::PointCloud<PointT> PointCloud;
using namespace std;class Keypoint_core
{
private:/* data */ros::Subscriber sub_point_cloud_;ros::Publisher pub_keypoints_;// 降采样的leaf_sizedouble leaf_size = 0.3;// iss特征计算的邻域double iss_size = 0.3;void point_cb(const sensor_msgs::PointCloud2ConstPtr& in_cloud);public:Keypoint_core(ros::NodeHandle &nh);~Keypoint_core();
};Keypoint_core::Keypoint_core(ros::NodeHandle &nh)
{ std::cout<<"-----------------keypoint_node start-----------------"<<std::endl;cout<<"leaf_size: "<<leaf_size<<", "<<"iss_size: "<<iss_size<<endl;sub_point_cloud_ = nh.subscribe("/rslidar_points",10, &Keypoint_core::point_cb, this);pub_keypoints_ = nh.advertise<sensor_msgs::PointCloud2>("/key_points", 10);ros::spin();}Keypoint_core::~Keypoint_core()
{
}void Keypoint_core::point_cb(const sensor_msgs::PointCloud2ConstPtr& in_cloud_ptr)
{pcl::PointCloud<pcl::PointXYZ>::Ptr current_pc_ptr(new pcl::PointCloud<pcl::PointXYZ>);pcl::fromROSMsg(*in_cloud_ptr, *current_pc_ptr);clock_t start = clock();// 下采样PointCloud::Ptr cloud_src_out(new PointCloud);pcl::VoxelGrid<pcl::PointXYZ> filter;filter.setInputCloud(current_pc_ptr);filter.setLeafSize(leaf_size,leaf_size,leaf_size);filter.filter(*cloud_src_out);//issPointCloud::Ptr cloud_src_is(new PointCloud);pcl::ISSKeypoint3D<pcl::PointXYZ, pcl::PointXYZ> iss_det;pcl::search::KdTree<pcl::PointXYZ>::Ptr tree_1(new pcl::search::KdTree<pcl::PointXYZ>());double model_solution = 0.2;//iss参数设置iss_det.setSearchMethod(tree_1);iss_det.setSalientRadius(iss_size); // 0.5iss_det.setNonMaxRadius(0.5);iss_det.setThreshold21(0.975);iss_det.setThreshold32(0.975);iss_det.setMinNeighbors(5);iss_det.setNumberOfThreads(4);iss_det.setInputCloud(cloud_src_out);iss_det.compute(*cloud_src_is);clock_t end = clock();cout << "iss关键点提取时间:" << (double)(end - start) / CLOCKS_PER_SEC <<endl;cout << "iss关键点数量" << cloud_src_is->size() << endl;PointCloud::Ptr cloud_key(new PointCloud);pcl::copyPointCloud(*cloud_src_is, *cloud_key);sensor_msgs::PointCloud2 pub_pc;pcl::toROSMsg(*cloud_key, pub_pc);pub_pc.header = in_cloud_ptr->header;pub_keypoints_.publish(pub_pc);}#endif
keypoint_node.cpp
#include "keypoint_core.h"int main(int argc, char *argv[])
{ros::init(argc, argv, "keypoint_node"); // 节点名称 launch中的 type="aiimooc_syz4_node"是可执行文件名称ros::NodeHandle nh;// 创建对象Keypoint_core core(nh);return 0;
}
完整代码
这篇关于ROS实现可视化点云关键点(iss)的文章就介绍到这儿,希望我们推荐的文章对编程师们有所帮助!