👑主页：吾名招财
👓简介：工科学硕，研究方向机器视觉，爱好较广泛…
​💫签名：面朝大海，春暖花开！

1，引言

  针对彩色图和深度图以及相机内参可以合成点云，而对某一区域的点云数据截取，可以通过二维ROI区域截取深度图及彩色图出来。不过不能整张图像裁剪，其图像大小不能变，把除了截取区域外的给置为0就行了。

2，相关测试数据资源如下

  本人上传了一个用于三维重建测试的公开数据集，内含彩色图、深度图、相机内参、相机位姿等相关数据，可用于相关测试
https://download.csdn.net/download/qq_44870829/90236553

3，彩色图、深度图和相机内参

4，C++代码

#include <string>
#include <iostream>
#include "Open3D/Open3D.h"
#include <opencv2/opencv.hpp>using namespace std;open3d::geometry::Image o3d_cv(const cv::Mat& A) {open3d::geometry::Image B;int bytes_per_channel = (A.depth() / 2 + 1);//refer to the fuction depthB.Prepare(A.cols, A.rows, A.channels(), bytes_per_channel);std::memcpy(B.data_.data(), A.data, A.total() * A.channels() * bytes_per_channel);return B;
}int main(int argc, char* argv[]) {//--------------------------------------1，相关参数----------------------------------//要截取的ROI区域大小int x = 285;int y = 363;int w = 150;int h = 120;//相机内参设置int width = 640;   // 输入图像的宽度int height = 480;  // 输入图像的高度double fx = 585; // x轴焦距 double fy = 585; // y轴焦距double cx = 320; // 相机原点的x坐标double cy = 240; // 相机原点的y坐标// 方式一auto intrinsic = open3d::camera::PinholeCameraIntrinsic(width, height, fx, fy, cx, cy); // 使用自定义内参方式二open3d::camera::PinholeCameraIntrinsic intrinsic = open3d::camera::PinholeCameraIntrinsic();intrinsic.SetIntrinsics(width, height, fx, fy, cx, cy);//open3d::camera::PinholeCameraIntrinsic intrinsic = open3d::camera::PinholeCameraIntrinsic(//	open3d::camera::PinholeCameraIntrinsicParameters::PrimeSenseDefault); // 使用默认内参// -RGBD图像参数设置double depth_scale = 1000.0; // 深度值的缩放倍数double depth_trunc = 3.0;    // 深度值的截断系数bool convert_rgb_to_intensity = false; // 是否将彩色图转为强度图//-------------------------------------2,读取图像并裁剪-------------------------------------//用opencv读取二维彩色图及深度图像，将某一ROI区域裁剪出来，然后再进行点云生成cv::Mat colorimg = cv::imread("frame-000276.color.jpg");cv::Mat depthimg = cv::imread("frame-000276.depth.png", -1);  //深度图要读取原图//制作ROI区域掩膜cv::Mat templateImg_color = cv::Mat::zeros(colorimg.size(), colorimg.type());cv::Mat mask = cv::Mat::zeros(colorimg.size(), colorimg.type());cv::rectangle(mask, cv::Point(x, y), cv::Point(x + w, y + h), cv::Scalar(255, 255, 255), -1);//矩形的两个顶点，两个顶点都包括在矩形内部colorimg.copyTo(templateImg_color, mask);cv::Mat templateImg_depth = cv::Mat::zeros(depthimg.size(), depthimg.type());cv::Mat mask2 = cv::Mat::zeros(depthimg.size(), CV_8UC1);cv::rectangle(mask2, cv::Point(x, y), cv::Point(x + w, y + h), cv::Scalar(255, 255, 255), -1);//矩形的两个顶点，两个顶点都包括在矩形内部//cv::bitwise_and();depthimg.copyTo(templateImg_depth, mask2);//-------------------------------------3，显示未裁剪前的原始点云-------------------------------------open3d::geometry::Image color_o = o3d_cv(colorimg);open3d::geometry::Image depth_o = o3d_cv(depthimg);// 生成RGBD图像std::shared_ptr<open3d::geometry::RGBDImage> rgbd_image_o = open3d::geometry::RGBDImage::CreateFromColorAndDepth(color_o,                    // 输入的彩色图像depth_o,                    // 输入的深度图像depth_scale,              // 深度值的缩放倍数depth_trunc,              // 深度值大于该值将被截断为0convert_rgb_to_intensity);// 设置是否将彩色图像转为强度图// RGBD转点云auto pcd_o = open3d::geometry::PointCloud::CreateFromRGBDImage(*rgbd_image_o, intrinsic);open3d::visualization::DrawGeometries({ pcd_o });//-------------------------------------4，显示并保存裁剪后的点云-------------------------------------open3d::geometry::Image color = o3d_cv(templateImg_color);open3d::geometry::Image depth = o3d_cv(templateImg_depth);直接使用open3d读取图像//open3d::geometry::Image color, depth;//open3d::io::ReadImage("0.png", color); // 读取彩色图像//open3d::io::ReadImage("0_depth.png", depth); // 读取深度图像//将裁剪后的深度图及彩色图转换成点云并显示保存//输出图像基本信息open3d::utility::LogInfo("Reading RGBD image : ");open3d::utility::LogInfo("     Color : {:d} x {:d} x {:d} ({:d} bits per channel)",color.width_, color.height_, color.num_of_channels_,color.bytes_per_channel_ * 8);open3d::utility::LogInfo("Depth : {:d} x {:d} x {:d} ({:d} bits per channel)",depth.width_, depth.height_, depth.num_of_channels_,depth.bytes_per_channel_ * 8);// 生成RGBD图像std::shared_ptr<open3d::geometry::RGBDImage> rgbd_image = open3d::geometry::RGBDImage::CreateFromColorAndDepth(color,                    // 输入的彩色图像depth,                    // 输入的深度图像depth_scale,              // 深度值的缩放倍数depth_trunc,              // 深度值大于该值将被截断为0convert_rgb_to_intensity);// 设置是否将彩色图像转为强度图// RGBD转点云auto pcd = open3d::geometry::PointCloud::CreateFromRGBDImage(*rgbd_image, intrinsic);open3d::visualization::DrawGeometries({ pcd });//5.保存点云文件 保存open3d::io::WritePointCloudToPCD("moban.pcd", *pcd, false);//open3d::io::WritePointCloudToPCD("search.pcd", *pcd, false);return 0;
}

5，最终效果

未裁剪前的点云

裁剪后的点云