基于图片的三维重建深度图与原图

                      基于图片的三维重建深度图与原图

点云重建基本流程

代码

#include
#include
#include
#include
#include

using namespace cv;
using namespace std;
using namespace pcl;

int user_data;
//相机内参，根据输入改动
const double u0 = 1329.49 / 4;//由于后面resize成原图的1/4所以有些参数要缩小相同倍数
const double v0 = 954.485 / 4;
const double fx = 6872.874 / 4;
const double fy = 6872.874 / 4;
const double Tx = 174.724;
const double doffs = 293.97 / 4;

void viewerOneOff(visualization::PCLVisualizer& viewer)
{
viewer.setBackgroundColor(0.0, 0.0, 0.0);
}

int main()
{
PointCloud cloud_a;
PointCloud::Ptr cloud(new PointCloud);

Mat color1 = imread("im0.png");
Mat depth = imread("Sword1_perfect_d.png");
////Resize
//color1.resize();
Mat color;
resize(color1, color, Size(color1.cols / 4, color1.rows / 4), 0, 0, CV_INTER_LINEAR);
//imshow("h",color);
//waitKey(0);

int rowNumber = color.rows;
int colNumber = color.cols;

cloud_a.height = rowNumber;
cloud_a.width = colNumber;
cloud_a.points.resize(cloud_a.width * cloud_a.height);

for (unsigned int u = 0; u < rowNumber; ++u)
{
	for (unsigned int v = 0; v < colNumber; ++v)
	{
		/*unsigned int num = rowNumber*colNumber-(u*colNumber + v)-1;*/
		unsigned int num = u*colNumber + v;
		double Xw = 0, Yw = 0, Zw = 0;


		Zw = fx*Tx / (((double)depth.at(u, v)[0]) + doffs);
		Xw = (v + 1 - u0) * Zw / fx;
		Yw = (u + 1 - v0) * Zw / fy;

		cloud_a.points[num].b = color.at(u, v)[0];
		cloud_a.points[num].g = color.at(u, v)[1];
		cloud_a.points[num].r = color.at(u, v)[2];

		cloud_a.points[num].x = Xw;
		cloud_a.points[num].y = Yw;
		cloud_a.points[num].z = Zw;
	}
}

*cloud = cloud_a;

visualization::CloudViewer viewer("Cloud Viewer");

viewer.showCloud(cloud);

viewer.runOnVisualizationThreadOnce(viewerOneOff);

while (!viewer.wasStopped())
{
	user_data = 9;
}

return 0;

}
原图“im0.png”可以http://vision.middlebury.edu/stereo/data/scenes2014/datasets/Sword1-perfect/im0.png 下载
深度图：

重建效果图：