Open3D从RGB图与depth图产生RGB |
您所在的位置：网站首页 › 灰度图转3d模型 › Open3D从RGB图与depth图产生RGB |
Open3D从RGB图与depth图产生RGB
|
Open3D
Open3D是一个开源库，它支持处理3D数据的软件的快速开发。Open3D前端在C++和Python中公开了一组精心选择的数据结构和算法。后端经过高度优化，并设置为并行化。 前期也用了一段时间的PCL，但是C++真的啃不动，转而寻求python版的点云处理库，就找到了Open3D。Open3D国内的相关资源还是比较少的，因此我准备在学习过程中，在CSDN上记录遇到的问题以及解决方法，方便后续查找。 下面代码可以利用Open3D生成RGB-D数据 import open3d as o3d import numpy as np from matplotlib import pyplot as plt color_raw = o3d.io.read_image("F:/Depth_image_C++/open_and_save/open_and_save/3Dprintcarrot/rgb/(1).png") depth_raw = o3d.io.read_image("F:/Depth_image_C++/open_and_save/open_and_save/3Dprintcarrot/depth/(1).png") rgbd_image = o3d.geometry.RGBDImage.create_from_color_and_depth(color_raw, depth_raw, convert_rgb_to_intensity=False) print(rgbd_image) plt.subplot(1, 2, 1) plt.title('Redwood grayscale image') plt.imshow(rgbd_image.color) plt.subplot(1, 2, 2) plt.title('Redwood depth image') plt.imshow(rgbd_image.depth) plt.show() inter = o3d.camera.PinholeCameraIntrinsic() inter.set_intrinsics(1280, 720, 599.795593, 599.633118, 645.792786, 372.238983) pcd = o3d.geometry.PointCloud().create_from_rgbd_image( rgbd_image, inter) # Flip it, otherwise the pointcloud will be upside down pcd.transform([[1, 0, 0, 0], [0, -1, 0, 0], [0, 0, -1, 0], [0, 0, 0, 1]]) o3d.visualization.draw_geometries([pcd]) 遇到的问题因为我使用的是Azure kinect SDK获取的rgb图与深度图，所以相机的内参需要自己设定下。 内参获取我看相关资料可以在Azure-Kinect-Sensor-SDK/examples/calibration at develop · microsoft/Azure-Kinect-Sensor-SDK · 找到。只需在Visual Studio 2017中加载CmakeList.txt并进行构建，将生成Calibration_info.exe。连接并执行Azure Kinect后，输出如下。 C:\GitHub\Azure-Kinect-Sensor-SDK\build\Win-x64-Release-Ninja\bin>calibration_info.exe > out.txt下面是Azure kinect sdk的一些内参，供大家参考 图像尺寸为640×576的内参，格式为K4A_DEPTH_MODE_NFOV_UNBINNED Found 1 connected devices: ===== Device 0: {DeviceID:XXXXXX} ===== resolution width: 640 resolution height: 576 principal point x: 309.301086 principal point y: 330.656738 focal length x: 505.119751 focal length y: 505.163605 radial distortion coefficients: k1: 3.063722 k2: 1.988396 k3: 0.102371 k4: 3.397454 k5: 2.984188 k6: 0.544097 center of distortion in Z=1 plane, x: 0.000000 center of distortion in Z=1 plane, y: 0.000000 tangential distortion coefficient x: -0.000098 tangential distortion coefficient y: 0.000048 metric radius: 0.000000也可以进行自动校准 auto calib = calibration.depth_camera_calibration和 auto calib = calibration.color_camera_calibration这是1080p的内参 Found 1 connected devices: ===== Device 0: {DeviceID:XXXXXX} ===== resolution width: 1920 resolution height: 1080 principal point x: 968.939209 principal point y: 558.608459 focal length x: 899.693420 focal length y: 899.449646 radial distortion coefficients: k1: 0.678679 k2: -2.779486 k3: 1.569404 k4: 0.554819 k5: -2.610379 k6: 1.500811 center of distortion in Z=1 plane, x: 0.000000 center of distortion in Z=1 plane, y: 0.000000 tangential distortion coefficient x: 0.000476 tangential distortion coefficient y: 0.000104 metric radius: 0.000000这是720p内参 Found 1 connected devices: ===== Device 0: {DeviceID:XXXXXX} ===== resolution width: 1280 resolution height: 720 principal point x: 645.792786 principal point y: 372.238983 focal length x: 599.795593 focal length y: 599.633118 radial distortion coefficients: k1: 0.678679 k2: -2.779486 k3: 1.569404 k4: 0.554819 k5: -2.610379 k6: 1.500811 center of distortion in Z=1 plane, x: 0.000000 center of distortion in Z=1 plane, y: 0.000000 tangential distortion coefficient x: 0.000476 tangential distortion coefficient y: 0.000104 metric radius: 0.000000 有了内参，怎么用Open3D读取内参并用于点云生成中有两种方法 第一种 inter = o3d.camera.PinholeCameraIntrinsic() inter.set_intrinsics(1280, 720, 599.795593, 599.633118, 645.792786, 372.238983)这里要注意的是inter = o3d.camera.PinholeCameraIntrinsic()，括号一定不能少，否则会报错 inter.set_intrinsics(1280, 720, 599.795593, 599.633118, 645.792786, 372.238983) TypeError: set_intrinsics(): incompatible function arguments. The following argument types are supported: 1. (self: open3d.cpu.pybind.camera.PinholeCameraIntrinsic, width: int, height: int, fx: float, fy: float, cx: float, cy: float) -> None Invoked with: 1280, 720, 599.795593, 599.633118, 645.792786, 372.238983第二种 inter= o3d.camera.PinholeCameraIntrinsic(1280, 720, 599.795593, 599.633118, 645.792786, 372.238983) 保存、读取内参 o3d.io.write_pinhole_camera_intrinsic('./file.json', inter) pinholeCamera = o3d.io.read_pinhole_camera_intrinsic('./file.json') 最后放几张处理过程中的图片
|
最后的点云图有点失真，可能是因为我的彩色图是RGBA四通道图，转成RGB三通道图就可以正常显示了【本文地址】
今日新闻 |
推荐新闻 |