d455 Gets the point cloud
Hi，I used realsense.viewer to get the rgb and depth maps to get the point cloud data. But it failed, and the effect of the point cloud looked like it was stretching out from one point, as shown below. i wondering why that was，and hope to find a case to learn from.
Thanks
-
Thanks very much for the RGB image. It looks as though that image and the 2D mode's depth image are okay but the problem is occurring in 3D mode when depth and RGB are mapped together to produce a pointcloud. It appears that the curtains behind the screen are being drawn in front of the screen.
Does the problem still occur if you disable the Viewer's GLSL settings using instructions at the link below, please?
https://github.com/IntelRealSense/librealsense/issues/8110#issuecomment-754705023
-
I followed the instructions in the link to disable GLSL, but the problem persists.
And the code I'm using is the open3d teaching case, the camera matrix uses a csv file on save.
color_raw = o3d.io.read_image('./test_Color.png')
depth_raw = o3d.io.read_image('./test_Depth.png')
rgbd_image = o3d.geometry.RGBDImage.create_from_color_and_depth(
color_raw, depth_raw)
pcd = o3d.geometry.PointCloud.create_from_rgbd_image(
rgbd_image,
o3d.camera.PinholeCameraIntrinsic(
width=848, height=480, fx=386.578705, fy=386.578705, cx=318.531158, cy=244.414246))
pcd.transform([[1, 0, 0, 0], [0, -1, 0, 0], [0, 0, -1, 0], [0, 0, 0, 1]])
o3d.visualization.draw_geometries([pcd]) -
Are the fx, fy, cx and cy values in the script taken from a csv file exported from the Viewer using the Snapshot button on the depth stream? The inside of the Viewer's csv file looks like the image below (the values are unique for each individual RealSense camera).
It may be helpful to look at a RealSense Open3D pointcloud script by another RealSense user at the link below that is also generating a pointcloud from depth and color PNGs.
https://github.com/IntelRealSense/librealsense/issues/11960#issuecomment-1763474993
-
Yes, my parameters are all taken from the viewer exported csv file and I did notice the difference in the camera.
At the moment, I'm wondering if there's something wrong with the way I'm saving the images from realsense.viewer? Because the depth maps and rgb maps I get from pyrealsense2 are perfectly fine to get the point cloud, while the ones I save from realsense.viewer still have the same problem, and they all use the same camera matrix, the same 640x480 camera and the same code.The way I get the depth map and RGB map from realsense.viewer is to pause the screen in 3D mode and then go back to 2D and click save snapshot on the RGB window and depth window.
I've pretty much solved my problem, it was a minor query I had, thank you very much for your help. -
You are very welcome!
The way that you are saving the images in the RealSense Viewer is fine. The Viewer applies a range of post-processing filters and depth colorization settings by default, whilst a depth image generated by a Python script will not have these settings unless they have been deliberately programmed into the script. So there can be noticeable differences between a Viewer image and a Python one unless filters and colorization are incorporated into the script to bring the image closer to the Viewer's.
Please sign in to leave a comment.
Comments
11 comments