RGBD Images

Open3Dには画像用のデータ構造がある。 read_image、write_image、filter_image、draw_geometriesなどのさまざまな関数を用意している。 Open3Dイメージは、numpy配列に直接変換することができる。

Open3Dの RGBDImageは、RGBDImage.depthとRGBDImage.colorの2つの画像で構成されている。 ただし2つの画像を同じカメラフレームで位置合わせし、同じ解像度にする必要がある。 以下のチュートリアルでは、よく知られているいくつかのRGBDデータセットからRGBDイメージを読み込んで使用する方法を示す。

注意: コードを走らせる環境に注意すること。TestDataディレクトリがカレント(作業)ディレクトリからみて、祖先ディレクトリの下にあることを確認しよう


Redwood Dataset

ここでは Redwoodデータセット Choi et al.(2015)からGBDImageを読み込み、可視化する。

  • Choi, S., Zhou, Q.-Y., & V. Koltun. (2015). Robust Reconstruction of Indoor Scenes, CVPR.
In [15]:
# examples/Python/Basic/rgbd_redwood.py

import open3d as o3d
import matplotlib.pyplot as plt

if __name__ == "__main__":
    print("Read Redwood dataset")
    color_raw = o3d.io.read_image("../../TestData/RGBD/color/00000.jpg")
    depth_raw = o3d.io.read_image("../../TestData/RGBD/depth/00000.png")
    rgbd_image = o3d.geometry.RGBDImage.create_from_color_and_depth(
        color_raw, depth_raw)
    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)


    pcd = o3d.geometry.PointCloud.create_from_rgbd_image(
        rgbd_image,
        o3d.camera.PinholeCameraIntrinsic(
            o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault))
    # 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])
Read Redwood dataset
RGBDImage of size 
Color image : 640x480, with 1 channels.
Depth image : 640x480, with 1 channels.
Use numpy.asarray to access buffer data.

Redwoodフォーマットは、16ビットのシングルチャンネル画像に深度を格納している。 整数値はミリメートル単位の深度測定値を表す。 Open3Dが奥行き画像を解析するためのデフォルトのフォーマットである。

次は今挙げた examples/Python/Basic/rgbd_redwood.py の最初の部分コード:

In [ ]:
    print("Read Redwood dataset")
    color_raw = o3d.io.read_image("../../TestData/RGBD/color/00000.jpg")
    depth_raw = o3d.io.read_image("../../TestData/RGBD/depth/00000.png")
    rgbd_image = o3d.geometry.RGBDImage.create_from_color_and_depth(
        color_raw, depth_raw)
    print(rgbd_image)

デフォルトの変換関数create_rgbd_image_from_color_and_depthは、色と奥行き画像のペアからRGBDImageを作成する。 カラー画像はグレースケール画像に変換され、[0、1]の範囲のfloatに格納される。 奥行き画像は、深度値をメートルで表した浮動小数点数で格納さる。 print(rgbd_image)は以下を表示する:

RGBDImage of size
Color image : 640x480, with 1 channels.
Depth image : 640x480, with 1 channels.
Use numpy.asarray to access buffer data.

変換された画像は、numpy配列としてレンダリングすることができる。

In [ ]:
    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()

その出力:

http://www.open3d.org/docs/release/_images/redwood_rgbd1.png

RGBD画像は、カメラ・パラメータにより、ポイントクラウド(点群)に変換することができる。

In [ ]:
    pcd = o3d.geometry.PointCloud.create_from_rgbd_image(
        rgbd_image,
        o3d.camera.PinholeCameraIntrinsic(
            o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault))
    # 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])

ここでは、デフォルトのカメラ・パラメータとしてPinholeCameraIntrinsic.prime_sense_defaultを使用する。 画像の解像度は640x480、焦点距離(fx、fy)=(525.0,525.0)、光学中心(cx、cy)=(319.5,239.5)である。 恒等行列(単位行列)がデフォルトの外部パラメータとして使用される。 pcd.transformは、わかりやすい視点にするため、ポイントクラウド上で上下反転変換を適用する。 この結果は: http://www.open3d.org/docs/release/_images/redwood_pcd1.png

SUN Dataset

ここではSUN dataset (Song et al. 2015)を読み、可視化する。

Song,S., Lichtenberg, S. & J. Xiao (2015). SUN RGB-D: A RGB-D Scene Understanding Benchmark Suite, CVPR.

In [1]:
%matplotlib inline
In [16]:
# examples/Python/Basic/rgbd_sun.py

import open3d as o3d
import matplotlib.pyplot as plt

if __name__ == "__main__":
    print("Read SUN dataset")
    color_raw = o3d.io.read_image(
        "../../TestData/RGBD/other_formats/SUN_color.jpg")
    depth_raw = o3d.io.read_image(
        "../../TestData/RGBD/other_formats/SUN_depth.png")
    rgbd_image = o3d.geometry.RGBDImage.create_from_sun_format(
        color_raw, depth_raw)
    print(rgbd_image)
    plt.figure(figsize=(12,5))    
    plt.subplot(1, 3, 1)
    plt.title('SUN original image')
    plt.imshow(color_raw)
    plt.subplot(1, 3,  2)
    plt.title('SUN grayscale image')
    plt.imshow(rgbd_image.color,cmap='gray')
    plt.subplot(1, 3, 3)
    plt.title('SUN depth image')
    plt.imshow(rgbd_image.depth)
    plt.show()
    pcd = o3d.geometry.PointCloud.create_from_rgbd_image(
        rgbd_image,
        o3d.camera.PinholeCameraIntrinsic(
            o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault))
    # 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])
Read SUN dataset
RGBDImage of size 
Color image : 640x480, with 1 channels.
Depth image : 640x480, with 1 channels.
Use numpy.asarray to access buffer data.

このチュートリアルは、Redwoodデータセットを処理するチュートリアルとほぼ同じであるが、唯一の違いは、変換関数create_rgbd_image_from_sun_formatを使用して、SUNデータセットの深度画像を解析することである。

RGBDImagenumpy配列としてレンダリングすることもできる:

ポイントクラウドとしても出力できる: http://www.open3d.org/docs/release/_images/sun_pcd.png

NYU Dataset

このチュートリアルは、NYU dataset (Silberman et al. 2012)のRGBDImageを読み込み可視化する。

Silberman, N., Hoiem, D., Kohli, P., & R. Fergus (2012) Indoor Segmentation and Support Inference from RGBD Images, ECCV.

Redwoodデータセットを処理するチュートリアルとほぼ同じであるが、2つの違いがある。 まず、NYUの画像は標準のjpg形式でもpng形式でもない。 したがって、mpimg.imreadを使用してカラー画像を数値配列として読み込み、それをOpen3D Imageに変換する。 追加のヘルパー関数read_nyu_pgmを呼び出し、NYUデータセットで使用される特別なビッグエンディアンのpgm形式から深度イメージを読み込む。 次に、変換関数create_rgbd_image_from_nyu_formatを使用して、SUNデータセットの深度イメージを解析する。

同様に、RGBDImageをnumpy配列としてレンダリングすることもできる:

In [17]:
# examples/Python/Basic/rgbd_nyu.py

import open3d as o3d
import numpy as np
import re
import matplotlib.image as mpimg
import matplotlib.pyplot as plt


# This is special function used for reading NYU pgm format
# as it is written in big endian byte order.
def read_nyu_pgm(filename, byteorder='>'):
    with open(filename, 'rb') as f:
        buffer = f.read()
    try:
        header, width, height, maxval = re.search(
            b"(^P5\s(?:\s*#.*[\r\n])*"
            b"(\d+)\s(?:\s*#.*[\r\n])*"
            b"(\d+)\s(?:\s*#.*[\r\n])*"
            b"(\d+)\s(?:\s*#.*[\r\n]\s)*)", buffer).groups()
    except AttributeError:
        raise ValueError("Not a raw PGM file: '%s'" % filename)
    img = np.frombuffer(buffer,
                        dtype=byteorder + 'u2',
                        count=int(width) * int(height),
                        offset=len(header)).reshape((int(height), int(width)))
    img_out = img.astype('u2')
    return img_out


if __name__ == "__main__":
    print("Read NYU dataset")
    # Open3D does not support ppm/pgm file yet. Not using o3d.io.read_image here.
    # MathplotImage having some ISSUE with NYU pgm file. Not using imread for pgm.
    color_raw = mpimg.imread("../../TestData/RGBD/other_formats/NYU_color.ppm")
    depth_raw = read_nyu_pgm("../../TestData/RGBD/other_formats/NYU_depth.pgm")
    color = o3d.geometry.Image(color_raw)
    depth = o3d.geometry.Image(depth_raw)
    rgbd_image = o3d.geometry.RGBDImage.create_from_nyu_format(color, depth)
    print(rgbd_image)
    plt.figure(figsize=(12,5))
    plt.subplot(1,3,1)
    plt.title('NYU original image')
    plt.imshow(color_raw)
    plt.subplot(1, 3, 2)
    plt.title('NYU grayscale image')
    plt.imshow(rgbd_image.color,cmap='gray')
    plt.subplot(1, 3,3)
    plt.title('NYU depth image')
    plt.imshow(rgbd_image.depth)
    plt.show()
    pcd = o3d.geometry.PointCloud.create_from_rgbd_image(
        rgbd_image,
        o3d.camera.PinholeCameraIntrinsic(
            o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault))
    # 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])
Read NYU dataset
RGBDImage of size 
Color image : 640x480, with 1 channels.
Depth image : 640x480, with 1 channels.
Use numpy.asarray to access buffer data.

ポイントクラウドとしても表示できる: http://www.open3d.org/docs/release/_images/nyu_pcd.png

TUM Dataset

TUM dataset (Strum et al. 2012)を用いた紹介。

Sturm, J., Engelhard, N., Endres, F., Burgard W., & D. Cremers (2012) A Benchmark for the Evaluation of RGB-D SLAM Systems, IROS.

これもRedwoodデータセットを処理するチュートリアルとほぼ同じであるが、唯一の違いは、変換関数create_rgbd_image_from_tum_formatを使用して、TUMデータセットの深度画像を解析することである。Redwoodデータセットと同様に、RGBDImageをnumpy配列としてレンダリングすることもできる。

In [18]:
	

# examples/Python/Basic/rgbd_tum.py

import open3d as o3d
import matplotlib.pyplot as plt

if __name__ == "__main__":
    print("Read TUM dataset")
    color_raw = o3d.io.read_image(
        "../../TestData/RGBD/other_formats/TUM_color.png")
    depth_raw = o3d.io.read_image(
        "../../TestData/RGBD/other_formats/TUM_depth.png")
    rgbd_image = o3d.geometry.RGBDImage.create_from_tum_format(
        color_raw, depth_raw)
    print(rgbd_image)
    plt.figure(figsize=(12,5))
    plt.subplot(1,3,1)
    plt.title('TUM original image') 
    plt.imshow(color_raw)
    plt.subplot(1,3,2)
    plt.title('TUM grayscale image')
    plt.imshow(rgbd_image.color,cmap='gray')
    plt.subplot(1, 3,3)
    plt.title('TUM depth image')
    plt.imshow(rgbd_image.depth)
    plt.show()
    pcd = o3d.geometry.PointCloud.create_from_rgbd_image(
        rgbd_image,
        o3d.camera.PinholeCameraIntrinsic(
            o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault))
    # 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])
Read TUM dataset
RGBDImage of size 
Color image : 640x480, with 1 channels.
Depth image : 640x480, with 1 channels.
Use numpy.asarray to access buffer data.

ポイントクラウドとしての表示: http://www.open3d.org/docs/release/_images/tum_pcd.png