how to maintain the surround view cameras (4 camera images) position in the point cloud

### Checklist

- [X] I have searched for [similar issues](https://github.com/isl-org/Open3D/issues).
- [X] For Python issues, I have tested with the [latest development wheel](http://www.open3d.org/docs/latest/getting_started.html#development-version-pip).
- [X] I have checked the [release documentation](http://www.open3d.org/docs/release/) and the [latest documentation](http://www.open3d.org/docs/latest/) (for `master` branch).

### My Question

Hi,

I am trying to generate 3D point cloud from 4 RGB-D images. I am able to do that with Open3D but I am unable to maintain the position of the images. You can find the camera_parameters.json [here](https://drive.google.com/file/d/1euk0oB7BXyurH6Bhb9ropMpmbJSjnXY4/view?usp=sharing).

```
import open3d as o3d
import cv2
import os
import numpy as np
import matplotlib.pyplot as plt
import json

def load_json(path):
with open(path) as f:
return json.load(f)

def parse_camera_params(camera):
param = {}
param['camera'] = camera['camera']
param['depth'] = {}
param['depth']['fx'] = camera['K_depth'][0][0]
param['depth']['fy'] = camera['K_depth'][1][1]
param['depth']['cx'] = camera['K_depth'][0][2]
param['depth']['cy'] = camera['K_depth'][1][2]
param['depth']['K'] = camera['K_depth']
# ignore distCoeffs_depth's 5th (1000) and 6th (0) element
# since they are strange
param['depth']['distCoeffs'] = np.array(camera['distCoeffs_depth'][:5])

param['depth_width'] = camera['depth_width']
param['depth_height'] = camera['depth_height']


param['color'] = {}
param['color']['fx'] = camera['K_color'][0][0]
param['color']['fy'] = camera['K_color'][1][1]
param['color']['cx'] = camera['K_color'][0][2]
param['color']['cy'] = camera['K_color'][1][2]
param['color']['K'] = camera['K_color']
# ignore distCoeffs_color's 5th (1000) and 6th (0) element
# since they are strange
param['color']['distCoeffs'] = np.array(camera['distCoeffs_color'][:5])

param['color_width'] = camera['color_width']
param['color_height'] = camera['color_height']

# world to depth
w2d_T = np.array(camera['M_world2sensor'])
param['w2d_R'] = w2d_T[0:3, 0:3]
param['w2d_t'] = w2d_T[0:3, 3]
param['w2d_T'] = camera['M_world2sensor']

d2w_T = np.linalg.inv(w2d_T)
param['d2w_R'] = d2w_T[0:3, 0:3]
param['d2w_t'] = d2w_T[0:3, 3]
param['d2w_T'] = d2w_T

return param

if __name__ == '__main__':
data_dir = "data/"
camera_params = load_json(os.path.join(data_dir,
'camera_parameters.json'))

SVC_NUM = 4
pcd_combined = o3d.geometry.PointCloud()

for i in range(SVC_NUM):

param = parse_camera_params(camera_params['sensors'][i])

color = cv2.imread(os.path.join(data_dir, 'color_{:05d}.png'.format(i)))
color = cv2.cvtColor(color, cv2.COLOR_BGR2RGB)

depth = cv2.imread(os.path.join(data_dir, 'depth_{:05d}.png'.format(i)), -1)
# depth = depth * 0.001 # factor to scale the depth image from carla

o3d_color = o3d.geometry.Image(color)
o3d_depth = o3d.geometry.Image(depth)

rgbd_image = o3d.geometry.RGBDImage.create_from_tum_format(o3d_color, o3d_depth, False)

h, w = depth.shape

dfx, dfy, dcx, dcy = param['depth']['fx'], param['depth']['fy'], param['depth']['cx'], param['depth']['cy']

intrinsic = o3d.camera.PinholeCameraIntrinsic(w, h, dfx,dfy, dcx, dcy)
intrinsic.intrinsic_matrix = param['depth']['K']
cam = o3d.camera.PinholeCameraParameters()
cam.intrinsic = intrinsic
cam.extrinsic = np.array(param['w2d_T'])

pcd = o3d.geometry.PointCloud.create_from_rgbd_image(rgbd_image, cam.intrinsic, cam.extrinsic)

o3d.io.write_point_cloud("svc_{:05d}_v13.pcd".format(i), pcd)

pcd_combined += pcd

o3d.io.write_point_cloud("svc_global_v13.pcd", pcd_combined)
```

With the above code I am getting output of ```svc_global_v13.pcd``` like below


As you can see, all the images are projected into center. As indicated in the json file, I would like the images to be positioned as left, right, front and rear in the 3D point cloud.

It is highly appreciated, if someone can help me to fix this issue.

Thanks.