1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
21
22
23
24
25
26
27
28
29
30
31
32
33
34
35
36
37
38
39
40
41
42
43
44
45
46
47
48
49
50
51
52
53
54
55
56
57
58
59
60
61
62
63
64
65
66
67
68
69
70
71
72
73
74
75
76
77
78
79
80
81
82
83
84
85
86
87
88
89
90
91
92
93
94
95
96
97
98
99
100
101
102
103
104
105
106
107
108
109
110
111
112
113
114
115
116
117
118
119
120
121
122
123
124
125
126
127
128
129
130
131
132
133
134
135
136
137
138
139
140
141
142
143
144
145
146
147
148
149
150
151
152
153
154
155
156
157
158
159
160
161
162
163
164
165
166
167
168
169
170
171
172
173
174
175
176
177
178
179
180
181
182
183
184
185
186
187
188
189
190
191
192
193
194
195
196
197
|
# ----------------------------------------------------------------------------
# - Open3D: www.open3d.org -
# ----------------------------------------------------------------------------
# Copyright (c) 2018-2024 www.open3d.org
# SPDX-License-Identifier: MIT
# ----------------------------------------------------------------------------
from tqdm import tqdm
import numpy as np
import open3d as o3d
import open3d.core as o3c
from config import ConfigParser
from common import load_rgbd_file_names, load_depth_file_names, save_poses, load_intrinsic, load_extrinsics, get_default_dataset
def read_legacy_rgbd_image(color_file, depth_file, convert_rgb_to_intensity):
color = o3d.io.read_image(color_file)
depth = o3d.io.read_image(depth_file)
rgbd_image = o3d.geometry.RGBDImage.create_from_color_and_depth(
color,
depth,
depth_scale=1000.0,
depth_trunc=3.0,
convert_rgb_to_intensity=convert_rgb_to_intensity)
return rgbd_image
def rgbd_loop_closure(depth_list, color_list, intrinsic, config):
# TODO: load it from config
device = o3c.Device('CUDA:0')
interval = config.odometry_loop_interval
n_files = len(depth_list)
key_indices = list(range(0, n_files, interval))
n_key_indices = len(key_indices)
edges = []
poses = []
infos = []
pairs = []
criteria_list = [
o3d.t.pipelines.odometry.OdometryConvergenceCriteria(20),
o3d.t.pipelines.odometry.OdometryConvergenceCriteria(10),
o3d.t.pipelines.odometry.OdometryConvergenceCriteria(5)
]
method = o3d.t.pipelines.odometry.Method.PointToPlane
for i in range(n_key_indices - 1):
key_i = key_indices[i]
depth_curr = o3d.t.io.read_image(depth_list[key_i]).to(device)
color_curr = o3d.t.io.read_image(color_list[key_i]).to(device)
rgbd_curr = o3d.t.geometry.RGBDImage(color_curr, depth_curr)
for j in range(i + 1, n_key_indices):
key_j = key_indices[j]
depth_next = o3d.t.io.read_image(depth_list[key_j]).to(device)
color_next = o3d.t.io.read_image(color_list[key_j]).to(device)
rgbd_next = o3d.t.geometry.RGBDImage(color_next, depth_next)
# TODO: add OpenCV initialization if necessary
# TODO: better failure check
try:
res = o3d.t.pipelines.odometry.rgbd_odometry_multi_scale(
rgbd_curr, rgbd_next, intrinsic, o3c.Tensor(np.eye(4)),
1000.0, 3.0, criteria_list, method)
info = o3d.t.pipelines.odometry.compute_odometry_information_matrix(
depth_curr, depth_next, intrinsic, res.transformation, 0.07,
1000.0, 3.0)
except Exception as e:
pass
else:
if info[5, 5] / (depth_curr.columns * depth_curr.rows) > 0.3:
edges.append((key_i, key_j))
poses.append(res.transformation.cpu().numpy())
infos.append(info.cpu().numpy())
# pcd_src = o3d.t.geometry.PointCloud.create_from_rgbd_image(
# rgbd_curr, intrinsic)
# pcd_dst = o3d.t.geometry.PointCloud.create_from_rgbd_image(
# rgbd_next, intrinsic)
# o3d.visualization.draw([pcd_src, pcd_dst])
# o3d.visualization.draw(
# [pcd_src.transform(res.transformation), pcd_dst])
return edges, poses, infos
def rgbd_odometry(depth_list, color_list, intrinsic, config):
# TODO: load it from config
device = o3c.Device('CUDA:0')
n_files = len(depth_list)
depth_curr = o3d.t.io.read_image(depth_list[0]).to(device)
color_curr = o3d.t.io.read_image(color_list[0]).to(device)
rgbd_curr = o3d.t.geometry.RGBDImage(color_curr, depth_curr)
# TODO: load all params and scale/max factors from config
edges = []
poses = []
infos = []
criteria_list = [
o3d.t.pipelines.odometry.OdometryConvergenceCriteria(20),
o3d.t.pipelines.odometry.OdometryConvergenceCriteria(10),
o3d.t.pipelines.odometry.OdometryConvergenceCriteria(5)
]
method = o3d.t.pipelines.odometry.Method.PointToPlane
for i in tqdm(range(0, n_files - 1)):
depth_next = o3d.t.io.read_image(depth_list[i + 1]).to(device)
color_next = o3d.t.io.read_image(color_list[i + 1]).to(device)
rgbd_next = o3d.t.geometry.RGBDImage(color_next, depth_next)
res = o3d.t.pipelines.odometry.rgbd_odometry_multi_scale(
rgbd_curr, rgbd_next, intrinsic, o3c.Tensor(np.eye(4)), 1000.0, 3.0,
criteria_list, method)
info = o3d.t.pipelines.odometry.compute_odometry_information_matrix(
depth_curr, depth_next, intrinsic, res.transformation, 0.07, 1000.0,
3.0)
edges.append((i, i + 1))
poses.append(res.transformation.cpu().numpy())
infos.append(info.cpu().numpy())
color_curr = color_next
depth_curr = depth_next
rgbd_curr = rgbd_next
return edges, poses, infos
if __name__ == '__main__':
parser = ConfigParser()
parser.add('--config',
is_config_file=True,
help='YAML config file path.'
'Please refer to config.py for the options,'
'and default_config.yml for default settings '
'It overrides the default config file, but will be '
'overridden by other command line inputs.')
parser.add('--default_dataset',
help='Default dataset is used when config file is not provided. '
'Default dataset may be selected from the following options: '
'[lounge, jack_jack]',
default='lounge')
config = parser.get_config()
if config.path_dataset == '':
config = get_default_dataset(config)
depth_file_names, color_file_names = load_rgbd_file_names(config)
intrinsic = load_intrinsic(config)
i = 0
j = 10
depth_src = o3d.t.io.read_image(depth_file_names[i])
color_src = o3d.t.io.read_image(color_file_names[i])
depth_dst = o3d.t.io.read_image(depth_file_names[j])
color_dst = o3d.t.io.read_image(color_file_names[j])
rgbd_src = o3d.t.geometry.RGBDImage(color_src, depth_src)
rgbd_dst = o3d.t.geometry.RGBDImage(color_dst, depth_dst)
# RGBD odmetry and information matrix computation
res = o3d.t.pipelines.odometry.rgbd_odometry_multi_scale(
rgbd_src, rgbd_dst, intrinsic)
info = o3d.t.pipelines.odometry.compute_odometry_information_matrix(
depth_src, depth_dst, intrinsic, res.transformation, 0.07)
print(res.transformation, info)
print(info[5, 5] / (depth_src.columns * depth_src.rows))
# Legacy for reference, can be a little bit different due to minor implementation discrepancies
rgbd_src_legacy = read_legacy_rgbd_image(color_file_names[i],
depth_file_names[i], True)
rgbd_dst_legacy = read_legacy_rgbd_image(color_file_names[j],
depth_file_names[j], True)
intrinsic_legacy = o3d.camera.PinholeCameraIntrinsic(
o3d.camera.PinholeCameraIntrinsicParameters.PrimeSenseDefault)
success, trans, info = o3d.pipelines.odometry.compute_rgbd_odometry(
rgbd_src_legacy, rgbd_dst_legacy, intrinsic_legacy, np.eye(4),
o3d.pipelines.odometry.RGBDOdometryJacobianFromHybridTerm())
print(trans, info)
# Visualization
pcd_src = o3d.t.geometry.PointCloud.create_from_rgbd_image(
rgbd_src, intrinsic)
pcd_dst = o3d.t.geometry.PointCloud.create_from_rgbd_image(
rgbd_dst, intrinsic)
o3d.visualization.draw([pcd_src, pcd_dst])
o3d.visualization.draw([pcd_src.transform(res.transformation), pcd_dst])
|
