forked from uoip/g2opy
-
Notifications
You must be signed in to change notification settings - Fork 0
/
Copy pathgicp_sba_demo.py
144 lines (106 loc) · 4.42 KB
/
gicp_sba_demo.py
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
# https://github.com/RainerKuemmerle/g2o/blob/master/g2o/examples/icp/gicp_sba_demo.cpp
import numpy as np
import g2o
import argparse
parser = argparse.ArgumentParser()
parser.add_argument('--num_points', type=int, help='num of points to use in projection SBA', default=0)
parser.add_argument('--pos_noise', type=float, help='noise in 3d position', default=0.2)
parser.add_argument('--pixel_noise', type=float, help='pixel noise', default=0.5)
parser.add_argument('--seed', type=int, help='random seed', default=0)
args = parser.parse_args()
def main():
optimizer = g2o.SparseOptimizer()
solver = g2o.BlockSolverX(g2o.LinearSolverCSparseX())
solver = g2o.OptimizationAlgorithmLevenberg(solver)
optimizer.set_algorithm(solver)
true_points = np.hstack([
np.random.random((1000, 1)) * 3 - 1.5,
np.random.random((1000, 1)) - 0.5,
np.random.random((1000, 1)) + 10])
focal_length = (500, 500)
principal_point = (320, 240)
baseline = 0.075
g2o.VertexSCam.set_cam(*focal_length, *principal_point, baseline)
for i in range(2):
t = np.array([0, 0, i])
cam = g2o.Isometry3d(np.identity(3), t)
vc = g2o.VertexSCam()
vc.set_id(i)
vc.set_estimate(cam)
if i == 0:
vc.set_fixed(True)
vc.set_all()
optimizer.add_vertex(vc)
trans0 = optimizer.vertex(0).estimate().inverse()
trans1 = optimizer.vertex(1).estimate().inverse()
for i in range(len(true_points)):
pt0 = trans0 * true_points[i]
pt1 = trans1 * true_points[i]
# add noise
pt0 += np.random.randn(3) * args.pos_noise
pt1 += np.random.randn(3) * args.pos_noise
# form edge, with normals in varioius positions
nm0 = np.array([0, i, 1])
nm0 = nm0 / np.linalg.norm(nm0)
nm1 = np.array([0, i, 1])
nm1 = nm1 / np.linalg.norm(nm1)
meas = g2o.EdgeGICP()
meas.pos0 = pt0
meas.pos1 = pt1
meas.normal0 = nm0
meas.normal1 = nm1
edge = g2o.Edge_V_V_GICP()
edge.set_vertex(0, optimizer.vertex(0))
edge.set_vertex(1, optimizer.vertex(1))
edge.set_measurement(meas)
edge.set_information(meas.prec0(0.01))
optimizer.add_edge(edge)
# set up SBA projections
if args.num_points > 0:
true_points = np.hstack([
np.random.random((args.num_points, 1)) * 3 - 1.5,
np.random.random((args.num_points, 1)) - 0.5,
np.random.random((args.num_points, 1)) + 10])
cam_num = 2
for i, point in enumerate(true_points):
vp = g2o.VertexSBAPointXYZ()
vp.set_id(cam_num + i)
vp.set_marginalized(True)
vp.set_estimate(point + np.random.randn(3))
optimizer.add_vertex(vp)
for j in range(cam_num):
z = optimizer.vertex(j).map_point(point)
if 0 <= z[0] < 640 and 0 <= z[1] < 480:
z += np.random.randn(3) * args.pixel_noise * [1, 1, 1/16.]
edge = g2o.Edge_XYZ_VSC()
edge.set_vertex(0, vp)
edge.set_vertex(1, optimizer.vertex(j))
edge.set_measurement(z)
edge.set_information(np.identity(3))
edge.set_robust_kernel(g2o.RobustKernelHuber())
optimizer.add_edge(edge)
# move second cam off of its true position
vc = optimizer.vertex(1)
cam = g2o.Isometry3d(vc.estimate().R, np.array([-0.1, -0.1, 0.2]))
vc.set_estimate(cam)
optimizer.initialize_optimization()
optimizer.compute_active_errors()
print('Initial chi2 =', optimizer.chi2())
optimizer.set_verbose(True)
optimizer.optimize(20)
print('\nSecond vertex should be near [0, 0, 1]')
print('before optimization:', cam.t)
print('after optimization:', optimizer.vertex(1).estimate().t)
print('error:', optimizer.vertex(1).estimate().t - [0, 0, 1])
'''
Mean squared error (average over 100 loops):
num_points 0: [ 0.29733384 0.40814327 0.03907623]
10: [ 0.14172112 0.31366953 0.03670497]
100: [ 0.05319327 0.37413272 0.01637925]
1000: [ 0.01652139 0.19501433 0.00502872]
10000: [ 0.00901297 0.04055765 0.00130791] -> 1/3 : 1 : 1/16
'''
if __name__ == '__main__':
if args.seed > 0:
np.random.seed(args.seed)
main()