-
Notifications
You must be signed in to change notification settings - Fork 2
/
Copy pathretinas.py
251 lines (204 loc) · 8.89 KB
/
retinas.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
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
198
199
200
201
202
203
204
205
206
207
208
209
210
211
212
213
214
215
216
217
218
219
220
221
222
223
224
225
226
227
228
229
230
231
232
233
234
235
236
237
238
239
240
241
242
243
244
245
246
247
248
249
250
251
import apriltag
import cv2
import utils.camera_streamer as cs
import numpy as np
from threading import Thread
import time
import networkx as nx
import numpy as np
import matplotlib.pyplot as plt
from objects import RetinaCamera, RetinaBody
from pose import Pose, get_cam_pose
from test_bodies.cube_body import cube0_body, cube1_body
from test_bodies.world_body_4_corners import world_body
from utils.convex_hull import get_convex_hull_area
from world import World
DEFAULT_APRILTAG_DETECTOR = apriltag.Detector()
STRENGTH_CONSTANT = 1 # k
class ApriltagObserver:
def __init__(self, camera_streamer, threshold=True):
self.camera_streamer = camera_streamer
self.threshold = threshold
self.detector = apriltag.Detector(apriltag.DetectorOptions(
families='tag36h11',
border=1,
nthreads=8,
quad_decimate=1.0,
quad_blur=0.0,
refine_edges=True,
refine_decode=False,
refine_pose=False,
debug=False,
quad_contours=True
))
self.grayscale_frame = None
self.frame = None
def get_observation(self):
ret, frame = self.camera_streamer.read()
if not ret:
return [], np.empty((0,2))
grayscale_frame = cv2.cvtColor(frame, cv2.COLOR_BGR2GRAY)
if self.threshold:
# grayscale_frame = cv2.GaussianBlur(grayscale_frame, (3,3),0)
_ , grayscale_frame = cv2.threshold(grayscale_frame, 0, 255, cv2.THRESH_BINARY+cv2.THRESH_OTSU)
results = self.detector.detect(grayscale_frame)
# print("After detect")
tag_list = []
label_list = []
point_list = []
for r in results:
tag_list.append(r.tag_id)
for corner in range(4):
label_list.append((r.tag_id, corner))
frame = cv2.circle(frame, np.array(r.corners[corner], dtype=np.int32), radius=0, color=(0,0, corner*63), thickness=16)
point_list.append(r.corners[corner])
# cv2.imshow(self.camera_streamer.name, grayscale_frame)
# print("Showed")
# print(tag_list)
if len(tag_list) < 3:
return self.get_observation()
self.frame = frame
self.grayscale_frame = grayscale_frame
return label_list, np.array(point_list)
class Retinas(Thread):
def __init__(self, observers, bodies, cameras, sticky=True):
super().__init__()
self.observers = observers
self.bodies = bodies
self.cameras = cameras
self.J = len(observers)
self.I = len(bodies)
self.sticky = sticky
self.tag_map = {}
for body in bodies:
point_dict = body.point_dict
for label in point_dict:
self.tag_map[label] = bodies.index(body)
self.world_camera_poses = {}
self.world_body_poses = {}
self.__is__running__ = True
self.start()
def run(self):
I = self.I
J = self.J
while self.__is__running__:
# All the following are J x I
N = {}
A = {}
T = {}
E = {}
G = {}
if self.sticky:
world_camera_poses = self.world_camera_poses
world_body_poses = self.world_body_poses
else:
world_camera_poses = {}#self.world_camera_poses
world_body_poses = {}#self.world_body_poses
counter = 0
for j in range(J):
k_labels, k_points = self.observers[j].get_observation()
# print(len(k_labels))
for k in range(len(k_labels)):
label, point = k_labels[k], k_points[k]
i = self.tag_map[label]
if (j, i) in N:
N[j, i][0].append(label)
N[j, i][1].append(point)
else:
counter += 1
N[j, i] = [label], [point]
for j, i in N:
observer = self.observers[j]
body = self.bodies[i]
labels, points = N[j, i]
A[j,i] = get_convex_hull_area(points)
T[j,i] = self.do_pnp(labels, points, observer, body)
E[j,i] = self.get_total_reprojection_error(labels, points, observer, body, T[j,i])[0]
# print(E[j,i])
temp = (len(N[j, i][0])**0.5) * E[j,i]
G[j,i] = - np.log(1 + np.exp(-STRENGTH_CONSTANT) * temp)
####################
nodes = ['b'+str(i) for i in range(I)] + ['c'+str(j) for j in range(J)]
graph = nx.Graph()
graph.add_nodes_from(nodes)
for j, i in G:
graph.add_edge('b'+str(i), 'c'+str(j))
graph.add_edge('c'+str(j), 'b'+str(i))
paths = nx.shortest_path(graph.to_undirected(), source='b0')
for node in paths:
path = paths[node]
pose = Pose(0,0,0,0,0,0)
cur = path[0]
for step in path[1:]:
# print()
source = int(cur[1])
destin = int(step[1])
if cur[0] == 'c':
pose = pose @ T[source, destin]
elif cur[0] == 'b':
pose = pose @ T[destin, source].invert()
else:
raise Exception()
cur = step
if node[0] == 'b':
world_body_poses[int(node[1])] = pose
self.bodies[int(node[1])].pose = pose
if node[0] == 'c':
world_camera_poses[int(node[1])] = pose
self.cameras[int(node[1])].pose = pose
####################
# for iteration in range(100):
# delta_world_body_poses = np.array((I, 6))
# for j in world_camera_poses:
# for i in world_body_poses:
# if (j, i) in T:
# labels, points = N[j, i]
# observer = self.observers[j]
# body = self.bodies[i]
# pose = world_body_poses[i].invert() @ world_camera_poses[j]
# error, jacobian = cv2.projectPoints(np.array(visible_body[1]), np.array(points), K, D)
# delta_T
####################
for i, body in enumerate(self.bodies):
if i not in world_body_poses:
world_body_poses[i] = None
self.bodies[i].pose = None
for j, camera in enumerate(self.cameras):
if j not in world_camera_poses:
world_camera_poses[j] = None
self.cameras[j].pose = None
# print(world_body_poses)
self.world_camera_poses = world_camera_poses
self.world_body_poses = world_body_poses
def get_total_reprojection_error(self, labels, points, observer, body, pose):
visible_body = labels, []
for label in labels:
visible_body[1].append(body.point_dict[label])
projected, jacobian = cv2.projectPoints(np.array(visible_body[1]), pose.rvec, pose.tvec, observer.camera_streamer.K, observer.camera_streamer.D)
error = np.sum(np.power(np.squeeze(projected)-points, 2))
jacobian_final = None
return error, jacobian_final
def do_pnp(self, labels, points, observer, body):
visible_body = labels, []
for label in labels:
visible_body[1].append(body.point_dict[label])
flag, rvec, tvec = cv2.solvePnP(np.array(visible_body[1]), np.array(points), observer.camera_streamer.K, observer.camera_streamer.D, flags=cv2.SOLVEPNP_EPNP)
# print(Pose(rvec, tvec))
# print(Pose(rvec, tvec).invert())
return Pose(rvec, tvec)
if __name__ == '__main__':
camera_streamers = []
# camera_streamers.append(cs.WebcamStreamer('rtsp://192.168.0.77:554', cs.iphone13_K))
# camera_streamers.append(cs.RemoteStreamer("http://192.168.0.222:8080/shot.jpg", cs.oneplus_8t_K))
camera_streamers.append(cs.WebcamStreamer(0, cs.oneplus_8t_K))
bodies = [world_body, cube0_body, cube1_body]
cameras = [RetinaCamera(camera_streamer) for camera_streamer in camera_streamers]
observers = [ApriltagObserver(camera_streamer) for camera_streamer in camera_streamers]
world = World("My World", bodies, cameras)
world.camera_pose = get_cam_pose((0.27, -1, 1),(0.27, 0.4, 0.3))
retinas = Retinas(observers, bodies, cameras)
while True:
for j, streamer in enumerate(camera_streamers):
if streamer.ret and (observers[j].frame is not None):
cv2.imshow(f"camera frame {j}", observers[j].frame)
world.display()