-
Notifications
You must be signed in to change notification settings - Fork 0
/
wrapper.py
214 lines (182 loc) · 7.59 KB
/
wrapper.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
#!/usr/bin/env python2
# -*- coding: utf-8 -*-
"""
Created on Mon Apr 6 12:08:49 2020
@author: kumar
"""
import cv2
import os
import numpy as np
import glob
import matplotlib.pyplot as plt
from scipy.optimize import curve_fit
def v_compute(i,j,H):
#print H
v_row=np.zeros((1,6), np.float32)
v_row[0][0]=H[0][i]*H[0][j]
v_row[0][1]=H[0][i]*H[1][j]+H[1][i]*H[0][j]
v_row[0][2]=H[1][i]*H[1][j]
v_row[0][3]=H[2][i]*H[0][j]+H[0][i]*H[2][j]
v_row[0][4]=H[2][i]*H[1][j]+H[1][i]*H[2][j]
v_row[0][5]=H[2][i]*H[2][j]
return v_row
"""def fn_compute(objpointsHomgenous,A0,A1,A2,A3,A4,A5,A6,A7,A8,RT0,RT1,RT2,RT3,RT4,RT5,RT6,RT7,RT8):
#def fn_compute(objpointsHomgenous,A,RT,k1,k2):
A=np.array([[A0,A1,A2],[A3,A4,A5],[A6,A7,A8]])
RT=np.array([[RT0,RT1,RT2],[RT3,RT4,RT5],[RT6,RT7,RT8]])
#A=A.reshape(3,3)
#RT=RT.reshape(3,3)
objpointsHomgenous=objpointsHomgenous.reshape(54,3)
#H1=np.dot(A,RT)
cameraFrameCoord=(np.dot(RT,objpointsHomgenous.T)).T
normalized_coord=cameraFrameCoord/(cameraFrameCoord[:,2].reshape(54,1))
normalizedSquareDist=np.square(normalized_coord[:,1])+np.square(normalized_coord[:,0])
u0,v0=A2,A5
#distortionFactor=k1*normalizedSquareDist+(k2*np.square(normalizedSquareDist))
imgPointEst=np.dot(A,normalized_coord.T).T
#print(objpointsHomgenous)
imgPointEst=imgPointEst/(imgPointEst[:,2].reshape(54,1))
#imgPointEst[:,0]=imgPointEst[:,0]+(imgPointEst[:,0]-u0)*distortionFactor
#imgPointEst[:,1]=imgPointEst[:,1]+(imgPointEst[:,1]-v0)*distortionFactor
#imgPointEstimated.append(imgPointEst)
return imgPointEst.ravel() """
imgPointsStore=[]
def fn_compute(objpointsHomgenous,A0,A1,A2,A3,A4,A5,A6,A7,A8,k1,k2,*args):
#def fn_compute(objpointsHomgenous,A,RT,k1,k2):
AT=np.array([[A0,A1,A2],[A3,A4,A5],[A6,A7,A8]])
RT=np.array(args[0:117]).reshape(13,3,3)
#A=A.reshape(3,3)
#print(len(kwargs))
imgPointEstimated=[]
objpointsHomgenous=objpointsHomgenous.reshape(13,54,3)
#H1=np.dot(A,RT)
for i in range(13):
cameraFrameCoord=(np.dot(RT[i],objpointsHomgenous[i].T)).T
normalized_coord=cameraFrameCoord/(cameraFrameCoord[:,2].reshape(54,1))
normalizedSquareDist=np.square(normalized_coord[:,1])+np.square(normalized_coord[:,0])
u0,v0=A2,A5
distortionFactor=k1*normalizedSquareDist+(k2*np.square(normalizedSquareDist))
imgPointEst=np.dot(AT,normalized_coord.T).T
#print(objpointsHomgenous)
imgPointEst=imgPointEst/(imgPointEst[:,2].reshape(54,1))
imgPointEst[:,0]=imgPointEst[:,0]+(imgPointEst[:,0]-u0)*distortionFactor
imgPointEst[:,1]=imgPointEst[:,1]+(imgPointEst[:,1]-v0)*distortionFactor
imgPointEstimated.append(imgPointEst)
imgPointsStore.append(imgPointEst)
return (np.array(imgPointEstimated,dtype=np.float32)).ravel()
path="./Calibration_Imgs/*.jpg"
# termination criteria
criteria = (cv2.TERM_CRITERIA_EPS + cv2.TERM_CRITERIA_MAX_ITER, 30, 0.001)
objp = np.zeros((9*6,3), np.float32)
objpHomgenous = np.ones((9*6,3), np.float32)
objp[:,:2] = np.mgrid[0:9,0:6].T.reshape(-1,2)
objpHomgenous[:,:2] = np.mgrid[0:9,0:6].T.reshape(-1,2)
objpoints = [] # 3d point in real world space
objpointsHomgenous=[]#3d homogenous points with z neglected as z=0
imgpoints = [] # 2d points in image plane.
H=[]
V=np.zeros((26,6),np.float32)
for i, fil in enumerate(glob.glob(path)):
image=cv2.imread(fil)
gray=cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
ret, corners=cv2.findChessboardCorners(gray,(9,6))
if ret==True :
objpoints.append(objp)
objpointsHomgenous.append(objpHomgenous)
corners2 = cv2.cornerSubPix(gray,corners,(11,11),(-1,-1),criteria)
corners3=corners2.reshape(-1,2)
corners3=np.column_stack((corners3,np.ones((54,1),np.float32)))
imgpoints.append(corners3)
h,_=cv2.findHomography(objp,imgpoints[i])
#h=cv2.getPerspectiveTransform(objp,imgpoints[i])
V[2*i]=v_compute(0,1,h)
V[2*i+1]=v_compute(0,0,h)-v_compute(1,1,h)
H.append(h)
#img=cv2.drawChessboardCorners(image,(9,6),np.array(corners3[:,0:2].reshape(54,1,2),dtype=np.float32),False)
#plt.figure(14)
#plt.imshow(img)
eigval,eigvec=np.linalg.eig(np.dot(V.T,V))
min_eigval_col=(np.argwhere(eigval==min(eigval))).item()
B=eigvec[:,min_eigval_col]
b11=B[0]
b12=B[1]
b22=B[2]
b13=B[3]
b23=B[4]
b33=B[5]
A=np.zeros((3,3),np.float32)
A[1][2]=((b12*b13)-(b11*b23))/((b11*b22)-np.power(b12,2))
l=b33-((np.power(b13,2)+(A[1][2]*((b12*b13)-(b11*b23))))/b11)
A[0][0]=np.sqrt(l/b11)
A[1][1]=np.sqrt((l*b11)/((b11*b22)-np.power(b12,2)))
A[0][1]=-(b12*(np.power(A[0][0],2))*A[1][1])/l
A[0][2]=(A[1][2]*A[0][1]/A[1][1])-(b13*np.square(A[0][0])/l)
A[2][2]=1.0
R=[]
T=[]
RT=[]
for i, hom in enumerate(H):
l=1/(np.linalg.norm(np.dot(np.linalg.inv(A),hom[:,0])))
r1=l*(np.dot(np.linalg.inv(A),hom[:,0]))
r2=l*(np.dot(np.linalg.inv(A),hom[:,1]))
r3=np.cross(r1,r2)
R.append(np.column_stack((r1,r2,r3)))
T.append(l*(np.dot(np.linalg.inv(A),hom[:,2])))
RT.append(np.array(np.column_stack((r1,r2,l*(np.dot(np.linalg.inv(A),hom[:,2])))),dtype=np.float32))
#imgPointEstimated=fn_compute(objpointsHomgenous, A,RT)
#imgPointEstimated=[]
k1=np.float32(0.0)
k2=np.float32(0.0)
"""for i in range (13):
p0=A.ravel(),RT[i].ravel()
p0=np.hstack(p0)
coeff,x=curve_fit(fn_compute,objpointsHomgenous[i].ravel(),imgpoints[i].ravel(),p0)
#print(coeff)
#A=coeff[0:9].reshape(3,3)
#k1=coeff[18]
#k2=coeff[19]
A_estimated=coeff[0:9].reshape(3,3)
print(A_estimated)"""
p0=A.ravel(),k1,k2,np.array(RT).ravel()
p0=np.hstack(p0)
objpointsHomgenous=np.array(objpointsHomgenous,dtype=np.float32)
imgpoints=np.array(imgpoints,dtype=np.float32)
coeff,x=curve_fit(fn_compute,objpointsHomgenous.ravel(),imgpoints.ravel(),p0)
#print(coeff)
#A=coeff[0:9].reshape(3,3)
#k1=coeff[18]
#k2=coeff[19]
A_estimated=coeff[0:9].reshape(3,3)
k_estimated=np.zeros((1,4),dtype=np.float32)
k_estimated[0][0]=coeff[9]
k_estimated[0][1]=coeff[10]
RT_estimated=coeff[11:128].reshape(13,3,3)
print("Initial Intrinsic Matrix", A)
print("Estimated Intrinsic Matrix",A_estimated)
print("Estimated Radial distortion coefficients", k_estimated)
for i, fil in enumerate(glob.glob(path)):
image=cv2.imread(fil)
gray=cv2.cvtColor(image,cv2.COLOR_BGR2GRAY)
undistort_img=cv2.undistort(image,A,(k_estimated[0][0].item(),k_estimated[0][1].item(),0,0),A_estimated)
undistort_img=cv2.drawChessboardCorners(undistort_img,(9,6),np.array(imgPointsStore[i][:,0:2].reshape(54,1,2),dtype=np.float32),False)
fig=plt.figure(i+1,(1,2))
fig.add_subplot(1,2,1)
fig.suptitle("Left: distorted image and Right:undistorted image")
image=cv2.drawChessboardCorners(image,(9,6),np.array(imgPointsStore[i][:,0:2].reshape(54,1,2),dtype=np.float32),False)
plt.imshow(image)
fig.add_subplot(1,2,2)
plt.imshow(undistort_img)
tot_error = 0
R_estimated=[]
T_estimated=[]
for i in range(13):
R_estimated.append(np.column_stack((RT_estimated[i][:,0],RT_estimated[i][:,1],np.cross(RT_estimated[i][:,0],RT_estimated[i][:,1]))))
T_estimated.append(RT_estimated[i][:,2])
rvec=cv2.Rodrigues(R_estimated[i])
imgpoints2, _ = cv2.projectPoints(objpointsHomgenous[i], rvec[0], T_estimated[i], A_estimated, k_estimated)
imgpoints2=np.array(imgpoints2.reshape(54,2),dtype=np.float32)
#print(imgpoints2.shape)
#print(imgpoints[i].shape)
error = cv2.norm(imgpoints[i][:,0:2],imgpoints2, cv2.NORM_L2)/len(imgpoints2)
tot_error += error
print "Reprojection error: ", tot_error/13.0