def convert_points(j, matches):
"""将匹配转换成齐次坐标点的函数"""
ndx = matches[j].nonzero()[0]
fp = homography.make_homog(kp1[ndx, :2].T)
ndx2 = [int(matches[j][i]) for i in ndx]
tp = homography.make_homog(kp2[ndx2, :2].T)
return fp, tp
def convert_points(matches, l, j):
ndx = matches[j].nonzero()[0]
fp = homography.make_homog(l[j + 1][ndx, :2].T)
ndx2 = [int(matches[j][i]) for i in ndx]
tp = homography.make_homog(l[j][ndx2, :2].T)
return fp, tp
def convert_points(j, matches):
"""将匹配转换成齐次坐标点的函数"""
ndx = matches[j].nonzero()[0]
fp = homography.make_homog(kp1[ndx, :2].T)
ndx2 = [int(matches[j][i]) for i in ndx]
tp = homography.make_homog(kp2[ndx2, :2].T)
return fp, tp
def convert_points(matches,kp1,kp2):
# ndx = matches.nonzero()[0]
fp=np.array([kp1[m.queryIdx].pt for m in matches])
fp = homography.make_homog(fp.T)
# ndx2 = [int(matches[i]) for i in ndx]
tp=np.array([kp2[m.trainIdx].pt for m in matches])
tp = homography.make_homog(tp.T)
return fp,tp
def convert_points(j): ndx = matches[j].nonzero()[0] fp = homography.make_homog(l[j+1][ndx,:2].T) ndx2 = [int(matches[j][i]) for i in ndx] tp = homography.make_homog(l[j][ndx2,:2].T) # switch x and y - TODO this should move elsewhere fp = vstack([fp[1],fp[0],fp[2]]) tp = vstack([tp[1],tp[0],tp[2]]) return fp,tp
def convert_points(j):
ndx = matches[j].nonzero()[0]
fp = homography.make_homog(l[j + 1][ndx, :2].T)
ndx2 = [int(matches[j][i]) for i in ndx]
tp = homography.make_homog(l[j][ndx2, :2].T)
# switch x and y - TODO this should move elsewhere
fp = vstack([fp[1], fp[0], fp[2]])
tp = vstack([tp[1], tp[0], tp[2]])
return fp, tp
def get_krt(im1, im2):
ims = [im1, im2]
sifts = []
for x in range(2):
sifts.append(ims[x][:-3]+"sift")
# compute features
#sift.process_image('../../data/book_frontal.JPG','../../data/im0.sift')
sift.process_image(ims[0],sifts[0])
l0,d0 = sift.read_features_from_file(sifts[0])
#sift.process_image('../../data/book_perspective.JPG','../../data/im1.sift')
sift.process_image(ims[1],sifts[1])
l1,d1 = sift.read_features_from_file(sifts[1])
# match features and estimate homography
matches = sift.match_twosided(d0,d1)
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx,:2].T)
ndx2 = [int(matches[i]) for i in ndx]
print len(ndx2)
tp = homography.make_homog(l1[ndx2,:2].T)
model = homography.RansacModel()
H,ransac_data = homography.H_from_ransac(fp,tp,model)
# camera calibration
#K = camera.my_calibration((747,1000))
K = camera.my_calibration((Image.open(im2).size))
# 3D points at plane z=0 with sides of length 0.2
box = cube.cube_points([0,0,0.1],0.1)
# project bottom square in first image
cam1 = camera.Camera( hstack((K,dot(K,array([[0],[0],[-1]])) )) )
# first points are the bottom square
box_cam1 = cam1.project(homography.make_homog(box[:,:5]))
# use H to transfer points to the second image
print dot(H,box_cam1)
box_trans = homography.normalize(dot(H,box_cam1))
# compute second camera matrix from cam1 and H
cam2 = camera.Camera(dot(H,cam1.P))
A = dot(linalg.inv(K),cam2.P[:,:3])
A = array([A[:,0],A[:,1],cross(A[:,0],A[:,1])]).T
cam2.P[:,:3] = dot(K,A)
# project with the second camera
box_cam2 = cam2.project(homography.make_homog(box))
# test: projecting point on z=0 should give the same
point = array([1,1,0,1]).T
print homography.normalize(dot(dot(H,cam1.P),point))
print cam2.project(point)
import pickle
with open('%s.pkl' % ims[1][:-4],'w') as f:
pickle.dump(K,f)
pickle.dump(dot(linalg.inv(K),cam2.P),f)
sys.stderr.write("K and Rt dumped to %s.pkl\n" % ims[1][:-4])
def get_krt(im1, im2):
ims = [im1, im2]
sifts = []
for x in range(2):
sifts.append(ims[x][:-3] + "sift")
# compute features
#sift.process_image('../../data/book_frontal.JPG','../../data/im0.sift')
sift.process_image(ims[0], sifts[0])
l0, d0 = sift.read_features_from_file(sifts[0])
#sift.process_image('../../data/book_perspective.JPG','../../data/im1.sift')
sift.process_image(ims[1], sifts[1])
l1, d1 = sift.read_features_from_file(sifts[1])
# match features and estimate homography
matches = sift.match_twosided(d0, d1)
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
print len(ndx2)
tp = homography.make_homog(l1[ndx2, :2].T)
model = homography.RansacModel()
H, ransac_data = homography.H_from_ransac(fp, tp, model)
# camera calibration
#K = camera.my_calibration((747,1000))
K = camera.my_calibration((Image.open(im2).size))
# 3D points at plane z=0 with sides of length 0.2
box = cube.cube_points([0, 0, 0.1], 0.1)
# project bottom square in first image
cam1 = camera.Camera(hstack((K, dot(K, array([[0], [0], [-1]])))))
# first points are the bottom square
box_cam1 = cam1.project(homography.make_homog(box[:, :5]))
# use H to transfer points to the second image
print dot(H, box_cam1)
box_trans = homography.normalize(dot(H, box_cam1))
# compute second camera matrix from cam1 and H
cam2 = camera.Camera(dot(H, cam1.P))
A = dot(linalg.inv(K), cam2.P[:, :3])
A = array([A[:, 0], A[:, 1], cross(A[:, 0], A[:, 1])]).T
cam2.P[:, :3] = dot(K, A)
# project with the second camera
box_cam2 = cam2.project(homography.make_homog(box))
# test: projecting point on z=0 should give the same
point = array([1, 1, 0, 1]).T
print homography.normalize(dot(dot(H, cam1.P), point))
print cam2.project(point)
import pickle
with open('%s.pkl' % ims[1][:-4], 'w') as f:
pickle.dump(K, f)
pickle.dump(dot(linalg.inv(K), cam2.P), f)
sys.stderr.write("K and Rt dumped to %s.pkl\n" % ims[1][:-4])
def get_camera_params(sz, H):
"""
Get camere parametes: K, Rts.
"""
# camera calibration
K = my_calibration((sz))
#print(K)
# 3D points at plane z=0 with sides of length 0.2
box = cube_points([0, 0, 0.1], 0.1)
# project bottom square in first image
cam1 = Camera(np.hstack((K, np.dot(K, np.array([[0], [0], [-1]])))))
# first points are the bottom square
box_cam1 = cam1.project(homography.make_homog(box[:, :5]))
# use H to transfer points to the second image
box_trans = homography.normalize(np.dot(H, box_cam1))
# compute second camera metrx from cam1 and H
cam2 = Camera(np.dot(H, cam1.P))
A = np.dot(linalg.inv(K), cam2.P[:, :3])
A = np.array([A[:, 0], A[:, 1], np.cross(A[:, 0], A[:, 1])]).T
cam2.P[:, :3] = np.dot(K, A)
Rt = np.dot(linalg.inv(K), cam2.P)
return K, Rt
def setUp(self):
points = array([
[-1.1, -1.1, -1.1],
[1.4, -1.4, -1.4],
[-1.5, 1.5, -1],
[1, 1.8, -1],
[-1.2, -1.2, 1.2],
[1.3, -1.3, 1.3],
[-1.6, 1.6, 1],
[1, 1.7, 1],
])
points = homography.make_homog(points.T)
P = hstack((eye(3), array([[0], [0], [0]])))
cam = camera.Camera(P)
self.x = cam.project(points)
r = [0.05, 0.1, 0.15]
rot = camera.rotation_matrix(r)
cam.P = dot(cam.P, rot)
cam.P[:, 3] = array([1, 0, 0])
self.x2 = cam.project(points)
print self.x2.shape
print self.x2
K, R, t = cam.factor()
self.expectedE = dot(sfm.skew(t), R)
self.expectedE /= self.expectedE[2, 2]
def compute_camera_matrix_1(K, H):
# 3D points at plane z=0 with sides of length 0.2
box = cube.cube_points([0, 0, 0.1], 0.1)
# project bottom square in first image
cam1 = camera.Camera(hstack((K, dot(K, array([[0], [0], [-1]])))))
# first points are the bottom square
box_cam1 = cam1.project(homography.make_homog(box[:, :5]))
# use H to transfer points to the second image
box_trans = homography.normalize(dot(H, box_cam1))
return cam1, box
def compute_camera_matrix_1(K, H):
# 3D points at plane z=0 with sides of length 0.2
box = cube.cube_points([0,0,0.1],0.1)
# project bottom square in first image
cam1 = camera.Camera( hstack((K,dot(K,array([[0],[0],[-1]])) )) )
# first points are the bottom square
box_cam1 = cam1.project(homography.make_homog(box[:,:5]))
# use H to transfer points to the second image
box_trans = homography.normalize(dot(H,box_cam1))
return cam1, box
def get_H(im0_path, im1_path):
"""
Get the Homography matrix.
"""
# compute features
sift.process_image(im0_path, 'im0.sift')
l0, d0 = sift.read_features_from_file('im0.sift')
sift.process_image(im1_path, 'im1.sift')
l1, d1 = sift.read_features_from_file('im1.sift')
# match features and estimate homography
matches = sift.match_twosided(d0, d1)
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l1[ndx2, :2].T)
model = homography.RansacModel()
H = homography.H_from_ransac(fp, tp, model)[0]
return H
def compute_homography():
# compute features
#sift.process_image('../../data/book_frontal.JPG','../../data/im0.sift')
im1 = '../../data/space_front.jpg'
im2 = '../../data/space_perspective.jpg'
im1 = '../../data/mag_front.jpg'
im2 = '../../data/mag_perspective.jpg'
ims = [im1, im2]
sifts = []
for k in range(2):
sifts.append(ims[k][:-4] + ".sift")
l0, d0 = sift.read_features_from_file(sifts[0])
l1, d1 = sift.read_features_from_file(sifts[1])
# match features and estimate homography
matches = sift.match_twosided(d0, d1)
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l1[ndx2, :2].T)
model = homography.RansacModel()
H, ransac_data = homography.H_from_ransac(fp, tp, model)
return H
def compute_homography():
# compute features
#sift.process_image('../../data/book_frontal.JPG','../../data/im0.sift')
im1='../../data/space_front.jpg'
im2='../../data/space_perspective.jpg'
im1='../../data/mag_front.jpg'
im2='../../data/mag_perspective.jpg'
ims = [im1, im2]
sifts = []
for k in range(2):
sifts.append(ims[k][:-4]+".sift")
l0,d0 = sift.read_features_from_file(sifts[0])
l1,d1 = sift.read_features_from_file(sifts[1])
# match features and estimate homography
matches = sift.match_twosided(d0,d1)
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx,:2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l1[ndx2,:2].T)
model = homography.RansacModel()
H,ransac_data = homography.H_from_ransac(fp,tp,model)
return H
def compute_camera_matrix_2(cam1, K, box, H):
# compute second camera matrix from cam1 and H
cam2 = camera.Camera(dot(H, cam1.P))
A = dot(linalg.inv(K), cam2.P[:, :3])
A = array([A[:, 0], A[:, 1], cross(A[:, 0], A[:, 1])]).T
cam2.P[:, :3] = dot(K, A)
# project with the second camera
box_cam2 = cam2.project(homography.make_homog(box))
# test: projecting point on z=0 should give the same
point = array([1, 1, 0, 1]).T
print(homography.normalize(dot(dot(H, cam1.P), point)))
print(cam2.project(point))
#H is correct as seen by the plots but cam2.P does not eem to be correct
#test cam2.P
return cam2
def compute_camera_matrix_2(cam1, K, box, H):
# compute second camera matrix from cam1 and H
cam2 = camera.Camera(dot(H,cam1.P))
A = dot(linalg.inv(K),cam2.P[:,:3])
A = array([A[:,0],A[:,1],cross(A[:,0],A[:,1])]).T
cam2.P[:,:3] = dot(K,A)
# project with the second camera
box_cam2 = cam2.project(homography.make_homog(box))
# test: projecting point on z=0 should give the same
point = array([1,1,0,1]).T
print(homography.normalize(dot(dot(H, cam1.P),point)))
print(cam2.project(point))
#H is correct as seen by the plots but cam2.P does not eem to be correct
#test cam2.P
return cam2
def setUp(self):
points = array([
[-1.1, -1.1, -1.1], [ 1.4, -1.4, -1.4], [-1.5, 1.5, -1], [ 1, 1.8, -1],
[-1.2, -1.2, 1.2], [ 1.3, -1.3, 1.3], [-1.6, 1.6, 1], [ 1, 1.7, 1],
])
points = homography.make_homog(points.T)
P = hstack((eye(3), array([[0], [0], [0]])))
cam = camera.Camera(P)
self.x = cam.project(points)
r = [0.05, 0.1, 0.15]
rot = camera.rotation_matrix(r)
cam.P = dot(cam.P, rot)
cam.P[:, 3] = array([1, 0, 0])
self.x2 = cam.project(points)
K, R, t = cam.factor()
self.expectedE = dot(sfm.skew(t), R)
self.expectedE /= self.expectedE[2, 2]
# --- import numpy as np import matplotlib.pyplot as plt import cv2 #自作モジュール import camera import image_processing as ip import desc_val as dv import homography import sfm im1, im2, points2D, points3D, corr, P = sfm.load_merton_data(show_detail=True) #3Dの点を同次座標系にして射影する X = homography.make_homog(points3D) x = P[0].project(X) points2D[0] # + #画像1の上に点を描写する ip.show_img(im1, figsize=(6, 4)) plt.plot(points2D[0][0], points2D[0][1], '*') plt.show() ip.show_img(im1, figsize=(6, 4)) plt.plot(x[0], x[1], '*') plt.show() # -
im1 = array(Image.open('out_ch4pics/h_image.jpg'))
#sift.plot_features(im1, l1, circle=True)
#show()
if not os.path.exists('out_ch04_markerpose.pickle'):
matches = sift.match(d0, d1)
pickle.dump(matches, open('out_ch04_markerpose.pickle', 'wb'))
matches = pickle.load(open('out_ch04_markerpose.pickle', 'rb'))
#figure()
#gray()
#sift.plot_matches(im0, im1, l0, l1, matches, show_below=False)
#show()
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l1[ndx2, :2].T)
#model = homography.RansacModel()
#H = homography.H_from_ransac(fp, tp, model)
# Not enough matches for H_from_ransac(), but all matches happen to be correct
# anyways, so no need for that.
H = homography.H_from_points(fp, tp)
K = camera.my_calibration(im0.shape[:2])
# How big this appears depends on the z translation in cam1 in the cam1 line.
box = cube_points([0, 0, 0.1], 0.1)
# project bottom square in first image
l1, d1 = sift.detectAndCompute(im2_gray, None)
#%%
''' Match points from img1 to points in img2 '''
bf = cv2.BFMatcher(crossCheck=True)
matches = bf.match(d0, d1)
#https://stackoverflow.com/questions/30716610/how-to-get-pixel-coordinates-from-feature-matching-in-opencv-python
kp0 = [l0[mat.queryIdx].pt for mat in matches]
kp1 = [l1[mat.trainIdx].pt for mat in matches]
kp0 = np.array(kp0)
kp1 = np.array(kp1)
#%%
''' Compute from and to points '''
fp = homography.make_homog(kp0[:100, :2].T)
tp = homography.make_homog(kp1[:100, :2].T)
model = homography.RansacModel()
H, _ = homography.H_from_ransac(fp, tp, model)
#%%
''' Draw cone '''
def cube_points(c, wid):
""" Creates a list of points for plotting
a cube with plot. (the first 5 points are
the bottom square, some sides repeated). """
p = []
#bottom
def processInput():
print ""
if inputArgs.left == "" or inputArgs.right == "":
print "Missing images!"
quit()
# here we go ...
# load image pair - we might need them for later
img_l = cv2.imread(inputArgs.left)
img_r = cv2.imread(inputArgs.right)
if img_l == None or img_r == None:
print "Missing images!"
quit()
### git them points - calibration
mkp_l = []
mkp_r = []
file_kp = open(inputArgs.nameCalibration, 'r')
if file_kp == None:
print "Missing matching points file"
quit()
for line in file_kp:
l_r = line.split(';')
mkp_l.append(
[float(l_r[0].split(',')[0]),
float(l_r[0].split(',')[1])])
mkp_r.append(
[float(l_r[1].split(',')[0]),
float(l_r[1].split(',')[1])])
file_kp.close()
mkp_l = np.float32(mkp_l)
mkp_r = np.float32(mkp_r)
### git them points - reconstruction
mkp_l_p = []
mkp_r_p = []
file_kp = open(inputArgs.namePoints, 'r')
if file_kp == None:
print "Missing matching points file"
quit()
for line in file_kp:
l_r = line.split(';')
mkp_l_p.append(
[float(l_r[0].split(',')[0]),
float(l_r[0].split(',')[1])])
mkp_r_p.append(
[float(l_r[1].split(',')[0]),
float(l_r[1].split(',')[1])])
file_kp.close()
mkp_l_p = np.float32(mkp_l_p)
mkp_r_p = np.float32(mkp_r_p)
### git them calibrations - left
K_l = []
file_c_l = open(inputArgs.leftCalibration, 'r')
if file_c_l == None:
print "Missing left calibration file"
quit()
for line in file_c_l:
c_l = line.split(' ')
K_l.append([float(c_l[0]), float(c_l[1]), float(c_l[2])])
file_c_l.close()
K_l = np.float32(K_l)
### git them calibrations - right
K_r = []
file_c_r = open(inputArgs.rightCalibration, 'r')
if file_c_r == None:
print "Missing right calibration file"
quit()
for line in file_c_r:
c_l = line.split(' ')
K_r.append([float(c_l[0]), float(c_l[1]), float(c_l[2])])
file_c_r.close()
K_r = np.float32(K_r)
### ok, now we start work
# make homogeneous and normalize with inv(K)
x1 = homography.make_homog(mkp_l_p.T)
x2 = homography.make_homog(mkp_r_p.T)
x1n = dot(inv(K_l), x1)
x2n = dot(inv(K_r), x2)
# compute E (E = (K_r)T * F * K_l)
#F, mask = cv2.findFundamentalMat(mkp_l, mkp_r, cv2.FM_8POINT)
F, mask = cv2.findFundamentalMat(mkp_l, mkp_r, cv2.FM_RANSAC, 1, 0.99)
# we select only inlier points - most pf the time this makes it worse
#mkp_l = mkp_l[mask.ravel()==1]
#mkp_r = mkp_r[mask.ravel()==1]
E = K_r.transpose()
E = E.dot(F)
E = E.dot(K_l)
# compute camera matrices (P2 will be list of four solutions)
P1 = array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0]])
P2 = sfm.compute_P_from_essential(E)
# pick the solution with points in front of cameras
ind = 0
maxres = 0
for i in range(4):
# triangulate inliers and compute depth for each camera
X = sfm.triangulate(x1n, x2n, P1, P2[i])
d1 = dot(P1, X)[2]
d2 = dot(P2[i], X)[2]
if sum(d1 > 0) + sum(d2 > 0) > maxres:
maxres = sum(d1 > 0) + sum(d2 > 0)
ind = i
infront = (d1 > 0) & (d2 > 0)
# triangulate inliers and remove points not in front of both cameras
X = sfm.triangulate(x1n, x2n, P1, P2[ind])
X = X[:, infront]
# visualization
if inputArgs.debug == 1:
# draw points
img_tmp = img_l.copy()
for kp in mkp_l_p:
x = int(kp[0])
y = int(kp[1])
cv2.circle(img_tmp, (x, y), 3, (0, 0, 255))
cv2.imwrite(inputArgs.namePoints + ".jpg", img_tmp)
# 3D plot
out_points = []
out_colors = []
for i in range(len(X[0])):
out_points.append([X[0][i], X[1][i], X[2][i]])
#out_colors.append(img_l[int(x1[1][i])][int(x1[0][i])])
out_colors.append([
img_l[int(x1[1][i])][int(x1[0][i])][2],
img_l[int(x1[1][i])][int(x1[0][i])][1],
img_l[int(x1[1][i])][int(x1[0][i])][0]
])
out_points = np.float32(out_points)
out_colors = np.float32(out_colors)
write_ply(inputArgs.namePoints + ".ply", out_points, out_colors)
if (np.all(homography_matrix == None)):
ip.imwrite(file_path, frame)
continue
# 第1の画像の底面の正方形を射影する
# カメラオブジェクトを定義
#平行移動でz軸に-1を定義しているのは
#立方体を手前に移動させるため
#X→x_q→x_tとしているため、z軸での移動は
#x_tでは拡大縮小に見える。
P = np.hstack((K, K @ np.array([[0], [0], [-5]])))
cam1 = camera.Camera(P)
# 最初の点群は、底面の正方形
pts = homography.make_homog(box[:, :5])
box_cam1 = cam1.project(pts)
# img_for_edit=img_query.copy()
# for i in range(5):
# circle_pt=tuple([int(pt) for pt in box_cam1[:2,i]])
# cv2.circle(img_for_edit,circle_pt,50,(255,255,255),thickness=-1)
# ip.show_img(img_for_edit,show_axis=True)
# cam1とHから第2のカメラ行列を計算する
P2 = homography_matrix @ cam1.P
cam2 = camera.Camera(P2)
# P=K (R|t)なので
#P[:,:3]は K R
#Pの前半3列にKの逆行列をかければRが算出できる。
import sys
sys.path.append("../homography")
sys.path.append("../camera")
sys.path.append("../local_image_descriptors")
import homography
import camera
import sift
# compute features
im1 = "../../data/book_frontal.JPG"
im2 = "../../data/book_perspective.JPG"
sift.process_image(im1,'../../data/im0.sift')
l0,d0 = sift.read_features_from_file('../../data/im0.sift')
sift.process_image(im2,'../../data/im1.sift')
l1,d1 = sift.read_features_from_file('../../data/im1.sift')
# match features and estimate homography
matches = sift.match_twosided(d0,d1)
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx,:2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l1[ndx2,:2].T)
model = homography.RansacModel()
H = homography.H_from_ransac(fp,tp,model)
for i in range(len(imname)):
l[i], d[i] = sift.read_or_compute(imname[i], siftname[i], histeq)
tic.k('loaded sifts')
print '{} / {} features'.format(len(d[0]), len(d[1]))
if not os.path.exists('out_ch05_recover_match.pickle'):
#matches = sift.match(d[0], d[1])
matches = sift.match_twosided(d[0], d[1])
pickle.dump(matches, open('out_ch05_recover_match.pickle', 'wb'))
matches = pickle.load(open('out_ch05_recover_match.pickle', 'rb'))
tic.k('matched')
ndx = matches.nonzero()[0]
x1 = homography.make_homog(l[0][ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
x2 = homography.make_homog(l[1][ndx2, :2].T)
print '{} matches'.format(len(ndx))
image = [numpy.array(Image.open(name)) for name in imname]
# calibration (FIXME?)
K = camera.my_calibration(image[0].shape[:2])
# Normalize with inv(K) (allows metric reconstruction).
x1n = numpy.dot(numpy.linalg.inv(K), x1)
x2n = numpy.dot(numpy.linalg.inv(K), x2)
tic.k('normalized')
from numpy import *
from matplotlib.pylab import *
# calibration
K = array([[2394,0,932],[0,2398,628],[0,0,1]])
print "load images and compute features"
im1 = array(Image.open('../data/alcatraz1.jpg'))
sift.process_image('../data/alcatraz1.jpg','im1.sift')
l1,d1 = sift.read_features_from_file('im1.sift')
im2 = array(Image.open('../data/alcatraz2.jpg'))
sift.process_image('../data/alcatraz2.jpg','im2.sift')
l2,d2 = sift.read_features_from_file('im2.sift')
print "match features"
matches = sift.match_twosided(d1,d2)
ndx = matches.nonzero()[0]
# make homogeneous and normalize with inv(K)
x1 = homography.make_homog(l1[ndx,:2].T)
ndx2 = [int(matches[i]) for i in ndx]
x2 = homography.make_homog(l2[ndx2,:2].T)
x1n = dot(inv(K),x1)
x2n = dot(inv(K),x2)
print "estimate E with RANSAC"
model = sfm.RansacModel()
E,inliers = sfm.F_from_ransac(x1n,x2n,model)
print "compute camera matrices (P2 will be list of four solutions)"
P1 = array([[1,0,0,0],[0,1,0,0],[0,0,1,0]])
P2 = sfm.compute_P_from_essential(E)
print "pick the solution with points in front of cameras"
ind = 0
maxres = 0
for i in range(4):
# triangulate inliers and compute depth for each camera
k = array([2394, 0, 932], [0, 2398, 628], [0, 0, 1])
# load images and compute features
im1 = array(Image.open('Building_1.jpg'))
sift.process_image('Building_1.jpg', 'im1.sift')
l1, d1 = sift.read_features_from_file('im1.sift')
im2 = array(Image.open('Building_1.jpg'))
sift.process_image('Building_2.jpg', 'im2.sift')
l2, d2 = sift.read_features_from_file('im2.sift')
# match features
matches = sift.match_twosided(d1, d2)
ndx = matches.nonzero()[0]
# make homogenous and normalize with inv(k)
x1 = homography.make_homo(l1[ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
x2 = homography.make_homog(l2[ndx2, :2].T)
x1n = dot(inv(k), x1)
x2n = dot(inv(k), x2)
#estunage E with RANSAC
model = sfm.RansacModel()
E, inliers = sfm.F_from_ransac(x1n, x2n, model)
# compute camera matrices (p2 will be list of four solutions)
P1 = array([1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0])
P2 = sfm.compute_P_from_essential(E)
for i in range(len(imname)):
l[i], d[i] = sift.read_or_compute(imname[i], siftname[i], histeq)
tic.k('loaded sifts')
print '{} / {} features'.format(len(d[0]), len(d[1]))
if not os.path.exists('out_ch05_recover_match.pickle'):
#matches = sift.match(d[0], d[1])
matches = sift.match_twosided(d[0], d[1])
pickle.dump(matches, open('out_ch05_recover_match.pickle', 'wb'))
matches = pickle.load(open('out_ch05_recover_match.pickle', 'rb'))
tic.k('matched')
ndx = matches.nonzero()[0]
x1 = homography.make_homog(l[0][ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
x2 = homography.make_homog(l[1][ndx2, :2].T)
print '{} matches'.format(len(ndx))
image = [numpy.array(Image.open(name)) for name in imname]
# calibration (FIXME?)
K = camera.my_calibration(image[0].shape[:2])
# Normalize with inv(K) (allows metric reconstruction).
x1n = numpy.dot(numpy.linalg.inv(K), x1)
x2n = numpy.dot(numpy.linalg.inv(K), x2)
tic.k('normalized')
l, d = {}, {}
for i in range(len(imname)):
l[i], d[i] = sift.read_or_compute(imname[i], siftname[i])
tic.k('loaded sifts')
if not os.path.exists('out_ch03_ex07_match.pickle'):
matches = sift.match(d[1], d[0])
pickle.dump(matches, open('out_ch03_ex07_match.pickle', 'wb'))
matches = pickle.load(open('out_ch03_ex07_match.pickle', 'rb'))
tic.k('matched')
ndx = matches.nonzero()[0]
fp = homography.make_homog(l[1][ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l[0][ndx2, :2].T)
tic.k('converted')
im1 = array(Image.open(imname[0]).convert('L'))
im2 = array(Image.open(imname[1]).convert('L'))
if len(im1.shape) == 2:
gray()
imshow(im1)
tic.k('imloaded')
model = homography.AffineRansacModel()
l, d = {}, {}
for i in range(len(imname)):
l[i], d[i] = sift.read_or_compute(imname[i], siftname[i])
tic.k('loaded sifts')
if not os.path.exists('out_ch03_ex06_match.pickle'):
matches = sift.match(d[1], d[0])
pickle.dump(matches, open('out_ch03_ex06_match.pickle', 'wb'))
matches = pickle.load(open('out_ch03_ex06_match.pickle', 'rb'))
tic.k('matched')
ndx = matches.nonzero()[0]
fp = homography.make_homog(l[1][ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l[0][ndx2, :2].T)
tic.k('converted')
im1 = array(Image.open(imname[0]).convert('L'))
im2 = array(Image.open(imname[1]).convert('L'))
if len(im1.shape) == 2:
gray()
imshow(im1)
tic.k('imloaded')
model = homography.RansacModel()
featlist = [imlist[i][:-3] + "sift" for i in range(imcount)]
with open("vocabulary.pkl", "rb") as f:
voc = pickle.load(f)
searcher = imagesearch.Searcher("test.db", voc)
query_imid = 50
res_count = 20
res = [w[1] for w in searcher.query(imlist[query_imid])[:res_count]]
print "regular results for query %d:" % query_imid, res
# Rerank by trying to fit a homography.
q_locs, q_descr = sift.read_features_from_file(featlist[query_imid])
fp = homography.make_homog(q_locs[:, :2].T)
model = homography.RansacModel()
rank = {}
for ndx in res[1:]:
locs, descr = sift.read_features_from_file(featlist[ndx - 1]) # res is 1-based
matches = sift.match(q_descr, descr)
ind = matches.nonzero()[0]
ind2 = [int(matches[i]) for i in ind]
tp = homography.make_homog(locs[:, :2].T)
try:
H, inliers = homography.H_from_ransac(fp[:, ind], tp[:, ind2], model, match_threshold=4)
except:
if len(sys.argv) != 2:
print 'usage: %prog image.jpeg'
sys.exit(1)
imname = sys.argv[1]
im = array(Image.open(imname))
imshow(im)
corners = array(ginput(4))
# FIXME: sort corners, currently expects tl, bl, br, tr order.
# Fake more realistic side lengths.
h = 400
w = int(np.linalg.norm(corners[3] - corners[0]) /
np.linalg.norm(corners[1] - corners[0]) * h)
corners = homography.make_homog(corners.T)
normalized = homography.make_homog(array([[0, 0, w, w], [0, h, h, 0]]))
H = homography.H_from_points(normalized, corners)
#normalized = homography.make_homog(array([[0, 0, 1, 1], [0, 1, 1, 0]]))
#print homography.H_from_points(normalized, corners)
#print H
if True:
overshoot = 0.8 # Adds a 80% border around the selected area.
s = (2 * overshoot + 1)
H = dot(H, array([[s, 0, -w*overshoot],
[0, s, -h*overshoot],
[0, 0, 1]]))
mapped = imtools.Htransform(im, H, (h, w))
tic.k('loaded sifts')
print '{} / {} features'.format(len(d[0]), len(d[1]))
immd5 = md5.md5(''.join(imname)).hexdigest()
matchcache = 'out_ch05ex02_cache_matches_%s.pickle' % immd5
if not os.path.exists(matchcache):
#matches = sift.match(d[0], d[1])
matches = sift.match_twosided(d[0], d[1])
pickle.dump(matches, open(matchcache, 'wb'))
matches = pickle.load(open(matchcache, 'rb'))
tic.k('matched')
ndx = matches.nonzero()[0]
x1 = homography.make_homog(l[0][ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
x2 = homography.make_homog(l[1][ndx2, :2].T)
print '{} matches'.format(len(ndx))
image = [numpy.array(Image.open(name)) for name in imname]
# calibration (FIXME?)
K = camera.my_calibration(image[0].shape[:2])
# Normalize with inv(K) (allows metric reconstruction).
x1n = numpy.dot(numpy.linalg.inv(K), x1)
x2n = numpy.dot(numpy.linalg.inv(K), x2)
tic.k('normalized')
im1 = array(Image.open(imname).convert('L'))
l1, d1 = sift.read_features_from_file('im1.sift')
figure()
gray()
sift.plot_features(im1, l1, circle=True)
show()
# 匹配特征
matches = sift.match_twosided(d0, d1)
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l1[ndx2, :2].T)
print 'fp', fp
print 'tp', tp
model = homography.RansacModel()
H = homography.H_from_ransac(fp, tp, model)
# 计算相机标定矩阵
K = my_calibration((747, 1000))
# 位于边长为0.2, z=0平面上的三维点
box = cube_points([0, 0, 0.1], 0.1)
# We use OpenCV instead of the calculus of the homography present in the book
H = find_homography(kp0, desc0, kp1, desc1)
# Note: always resize image to 747 x 1000 or change the K below
# camera calibration
K = my_calibration((480,640))
# 3D points at plane z=0 with sides of length 0.2
box = cube_points([0,0,0.1],0.1)
# project bottom square in first image
cam1 = camera.Camera( hstack((K,dot(K,array([[0],[0],[-1]])) )) )
# first points are the bottom square
box_cam1 = cam1.project(homography.make_homog(box[:,:5]))
# use H to transfer points to the second image
box_trans = homography.normalize(dot(H,box_cam1))
# compute second camera matrix from cam1 and H
cam2 = camera.Camera(dot(H,cam1.P))
A = dot(linalg.inv(K),cam2.P[:,:3])
A = array([A[:,0],A[:,1],cross(A[:,0],A[:,1])]).T
cam2.P[:,:3] = dot(K,A)
# project with the second camera
box_cam2 = cam2.project(homography.make_homog(box))
print(box_cam2)
import homography
import camera
import sift
#compute the feature
sift.process_image('/home/wangkai/Pictures/book_frontal.jpg','im0.sift')
l0,d0 = sift.read_feature_from_file('im0.sift')
sift.process_image('/home/wangkai/Pictures/book_prespective.jpg','im1.sift')
l1,d1 = sift.read_feature_from_file('im1.sift')
#match the feature, compute the homography
matches = sift.match_twoside(d0,d1)
ndx = matches.nozeros()[0]
fp = homography.make_homog(l0[ndx,:2].T)
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l1[ndx2,:2].T)
model = homography.RansacModel()
H = homography.H_from_ransac(fp,tp,model)
from numpy import * import homography f = array(([1, 1], [2, 1], [3, 2], [4, 4])) t = f + 1 f = f.T f = homography.make_homog(f) t = t.T t = homography.make_homog(t) #print f, t print homography.H_from_points(f, t)
K[0, 2] = 0.5 * col
K[1, 2] = 0.5 * row
return K
sift.process_image('book_frontal.JPG', 'im0.sift')
l0, d0 = sift.read_features_from_file('im0.sift')
sift.process_image('book_perspective.JPG', 'im1.sift')
l1, d1 = sift.read_features_from_file('im1.sift')
# 匹配特征，并计算单应性矩阵
# match features and estimate homography
matches = sift.match_twosided(d0, d1) #匹配
ndx = matches.nonzero()[0]
fp = homography.make_homog(l0[ndx, :2].T) #vstack
ndx2 = [int(matches[i]) for i in ndx]
tp = homography.make_homog(l1[ndx2, :2].T) #vstack
model = homography.RansacModel()
H, inliers = homography.H_from_ransac(fp, tp, model) #删除误匹配
# camera calibration
K = my_calibration((747, 1000)) #相机焦距光圈初始化
# 3D points at plane z=0 with sides of length 0.2
box = cube_points([0, 0, 0.1], 0.1)
# project bottom square in first image
cam1 = camera.Camera(hstack((K, dot(K, array([[0], [0], [-1]]))))) #相机初始化
# first points are the bottom square
H = h**o.H_from_ransac(obj1, obj2, model)
# # Draw first 60 matches.
# img3 = img2
# img3 = cv2.drawMatches(img1, kp1, img2, kp2, matches[:60], img3)
# cv2.imshow('result', img3)
# cv2.imwrite('result.jpg', img3)
# cv2.waitKey()
# 位于边长为0.2， z = 0平面上底部的正方形
Box = cube_points([0, 0, 0.1], 0.1)
cam1 = Cli.Camera(
numpy.hstack((K, numpy.dot(K, numpy.array([[0], [0], [-1]])))))
box_cam1 = cam1.project(h**o.make_homog(Box[:, :5]))
# 使用H将点变换到第二幅图像中
box_trans = h**o.normalize(numpy.dot(H[0], box_cam1))
# 从cam1和H中计算第二个照相机矩阵
cam2 = Cli.Camera(numpy.dot(H[0], cam1.P))
A = numpy.dot(numpy.linalg.inv(K), cam2.P[:, :3])
A = numpy.array([A[:, 0], A[:, 1], numpy.cross(A[:, 0], A[:, 1])]).T
cam2.P[:, :3] = numpy.dot(K, A)
# 使用第二个照相机矩阵投影
box_cam2 = cam2.project(h**o.make_homog(Box))
# 测试：将点投影在Z=0上，应该能够得到相同点
point = numpy.array([[1, 1, 0, 1]]).T
#限界よりmatch点が多かったら
if num_matches < num_inspect:
num_inspect = num_matches
x1 = []
x2 = []
for i in range(num_inspect):
m = good_matches[i]
k1 = kp1[m.queryIdx]
x1.append(k1.pt)
k2 = kp2[m.trainIdx]
x2.append(k2.pt)
#同次座標系にし、K^-1を使って3次元座標にする。
x1 = np.array(x1).T
x1 = homography.make_homog(x1)
x2 = np.array(x2).T
x2 = homography.make_homog(x2)
x1n = np.linalg.inv(K) @ x1
x2n = np.linalg.inv(K) @ x2
#RANSACでEを推定
model = sfm.RansacModel()
E, inliers = sfm.F_from_ransac(x1n, x2n, model)
#カメラ行列を計算する
P1 = np.array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0]])
P2 = sfm.compute_P_from_essential(E)
tic.k('loaded sifts')
print '{} / {} features'.format(len(d[0]), len(d[1]))
immd5 = md5.md5(''.join(imname)).hexdigest()
matchcache = 'out_ch05ex02_cache_matches_%s.pickle' % immd5
if not os.path.exists(matchcache):
#matches = sift.match(d[0], d[1])
matches = sift.match_twosided(d[0], d[1])
pickle.dump(matches, open(matchcache, 'wb'))
matches = pickle.load(open(matchcache, 'rb'))
tic.k('matched')
ndx = matches.nonzero()[0]
x1 = homography.make_homog(l[0][ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
x2 = homography.make_homog(l[1][ndx2, :2].T)
print '{} matches'.format(len(ndx))
image = [numpy.array(Image.open(name)) for name in imname]
# calibration (FIXME?)
K = camera.my_calibration(image[0].shape[:2])
# Normalize with inv(K) (allows metric reconstruction).
x1n = numpy.dot(numpy.linalg.inv(K), x1)
x2n = numpy.dot(numpy.linalg.inv(K), x2)
tic.k('normalized')
from numpy import * import homography f = array(([1,1], [2,1], [3,2], [4,4])) t = f+1 f = f.T f = homography.make_homog(f) t = t.T t = homography.make_homog(t) #print f, t print homography.H_from_points(f,t)
def convert_points(j):
ndx = matches[j].nonzero()[0]
fp = homography.make_homog(l[j + 1][ndx, :2].T)
ndx2 = [int(matches[j][i]) for i in ndx]
tp = homography.make_homog(l[j][ndx2, :2].T)
return fp, tp
featlist = [imlist[i][:-3] + 'sift' for i in range(imcount)]
with open('vocabulary.pkl', 'rb') as f:
voc = pickle.load(f)
searcher = imagesearch.Searcher('test.db', voc)
query_imid = 50
res_count = 20
res = [w[1] for w in searcher.query(imlist[query_imid])[:res_count]]
print 'regular results for query %d:' % query_imid, res
# Rerank by trying to fit a homography.
q_locs, q_descr = sift.read_features_from_file(featlist[query_imid])
fp = homography.make_homog(q_locs[:, :2].T)
model = homography.RansacModel()
rank = {}
for ndx in res[1:]:
locs, descr = sift.read_features_from_file(featlist[ndx -
1]) # res is 1-based
matches = sift.match(q_descr, descr)
ind = matches.nonzero()[0]
ind2 = [int(matches[i]) for i in ind]
tp = homography.make_homog(locs[:, :2].T)
try:
H, inliers = homography.H_from_ransac(fp[:, ind],
im2_path = './e2.jpg'
im1 = np.array(Image.open(im1_path))
sift.process_image(im1_path, 'im1.sift')
l1, d1 = sift.read_features_from_file('im1.sift')
im2 = np.array(Image.open(im2_path))
sift.process_image(im2_path, 'im2.sift')
l2, d2 = sift.read_features_from_file('im2.sift')
# match features
matches = sift.match_twosided(d1, d2)
ndx = matches.nonzero()[0]
# make homogeneous and normalize with inv(K)
x1 = homography.make_homog(l1[ndx, :2].T)
ndx2 = [int(matches[i]) for i in ndx]
x2 = homography.make_homog(l2[ndx2, :2].T)
x1n = np.dot(np.linalg.inv(K), x1)
x2n = np.dot(np.linalg.inv(K), x2)
# estimate E with RANSAC
model = sfm.RansacModel()
E, inliers = sfm.F_from_ransac(x1n, x2n, model)
# compute camera matrices (P2 will be list of four solutions)
P1 = np.array([[1, 0, 0, 0], [0, 1, 0, 0], [0, 0, 1, 0]])
P2 = sfm.compute_P_from_essential(E)
# pick the solution with points in front of cameras
from pylab import *
import numpy
import camera
import homography
points = homography.make_homog(loadtxt('house.p3d').T)
P = hstack((eye(3), array([[0], [0], [-10]])))
cam = camera.Camera(P)
x = cam.project(points)
figure()
plot(x[0], x[1], 'k.')
r = 0.05 * numpy.random.rand(3)
rot = camera.rotation_matrix(r)
figure()
for t in range(20):
cam.P = dot(cam.P, rot)
x = cam.project(points)
plot(x[0], x[1], 'k.')
show()
