camera_calibration.py 10.5 KB
Newer Older
wester committed
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 99 100 101 102 103 104 105 106 107 108 109 110 111 112 113 114 115 116 117 118 119 120 121 122 123 124 125 126 127 128 129 130 131 132 133 134 135 136 137 138 139 140 141 142 143 144 145 146 147 148 149 150 151 152 153 154 155 156 157 158 159 160 161 162 163 164 165 166 167 168 169 170 171 172 173 174 175 176 177 178 179 180 181 182 183 184 185 186 187 188 189 190 191 192 193 194 195 196 197 198 199 200 201 202 203 204 205 206 207 208 209 210 211 212 213 214 215 216 217 218 219 220 221 222 223 224 225 226 227 228 229 230 231 232 233 234 235 236 237 238 239 240 241 242 243 244 245 246 247 248 249 250 251 252 253 254 255 256 257 258 259 260 261 262 263 264 265 266 267 268 269 270 271 272 273 274 275 276 277 278 279 280 281 282 283 284 285 286 287 288 289 290 291 292 293 294 295 296 297 298 299 300 301 302 303 304 305 306 307 308 309 310 311 312 313 314 315 316 317 318 319 320 321 322 323 324 325 326 327 328 329 330 331 332 333 334 335 336 337 338 339 340 341 342 343 344 345 346 347 348 349 350 351 352 353 354 355 356 357 358 359 360
#!/usr/bin/env python

import sys
import math
import time
import random

import numpy
import transformations
import cv2.cv as cv

def clamp(a, x, b):
    return numpy.maximum(a, numpy.minimum(x, b))

def norm(v):
    mag = numpy.sqrt(sum([e * e for e in v]))
    return v / mag

class Vec3:
    def __init__(self, x, y, z):
        self.v = (x, y, z)
    def x(self):
        return self.v[0]
    def y(self):
        return self.v[1]
    def z(self):
        return self.v[2]
    def __repr__(self):
        return "<Vec3 (%s,%s,%s)>" % tuple([repr(c) for c in self.v])
    def __add__(self, other):
        return Vec3(*[self.v[i] + other.v[i] for i in range(3)])
    def __sub__(self, other):
        return Vec3(*[self.v[i] - other.v[i] for i in range(3)])
    def __mul__(self, other):
        if isinstance(other, Vec3):
            return Vec3(*[self.v[i] * other.v[i] for i in range(3)])
        else:
            return Vec3(*[self.v[i] * other for i in range(3)])
    def mag2(self):
        return sum([e * e for e in self.v])
    def __abs__(self):
        return numpy.sqrt(sum([e * e for e in self.v]))
    def norm(self):
        return self * (1.0 / abs(self))
    def dot(self, other):
        return sum([self.v[i] * other.v[i] for i in range(3)])
    def cross(self, other):
        (ax, ay, az) = self.v
        (bx, by, bz) = other.v
        return Vec3(ay * bz - by * az, az * bx - bz * ax, ax * by - bx * ay)


class Ray:

    def __init__(self, o, d):
        self.o = o
        self.d = d

    def project(self, d):
        return self.o + self.d * d

class Camera:

    def __init__(self, F):
        R = Vec3(1., 0., 0.)
        U = Vec3(0, 1., 0)
        self.center = Vec3(0, 0, 0)
        self.pcenter = Vec3(0, 0, F)
        self.up = U
        self.right = R

    def genray(self, x, y):
        """ -1 <= y <= 1 """
        r = numpy.sqrt(x * x + y * y)
        if 0:
            rprime = r + (0.17 * r**2)
        else:
            rprime = (10 * numpy.sqrt(17 * r + 25) - 50) / 17
        print "scale", rprime / r
        x *= rprime / r
        y *= rprime / r
        o = self.center
        r = (self.pcenter + (self.right * x) + (self.up * y)) - o
        return Ray(o, r.norm())

class Sphere:

    def __init__(self, center, radius):
        self.center = center
        self.radius = radius

    def hit(self, r):
        # a = mag2(r.d)
        a = 1.
        v = r.o - self.center
        b = 2 * r.d.dot(v)
        c = self.center.mag2() + r.o.mag2() + -2 * self.center.dot(r.o) - (self.radius ** 2)
        det = (b * b) - (4 * c)
        pred = 0 < det

        sq = numpy.sqrt(abs(det))
        h0 = (-b - sq) / (2)
        h1 = (-b + sq) / (2)

        h = numpy.minimum(h0, h1)

        pred = pred & (h > 0)
        normal = (r.project(h) - self.center) * (1.0 / self.radius)
        return (pred, numpy.where(pred, h, 999999.), normal)

def pt2plane(p, plane):
    return p.dot(plane) * (1. / abs(plane))

class Plane:

    def __init__(self, p, n, right):
        self.D = -pt2plane(p, n)
        self.Pn = n
        self.right = right
        self.rightD = -pt2plane(p, right)
        self.up = n.cross(right)
        self.upD = -pt2plane(p, self.up)

    def hit(self, r):
        Vd = self.Pn.dot(r.d)
        V0 = -(self.Pn.dot(r.o) + self.D)
        h = V0 / Vd
        pred = (0 <= h)

        return (pred, numpy.where(pred, h, 999999.), self.Pn)

    def localxy(self, loc):
        x = (loc.dot(self.right) + self.rightD)
        y = (loc.dot(self.up) + self.upD)
        return (x, y)

# lena = numpy.fromstring(cv.LoadImage("../samples/c/lena.jpg", 0).tostring(), numpy.uint8) / 255.0

def texture(xy):
    x,y = xy
    xa = numpy.floor(x * 512)
    ya = numpy.floor(y * 512)
    a = (512 * ya) + xa
    safe = (0 <= x) & (0 <= y) & (x < 1) & (y < 1)
    if 0:
        a = numpy.where(safe, a, 0).astype(numpy.int)
        return numpy.where(safe, numpy.take(lena, a), 0.0)
    else:
        xi = numpy.floor(x * 11).astype(numpy.int)
        yi = numpy.floor(y * 11).astype(numpy.int)
        inside = (1 <= xi) & (xi < 10) & (2 <= yi) & (yi < 9)
        checker = (xi & 1) ^ (yi & 1)
        final = numpy.where(inside, checker, 1.0)
        return numpy.where(safe, final, 0.5)

def under(vv, m):
    return Vec3(*(numpy.dot(m, vv.v + (1,))[:3]))

class Renderer:

    def __init__(self, w, h, oversample):
        self.w = w
        self.h = h

        random.seed(1)
        x = numpy.arange(self.w*self.h) % self.w
        y = numpy.floor(numpy.arange(self.w*self.h) / self.w)
        h2 = h / 2.0
        w2 = w / 2.0
        self.r = [ None ] * oversample
        for o in range(oversample):
            stoch_x = numpy.random.rand(self.w * self.h)
            stoch_y = numpy.random.rand(self.w * self.h)
            nx = (x + stoch_x - 0.5 - w2) / h2
            ny = (y + stoch_y - 0.5 - h2) / h2
            self.r[o] = cam.genray(nx, ny)

        self.rnds = [random.random() for i in range(10)]

    def frame(self, i):

        rnds = self.rnds
        roll = math.sin(i * .01 * rnds[0] + rnds[1])
        pitch = math.sin(i * .01 * rnds[2] + rnds[3])
        yaw = math.pi * math.sin(i * .01 * rnds[4] + rnds[5])
        x = math.sin(i * 0.01 * rnds[6])
        y = math.sin(i * 0.01 * rnds[7])

        x,y,z = -0.5,0.5,1
        roll,pitch,yaw = (0,0,0)

        z = 4 + 3 * math.sin(i * 0.1 * rnds[8])
        print z

        rz = transformations.euler_matrix(roll, pitch, yaw)
        p = Plane(Vec3(x, y, z), under(Vec3(0,0,-1), rz), under(Vec3(1, 0, 0), rz))

        acc = 0
        for r in self.r:
            (pred, h, norm) = p.hit(r)
            l = numpy.where(pred, texture(p.localxy(r.project(h))), 0.0)
            acc += l
        acc *= (1.0 / len(self.r))

        # print "took", time.time() - st

        img = cv.CreateMat(self.h, self.w, cv.CV_8UC1)
        cv.SetData(img, (clamp(0, acc, 1) * 255).astype(numpy.uint8).tostring(), self.w)
        return img

#########################################################################

num_x_ints = 8
num_y_ints = 6
num_pts = num_x_ints * num_y_ints

def get_corners(mono, refine = False):
    (ok, corners) = cv.FindChessboardCorners(mono, (num_x_ints, num_y_ints), cv.CV_CALIB_CB_ADAPTIVE_THRESH | cv.CV_CALIB_CB_NORMALIZE_IMAGE)
    if refine and ok:
        corners = cv.FindCornerSubPix(mono, corners, (5,5), (-1,-1), ( cv.CV_TERMCRIT_EPS+cv.CV_TERMCRIT_ITER, 30, 0.1 ))
    return (ok, corners)

def mk_object_points(nimages, squaresize = 1):
    opts = cv.CreateMat(nimages * num_pts, 3, cv.CV_32FC1)
    for i in range(nimages):
        for j in range(num_pts):
            opts[i * num_pts + j, 0] = (j / num_x_ints) * squaresize
            opts[i * num_pts + j, 1] = (j % num_x_ints) * squaresize
            opts[i * num_pts + j, 2] = 0
    return opts

def mk_image_points(goodcorners):
    ipts = cv.CreateMat(len(goodcorners) * num_pts, 2, cv.CV_32FC1)
    for (i, co) in enumerate(goodcorners):
        for j in range(num_pts):
            ipts[i * num_pts + j, 0] = co[j][0]
            ipts[i * num_pts + j, 1] = co[j][1]
    return ipts

def mk_point_counts(nimages):
    npts = cv.CreateMat(nimages, 1, cv.CV_32SC1)
    for i in range(nimages):
        npts[i, 0] = num_pts
    return npts

def cvmat_iterator(cvmat):
    for i in range(cvmat.rows):
        for j in range(cvmat.cols):
            yield cvmat[i,j]

cam = Camera(3.0)
rend = Renderer(640, 480, 2)
cv.NamedWindow("snap")

#images = [rend.frame(i) for i in range(0, 2000, 400)]
images = [rend.frame(i) for i in [1200]]

if 0:
    for i,img in enumerate(images):
        cv.SaveImage("final/%06d.png" % i, img)

size = cv.GetSize(images[0])
corners = [get_corners(i) for i in images]

goodcorners = [co for (im, (ok, co)) in zip(images, corners) if ok]

def checkerboard_error(xformed):
    def pt2line(a, b, c):
        x0,y0 = a
        x1,y1 = b
        x2,y2 = c
        return abs((x2 - x1) * (y1 - y0) - (x1 - x0) * (y2 - y1)) / math.sqrt((x2 - x1) ** 2 + (y2 - y1) ** 2)
    errorsum = 0.
    for im in xformed:
        for row in range(6):
            l0 = im[8 * row]
            l1 = im[8 * row + 7]
            for col in range(1, 7):
                e = pt2line(im[8 * row + col], l0, l1)
                #print "row", row, "e", e
                errorsum += e

    return errorsum

if True:
    from scipy.optimize import fmin

    def xf(pt, poly):
        x, y = pt
        r = math.sqrt((x - 320) ** 2 + (y - 240) ** 2)
        fr = poly(r) / r
        return (320 + (x - 320) * fr, 240 + (y - 240) * fr)
    def silly(p, goodcorners):
    #    print "eval", p

        d = 1.0 # - sum(p)
        poly = numpy.poly1d(list(p) + [d, 0.])

        xformed = [[xf(pt, poly) for pt in co] for co in goodcorners]

        return checkerboard_error(xformed)

    x0 = [ 0. ]
    #print silly(x0, goodcorners)
    print "initial error", silly(x0, goodcorners)
    xopt = fmin(silly, x0, args=(goodcorners,))
    print "xopt", xopt
    print "final error", silly(xopt, goodcorners)

    d = 1.0 # - sum(xopt)
    poly = numpy.poly1d(list(xopt) + [d, 0.])
    print "final polynomial"
    print poly

    for co in goodcorners:
        scrib = cv.CreateMat(480, 640, cv.CV_8UC3)
        cv.SetZero(scrib)
        cv.DrawChessboardCorners(scrib, (num_x_ints, num_y_ints), [xf(pt, poly) for pt in co], True)
        cv.ShowImage("snap", scrib)
        cv.WaitKey()

    sys.exit(0)

for (i, (img, (ok, co))) in enumerate(zip(images, corners)):
    scrib = cv.CreateMat(img.rows, img.cols, cv.CV_8UC3)
    cv.CvtColor(img, scrib, cv.CV_GRAY2BGR)
    if ok:
        cv.DrawChessboardCorners(scrib, (num_x_ints, num_y_ints), co, True)
    cv.ShowImage("snap", scrib)
    cv.WaitKey()

print len(goodcorners)
ipts = mk_image_points(goodcorners)
opts = mk_object_points(len(goodcorners), .1)
npts = mk_point_counts(len(goodcorners))

intrinsics = cv.CreateMat(3, 3, cv.CV_64FC1)
distortion = cv.CreateMat(4, 1, cv.CV_64FC1)
cv.SetZero(intrinsics)
cv.SetZero(distortion)
# focal lengths have 1/1 ratio
intrinsics[0,0] = 1.0
intrinsics[1,1] = 1.0
cv.CalibrateCamera2(opts, ipts, npts,
           cv.GetSize(images[0]),
           intrinsics,
           distortion,
           cv.CreateMat(len(goodcorners), 3, cv.CV_32FC1),
           cv.CreateMat(len(goodcorners), 3, cv.CV_32FC1),
           flags = 0) # cv.CV_CALIB_ZERO_TANGENT_DIST)
print "D =", list(cvmat_iterator(distortion))
print "K =", list(cvmat_iterator(intrinsics))
mapx = cv.CreateImage((640, 480), cv.IPL_DEPTH_32F, 1)
mapy = cv.CreateImage((640, 480), cv.IPL_DEPTH_32F, 1)
cv.InitUndistortMap(intrinsics, distortion, mapx, mapy)
for img in images:
    r = cv.CloneMat(img)
    cv.Remap(img, r, mapx, mapy)
    cv.ShowImage("snap", r)
    cv.WaitKey()