| 1 | // |
| 2 | // Copyright 2003 Google, Inc. |
| 3 | // |
| 4 | // |
| 5 | // A simple class to handle 3x3 matrices |
| 6 | // The aim of this class is to be able to manipulate 3x3 matrices |
| 7 | // and 3D vectors as naturally as possible and make calculations |
| 8 | // readable. |
| 9 | // For that reason, the operators +, -, * are overloaded. |
| 10 | // (Reading a = a + b*2 - c is much easier to read than |
| 11 | // a = Sub(Add(a, Mul(b,2)),c) ) |
| 12 | // |
| 13 | // Please be careful about overflows when using those matrices wth integer types |
| 14 | // The calculations are carried with VType. eg : if you are using uint8 as the |
| 15 | // base type, all values will be modulo 256. |
| 16 | // This feature is necessary to use the class in a more general framework with |
| 17 | // VType != plain old data type. |
| 18 | |
| 19 | #ifndef UTIL_MATH_MATRIX3X3_INL_H__ |
| 20 | #define UTIL_MATH_MATRIX3X3_INL_H__ |
| 21 | #include <iostream> |
| 22 | using std::ostream; |
| 23 | using std::cout; |
| 24 | using std::endl; |
| 25 | |
| 26 | |
| 27 | #include "util/math/mathutil.h" |
| 28 | #include "util/math/vector3-inl.h" |
| 29 | #include "util/math/matrix3x3.h" |
| 30 | #include "base/basictypes.h" |
| 31 | #include "base/logging.h" |
| 32 | |
| 33 | template <class VType> |
| 34 | class Matrix3x3 { |
| 35 | private: |
| 36 | VType m_[3][3]; |
| 37 | public: |
| 38 | typedef Matrix3x3<VType> Self; |
| 39 | typedef Vector3<VType> MVector; |
| 40 | // Initialize the matrix to 0 |
| 41 | Matrix3x3() { |
| 42 | m_[0][2] = m_[0][1] = m_[0][0] = VType(); |
| 43 | m_[1][2] = m_[1][1] = m_[1][0] = VType(); |
| 44 | m_[2][2] = m_[2][1] = m_[2][0] = VType(); |
| 45 | } |
| 46 | |
| 47 | // Constructor explicitly setting the values of all the coefficient of |
| 48 | // the matrix |
| 49 | Matrix3x3(const VType &m00, const VType &m01, const VType &m02, |
| 50 | const VType &m10, const VType &m11, const VType &m12, |
| 51 | const VType &m20, const VType &m21, const VType &m22) { |
| 52 | m_[0][0] = m00; |
| 53 | m_[0][1] = m01; |
| 54 | m_[0][2] = m02; |
| 55 | |
| 56 | m_[1][0] = m10; |
| 57 | m_[1][1] = m11; |
| 58 | m_[1][2] = m12; |
| 59 | |
| 60 | m_[2][0] = m20; |
| 61 | m_[2][1] = m21; |
| 62 | m_[2][2] = m22; |
| 63 | } |
| 64 | |
| 65 | // Copy constructor |
| 66 | Matrix3x3(const Self &mb) { |
| 67 | m_[0][0] = mb.m_[0][0]; |
| 68 | m_[0][1] = mb.m_[0][1]; |
| 69 | m_[0][2] = mb.m_[0][2]; |
| 70 | |
| 71 | m_[1][0] = mb.m_[1][0]; |
| 72 | m_[1][1] = mb.m_[1][1]; |
| 73 | m_[1][2] = mb.m_[1][2]; |
| 74 | |
| 75 | m_[2][0] = mb.m_[2][0]; |
| 76 | m_[2][1] = mb.m_[2][1]; |
| 77 | m_[2][2] = mb.m_[2][2]; |
| 78 | } |
| 79 | |
| 80 | // Casting constructor |
| 81 | template <class VType2> |
| 82 | static Self Cast(const Matrix3x3<VType2> &mb) { |
| 83 | return Self(static_cast<VType>(mb(0, 0)), |
| 84 | static_cast<VType>(mb(0, 1)), |
| 85 | static_cast<VType>(mb(0, 2)), |
| 86 | static_cast<VType>(mb(1, 0)), |
| 87 | static_cast<VType>(mb(1, 1)), |
| 88 | static_cast<VType>(mb(1, 2)), |
| 89 | static_cast<VType>(mb(2, 0)), |
| 90 | static_cast<VType>(mb(2, 1)), |
| 91 | static_cast<VType>(mb(2, 2))); |
| 92 | } |
| 93 | |
| 94 | // Change the value of all the coefficients of the matrix |
| 95 | inline Self & |
| 96 | Set(const VType &m00, const VType &m01, const VType &m02, |
| 97 | const VType &m10, const VType &m11, const VType &m12, |
| 98 | const VType &m20, const VType &m21, const VType &m22) { |
| 99 | m_[0][0] = m00; |
| 100 | m_[0][1] = m01; |
| 101 | m_[0][2] = m02; |
| 102 | |
| 103 | m_[1][0] = m10; |
| 104 | m_[1][1] = m11; |
| 105 | m_[1][2] = m12; |
| 106 | |
| 107 | m_[2][0] = m20; |
| 108 | m_[2][1] = m21; |
| 109 | m_[2][2] = m22; |
| 110 | return (*this); |
| 111 | } |
| 112 | |
| 113 | // Copy |
| 114 | inline Self& operator=(const Self &mb) { |
| 115 | m_[0][0] = mb.m_[0][0]; |
| 116 | m_[0][1] = mb.m_[0][1]; |
| 117 | m_[0][2] = mb.m_[0][2]; |
| 118 | |
| 119 | m_[1][0] = mb.m_[1][0]; |
| 120 | m_[1][1] = mb.m_[1][1]; |
| 121 | m_[1][2] = mb.m_[1][2]; |
| 122 | |
| 123 | m_[2][0] = mb.m_[2][0]; |
| 124 | m_[2][1] = mb.m_[2][1]; |
| 125 | m_[2][2] = mb.m_[2][2]; |
| 126 | return (*this); |
| 127 | } |
| 128 | |
| 129 | // Compare |
| 130 | inline bool operator==(const Self &mb) const { |
| 131 | return (m_[0][0] == mb.m_[0][0]) && |
| 132 | (m_[0][1] == mb.m_[0][1]) && |
| 133 | (m_[0][2] == mb.m_[0][2]) && |
| 134 | (m_[1][0] == mb.m_[1][0]) && |
| 135 | (m_[1][1] == mb.m_[1][1]) && |
| 136 | (m_[1][2] == mb.m_[1][2]) && |
| 137 | (m_[2][0] == mb.m_[2][0]) && |
| 138 | (m_[2][1] == mb.m_[2][1]) && |
| 139 | (m_[2][2] == mb.m_[2][2]); |
| 140 | } |
| 141 | |
| 142 | // Matrix addition |
| 143 | inline Self& operator+=(const Self &mb) { |
| 144 | m_[0][0] += mb.m_[0][0]; |
| 145 | m_[0][1] += mb.m_[0][1]; |
| 146 | m_[0][2] += mb.m_[0][2]; |
| 147 | |
| 148 | m_[1][0] += mb.m_[1][0]; |
| 149 | m_[1][1] += mb.m_[1][1]; |
| 150 | m_[1][2] += mb.m_[1][2]; |
| 151 | |
| 152 | m_[2][0] += mb.m_[2][0]; |
| 153 | m_[2][1] += mb.m_[2][1]; |
| 154 | m_[2][2] += mb.m_[2][2]; |
| 155 | return (*this); |
| 156 | } |
| 157 | |
| 158 | // Matrix subtration |
| 159 | inline Self& operator-=(const Self &mb) { |
| 160 | m_[0][0] -= mb.m_[0][0]; |
| 161 | m_[0][1] -= mb.m_[0][1]; |
| 162 | m_[0][2] -= mb.m_[0][2]; |
| 163 | |
| 164 | m_[1][0] -= mb.m_[1][0]; |
| 165 | m_[1][1] -= mb.m_[1][1]; |
| 166 | m_[1][2] -= mb.m_[1][2]; |
| 167 | |
| 168 | m_[2][0] -= mb.m_[2][0]; |
| 169 | m_[2][1] -= mb.m_[2][1]; |
| 170 | m_[2][2] -= mb.m_[2][2]; |
| 171 | return (*this); |
| 172 | } |
| 173 | |
| 174 | // Matrix multiplication by a scalar |
| 175 | inline Self& operator*=(const VType &k) { |
| 176 | m_[0][0] *= k; |
| 177 | m_[0][1] *= k; |
| 178 | m_[0][2] *= k; |
| 179 | |
| 180 | m_[1][0] *= k; |
| 181 | m_[1][1] *= k; |
| 182 | m_[1][2] *= k; |
| 183 | |
| 184 | m_[2][0] *= k; |
| 185 | m_[2][1] *= k; |
| 186 | m_[2][2] *= k; |
| 187 | return (*this); |
| 188 | } |
| 189 | |
| 190 | // Matrix addition |
| 191 | inline Self operator+(const Self &mb) const { |
| 192 | return Self(*this) += mb; |
| 193 | } |
| 194 | |
| 195 | // Matrix subtraction |
| 196 | inline Self operator-(const Self &mb) const { |
| 197 | return Self(*this) -= mb; |
| 198 | } |
| 199 | |
| 200 | // Change the sign of all the coefficients in the matrix |
| 201 | friend inline Self operator-(const Self &vb) { |
| 202 | return Self(-vb.m_[0][0], -vb.m_[0][1], -vb.m_[0][2], |
| 203 | -vb.m_[1][0], -vb.m_[1][1], -vb.m_[1][2], |
| 204 | -vb.m_[2][0], -vb.m_[2][1], -vb.m_[2][2]); |
| 205 | } |
| 206 | |
| 207 | // Matrix multiplication by a scalar |
| 208 | inline Self operator*(const VType &k) const { |
| 209 | return Self(*this) *= k; |
| 210 | } |
| 211 | |
| 212 | friend inline Self operator*(const VType &k, const Self &mb) { |
| 213 | return Self(mb)*k; |
| 214 | } |
| 215 | |
| 216 | // Matrix multiplication |
| 217 | inline Self operator*(const Self &mb) const { |
| 218 | return Self( |
| 219 | m_[0][0] * mb.m_[0][0] + m_[0][1] * mb.m_[1][0] + m_[0][2] * mb.m_[2][0], |
| 220 | m_[0][0] * mb.m_[0][1] + m_[0][1] * mb.m_[1][1] + m_[0][2] * mb.m_[2][1], |
| 221 | m_[0][0] * mb.m_[0][2] + m_[0][1] * mb.m_[1][2] + m_[0][2] * mb.m_[2][2], |
| 222 | |
| 223 | m_[1][0] * mb.m_[0][0] + m_[1][1] * mb.m_[1][0] + m_[1][2] * mb.m_[2][0], |
| 224 | m_[1][0] * mb.m_[0][1] + m_[1][1] * mb.m_[1][1] + m_[1][2] * mb.m_[2][1], |
| 225 | m_[1][0] * mb.m_[0][2] + m_[1][1] * mb.m_[1][2] + m_[1][2] * mb.m_[2][2], |
| 226 | |
| 227 | m_[2][0] * mb.m_[0][0] + m_[2][1] * mb.m_[1][0] + m_[2][2] * mb.m_[2][0], |
| 228 | m_[2][0] * mb.m_[0][1] + m_[2][1] * mb.m_[1][1] + m_[2][2] * mb.m_[2][1], |
| 229 | m_[2][0] * mb.m_[0][2] + m_[2][1] * mb.m_[1][2] + m_[2][2] * mb.m_[2][2]); |
| 230 | } |
| 231 | |
| 232 | // Multiplication of a matrix by a vector |
| 233 | inline MVector operator*(const MVector &v) const { |
| 234 | return MVector( |
| 235 | m_[0][0] * v[0] + m_[0][1] * v[1] + m_[0][2] * v[2], |
| 236 | m_[1][0] * v[0] + m_[1][1] * v[1] + m_[1][2] * v[2], |
| 237 | m_[2][0] * v[0] + m_[2][1] * v[1] + m_[2][2] * v[2]); |
| 238 | } |
| 239 | |
| 240 | // Return the determinant of the matrix |
| 241 | inline VType Det(void) const { |
| 242 | return m_[0][0] * m_[1][1] * m_[2][2] |
| 243 | + m_[0][1] * m_[1][2] * m_[2][0] |
| 244 | + m_[0][2] * m_[1][0] * m_[2][1] |
| 245 | - m_[2][0] * m_[1][1] * m_[0][2] |
| 246 | - m_[2][1] * m_[1][2] * m_[0][0] |
| 247 | - m_[2][2] * m_[1][0] * m_[0][1]; |
| 248 | } |
| 249 | |
| 250 | // Return the trace of the matrix |
| 251 | inline VType Trace(void) const { |
| 252 | return m_[0][0] + m_[1][1] + m_[2][2]; |
| 253 | } |
| 254 | |
| 255 | // Return a pointer to the data array for interface with other libraries |
| 256 | // like opencv |
| 257 | VType* Data() { |
| 258 | return reinterpret_cast<VType*>(m_); |
| 259 | } |
| 260 | const VType* Data() const { |
| 261 | return reinterpret_cast<const VType*>(m_); |
| 262 | } |
| 263 | |
| 264 | // Return matrix element (i,j) with 0<=i<=2 0<=j<=2 |
| 265 | inline VType &operator()(const int i, const int j) { |
| 266 | DCHECK(i >= 0); |
| 267 | DCHECK(i < 3); |
| 268 | DCHECK(j >= 0); |
| 269 | DCHECK(j < 3); |
| 270 | return m_[i][j]; |
| 271 | } |
| 272 | inline VType operator()(const int i, const int j) const { |
| 273 | DCHECK(i >= 0); |
| 274 | DCHECK(i < 3); |
| 275 | DCHECK(j >= 0); |
| 276 | DCHECK(j < 3); |
| 277 | return m_[i][j]; |
| 278 | } |
| 279 | |
| 280 | // Return matrix element (i/3,i%3) with 0<=i<=8 (access concatenated rows). |
| 281 | inline VType &operator[](const int i) { |
| 282 | DCHECK(i >= 0); |
| 283 | DCHECK(i < 9); |
| 284 | return reinterpret_cast<VType*>(m_)[i]; |
| 285 | } |
| 286 | inline VType operator[](const int i) const { |
| 287 | DCHECK(i >= 0); |
| 288 | DCHECK(i < 9); |
| 289 | return reinterpret_cast<const VType*>(m_)[i]; |
| 290 | } |
| 291 | |
| 292 | // Return the transposed matrix |
| 293 | inline Self Transpose(void) const { |
| 294 | return Self(m_[0][0], m_[1][0], m_[2][0], |
| 295 | m_[0][1], m_[1][1], m_[2][1], |
| 296 | m_[0][2], m_[1][2], m_[2][2]); |
| 297 | } |
| 298 | |
| 299 | // Return the transposed of the matrix of the cofactors |
| 300 | // (Useful for inversion for example) |
| 301 | inline Self ComatrixTransposed(void) const { |
| 302 | return Self( |
| 303 | m_[1][1] * m_[2][2] - m_[2][1] * m_[1][2], |
| 304 | m_[2][1] * m_[0][2] - m_[0][1] * m_[2][2], |
| 305 | m_[0][1] * m_[1][2] - m_[1][1] * m_[0][2], |
| 306 | |
| 307 | m_[1][2] * m_[2][0] - m_[2][2] * m_[1][0], |
| 308 | m_[2][2] * m_[0][0] - m_[0][2] * m_[2][0], |
| 309 | m_[0][2] * m_[1][0] - m_[1][2] * m_[0][0], |
| 310 | |
| 311 | m_[1][0] * m_[2][1] - m_[2][0] * m_[1][1], |
| 312 | m_[2][0] * m_[0][1] - m_[0][0] * m_[2][1], |
| 313 | m_[0][0] * m_[1][1] - m_[1][0] * m_[0][1]); |
| 314 | } |
| 315 | // Matrix inversion |
| 316 | inline Self Inverse(void) const { |
| 317 | VType det = Det(); |
| 318 | CHECK(det != 0) << " Can't inverse. Determinant = 0." ; |
| 319 | return (1/det) * ComatrixTransposed(); |
| 320 | } |
| 321 | |
| 322 | // Return the vector 3D at row i |
| 323 | inline MVector Row(const int i) const { |
| 324 | DCHECK(i >= 0); |
| 325 | DCHECK(i < 3); |
| 326 | return MVector(m_[i][0], m_[i][1], m_[i][2]); |
| 327 | } |
| 328 | |
| 329 | // Return the vector 3D at col i |
| 330 | inline MVector Col(const int i) const { |
| 331 | DCHECK(i >= 0); |
| 332 | DCHECK(i < 3); |
| 333 | return MVector(m_[0][i], m_[1][i], m_[2][i]); |
| 334 | } |
| 335 | |
| 336 | // Create a matrix from 3 row vectors |
| 337 | static inline Self FromRows(const MVector &v1, |
| 338 | const MVector &v2, |
| 339 | const MVector &v3) { |
| 340 | Self temp; |
| 341 | temp.Set(v1[0], v1[1], v1[2], |
| 342 | v2[0], v2[1], v2[2], |
| 343 | v3[0], v3[1], v3[2]); |
| 344 | return temp; |
| 345 | } |
| 346 | |
| 347 | // Create a matrix from 3 column vectors |
| 348 | static inline Self FromCols(const MVector &v1, |
| 349 | const MVector &v2, |
| 350 | const MVector &v3) { |
| 351 | Self temp; |
| 352 | temp.Set(v1[0], v2[0], v3[0], |
| 353 | v1[1], v2[1], v3[1], |
| 354 | v1[2], v2[2], v3[2]); |
| 355 | return temp; |
| 356 | } |
| 357 | |
| 358 | // Set the vector in row i to be v1 |
| 359 | void SetRow(int i, const MVector &v1) { |
| 360 | DCHECK(i >= 0); |
| 361 | DCHECK(i < 3); |
| 362 | m_[i][0] = v1[0]; |
| 363 | m_[i][1] = v1[1]; |
| 364 | m_[i][2] = v1[2]; |
| 365 | } |
| 366 | |
| 367 | // Set the vector in column i to be v1 |
| 368 | void SetCol(int i, const MVector &v1) { |
| 369 | DCHECK(i >= 0); |
| 370 | DCHECK(i < 3); |
| 371 | m_[0][i] = v1[0]; |
| 372 | m_[1][i] = v1[1]; |
| 373 | m_[2][i] = v1[2]; |
| 374 | } |
| 375 | |
| 376 | // Return a matrix M close to the original but verifying MtM = I |
| 377 | // (useful to compensate for errors in a rotation matrix) |
| 378 | Self Orthogonalize() const { |
| 379 | MVector r1, r2, r3; |
| 380 | r1 = Row(0).Normalize(); |
| 381 | r2 = (Row(2).CrossProd(r1)).Normalize(); |
| 382 | r3 = (r1.CrossProd(r2)).Normalize(); |
| 383 | return FromRows(r1, r2, r3); |
| 384 | } |
| 385 | |
| 386 | // Return the identity matrix |
| 387 | static inline Self Identity(void) { |
| 388 | Self temp; |
| 389 | temp.Set(1, 0, 0, |
| 390 | 0, 1, 0, |
| 391 | 0, 0, 1); |
| 392 | return temp; |
| 393 | } |
| 394 | |
| 395 | // Return a matrix full of zeros |
| 396 | static inline Self Zero(void) { |
| 397 | return Self(); |
| 398 | } |
| 399 | |
| 400 | // Return a diagonal matrix with the coefficients in v |
| 401 | static inline Self Diagonal(const MVector &v) { |
| 402 | return Self(v[0], VType(), VType(), |
| 403 | VType(), v[1], VType(), |
| 404 | VType(), VType(), v[2]); |
| 405 | } |
| 406 | |
| 407 | // Return the matrix vvT |
| 408 | static Self Sym3(const MVector &v) { |
| 409 | return Self( |
| 410 | v[0]*v[0], v[0]*v[1], v[0]*v[2], |
| 411 | v[1]*v[0], v[1]*v[1], v[1]*v[2], |
| 412 | v[2]*v[0], v[2]*v[1], v[2]*v[2]); |
| 413 | } |
| 414 | // Return a matrix M such that: |
| 415 | // for each u, M * u = v.CrossProd(u) |
| 416 | static Self AntiSym3(const MVector &v) { |
| 417 | return Self(VType(), -v[2], v[1], |
| 418 | v[2], VType(), -v[0], |
| 419 | -v[1], v[0], VType()); |
| 420 | } |
| 421 | |
| 422 | static Self Rodrigues(const MVector &rot) { |
| 423 | Self R; |
| 424 | VType theta = rot.Norm(); |
| 425 | MVector w = rot.Normalize(); |
| 426 | Self Wv = Self::AntiSym3(w); |
| 427 | Self I = Self::Identity(); |
| 428 | Self A = Self::Sym3(w); |
| 429 | R = (1 - cos(theta)) * A + sin(theta) * Wv + cos(theta) * I; |
| 430 | return R; |
| 431 | } |
| 432 | |
| 433 | // Returns v.Transpose() * (*this) * u |
| 434 | VType MulBothSides(const MVector &v, const MVector &u) const { |
| 435 | return ((*this) * u).DotProd(v); |
| 436 | } |
| 437 | |
| 438 | // Use the 3x3 matrix as a projective transform for 2d points |
| 439 | Vector2<VType> Project(const Vector2<VType> &v) const { |
| 440 | MVector temp = (*this) * MVector(v[0], v[1], 1); |
| 441 | return Vector2<VType>(temp[0] / temp[2], temp[1] / temp[2]); |
| 442 | } |
| 443 | |
| 444 | // Return the Frobenius norm of the matrix: sqrt(sum(aij^2)) |
| 445 | VType FrobeniusNorm() const { |
| 446 | VType sum = VType(); |
| 447 | for (int i = 0; i < 3; i++) { |
| 448 | for (int j = 0; j < 3; j++) { |
| 449 | sum += m_[i][j] * m_[i][j]; |
| 450 | } |
| 451 | } |
| 452 | return sqrt(sum); |
| 453 | } |
| 454 | |
| 455 | // Finds the eigen values of the matrix. Return the number of real eigenvalues |
| 456 | // found |
| 457 | int EigenValues(MVector *eig_val) { |
| 458 | long double r1, r2, r3; // NOLINT |
| 459 | // characteristic polynomial is |
| 460 | // x^3 + (a11*a22+a22*a33+a33*a11)*x^2 - trace(A)*x - det(A) |
| 461 | VType a = -Trace(); |
| 462 | VType b = m_[0][0]*m_[1][1] + m_[1][1]*m_[2][2] + m_[2][2]*m_[0][0] |
| 463 | - m_[1][0]*m_[0][1] - m_[2][1]*m_[1][2] - m_[0][2]*m_[2][0]; |
| 464 | VType c = -Det(); |
| 465 | bool res = MathUtil::RealRootsForCubic(a, b, c, &r1, &r2, &r3); |
| 466 | (*eig_val)[0] = r1; |
| 467 | if (res) { |
| 468 | (*eig_val)[1] = r2; |
| 469 | (*eig_val)[2] = r3; |
| 470 | return 3; |
| 471 | } |
| 472 | return 1; |
| 473 | } |
| 474 | |
| 475 | // Finds the eigen values and associated eigen vectors of a |
| 476 | // symmetric positive definite 3x3 matrix,eigen values are |
| 477 | // sorted in decreasing order, eig_val corresponds to the |
| 478 | // columns of the eig_vec matrix. |
| 479 | // Note: The routine will only use the lower diagonal part |
| 480 | // of the matrix, i.e. |
| 481 | // | a00, | |
| 482 | // | a10, a11, | |
| 483 | // | a20, a21, a22 | |
| 484 | void SymmetricEigenSolver(MVector *eig_val, |
| 485 | Self *eig_vec ) { |
| 486 | // Compute characteristic polynomial coefficients |
| 487 | double c2 = -(m_[0][0] + m_[1][1] + m_[2][2]); |
| 488 | double c1 = -(m_[1][0] * m_[1][0] - m_[0][0] * m_[1][1] |
| 489 | - m_[0][0] * m_[2][2] - m_[1][1] * m_[2][2] |
| 490 | + m_[2][0] * m_[2][0] + m_[2][1] * m_[2][1]); |
| 491 | double c0 = -(m_[0][0] * m_[1][1] * m_[2][2] - m_[2][0] |
| 492 | * m_[2][0] * m_[1][1] - m_[1][0] * m_[1][0] |
| 493 | * m_[2][2] - m_[0][0] * m_[2][1] * m_[2][1] |
| 494 | + 2 * m_[1][0] * m_[2][0] * m_[2][1]); |
| 495 | |
| 496 | // Root finding |
| 497 | double q = (c2*c2-3*c1)/9.0; |
| 498 | double r = (2*c2*c2*c2-9*c2*c1+27*c0)/54.0; |
| 499 | // Assume R^3 <Q^3 so there are three real roots |
| 500 | if (q < 0) q = 0; |
| 501 | double sqrt_q = -2.0 * sqrt(q); |
| 502 | double theta = acos(r / sqrt(q * q * q)); |
| 503 | double c2_3 = c2 / 3; |
| 504 | (*eig_val)[0] = sqrt_q * cos(theta / 3.0) - c2_3; |
| 505 | (*eig_val)[1] = sqrt_q * cos((theta + 2.0 * M_PI)/3.0) - c2_3; |
| 506 | (*eig_val)[2] = sqrt_q * cos((theta - 2.0 * M_PI)/3.0) - c2_3; |
| 507 | |
| 508 | // Sort eigen value in decreasing order |
| 509 | Vector3<int> d_order = eig_val->ComponentOrder(); |
| 510 | (*eig_val) = MVector((*eig_val)[d_order[2]], |
| 511 | (*eig_val)[d_order[1]], |
| 512 | (*eig_val)[d_order[0]]); |
| 513 | // Compute eigenvectors |
| 514 | for (int i = 0; i < 3; ++i) { |
| 515 | MVector r1 , r2 , r3 , e1 , e2 , e3; |
| 516 | r1[0] = m_[0][0] - (*eig_val)[i]; |
| 517 | r2[0] = m_[1][0]; |
| 518 | r3[0] = m_[2][0]; |
| 519 | r1[1] = m_[1][0]; |
| 520 | r2[1] = m_[1][1] - (*eig_val)[i]; |
| 521 | r3[1] = m_[2][1]; |
| 522 | r1[2] = m_[2][0]; |
| 523 | r2[2] = m_[2][1]; |
| 524 | r3[2] = m_[2][2] - (*eig_val)[i]; |
| 525 | |
| 526 | e1 = r1.CrossProd(r2); |
| 527 | e2 = r2.CrossProd(r3); |
| 528 | e3 = r3.CrossProd(r1); |
| 529 | |
| 530 | // Make e2 and e3 point in the same direction as e1 |
| 531 | if (e2.DotProd(e1) < 0) e2 = -e2; |
| 532 | if (e3.DotProd(e1) < 0) e3 = -e3; |
| 533 | MVector e = (e1 + e2 + e3).Normalize(); |
| 534 | eig_vec->SetCol(i,e); |
| 535 | } |
| 536 | } |
| 537 | |
| 538 | // Return true is one of the elements of the matrix is NaN |
| 539 | bool IsNaN() const { |
| 540 | for ( int i = 0; i < 3; ++i ) { |
| 541 | for ( int j = 0; j < 3; ++j ) { |
| 542 | if ( isnan(m_[i][j]) ) { |
| 543 | return true; |
| 544 | } |
| 545 | } |
| 546 | } |
| 547 | return false; |
| 548 | } |
| 549 | |
| 550 | friend std::ostream &operator <<(std::ostream &out, const Self &mb) { |
| 551 | int i, j; |
| 552 | for (i = 0; i < 3; i++) { |
| 553 | if (i ==0) { |
| 554 | out << "[" ; |
| 555 | } else { |
| 556 | out << " " ; |
| 557 | } |
| 558 | for (j = 0; j < 3; j++) { |
| 559 | out << mb(i, j) << " " ; |
| 560 | } |
| 561 | if (i == 2) { |
| 562 | out << "]" ; |
| 563 | } else { |
| 564 | out << endl; |
| 565 | } |
| 566 | } |
| 567 | return out; |
| 568 | } |
| 569 | }; |
| 570 | |
| 571 | // TODO(user): Matrix3x3<T> does not actually satisfy the definition of a |
| 572 | // POD type even when T is a POD. Pretending that Matrix3x3<T> is a POD |
| 573 | // probably won't cause immediate problems, but eventually this should be fixed. |
| 574 | PROPAGATE_POD_FROM_TEMPLATE_ARGUMENT(Matrix3x3); |
| 575 | |
| 576 | #endif // UTIL_MATH_MATRIX3X3_INL_H__ |
| 577 | |