Question

我想使用numpy / tensorflow实现cv2.solvePnP和cv2.projectPoints。但有意想不到的价值。我引用了这个公式：How to change text of a TextView in navigation drawer header?

来自formula

我的参数是：

OBJ_POINTS=np.array([[ 6.825897,  1.330353, -1.330353, -6.825897,  5.311432,  1.78993 ,
    -1.78993 , -5.311432,  2.005628, -2.005628,  2.774015, -2.774015,
     0.      ,  0.      ],
   [ 6.760612,  7.122144,  7.122144,  6.760612,  5.485328,  5.393625,
     5.393625,  5.485328,  1.409845,  1.409845, -2.080775, -2.080775,
    -3.116408, -7.415691],
   [ 4.402142,  6.903745,  6.903745,  4.402142,  3.987654,  4.413414,
     4.413414,  3.987654,  6.165652,  6.165652,  5.048531,  5.048531,
     6.097667,  4.070434],
   [ 1.      ,  1.      ,  1.      ,  1.      ,  1.      ,  1.      ,
     1.      ,  1.      ,  1.      ,  1.      ,  1.      ,  1.      ,
     1.      ,  1.      ]])
label=np.array([[391., 206.],
   [430., 197.],
   [459., 191.],
   [479., 195.],
   [408., 219.],
   [428., 217.],
   [457., 213.],
   [472., 211.],
   [441., 255.],
   [459., 253.],
   [430., 277.],
   [466., 275.],
   [452., 288.],
   [446., 312.]], dtype=float32)
cam_mat=np.array([[1.28e+03, 0.00e+00, 6.40e+02],
   [0.00e+00, 1.28e+03, 3.60e+02],
   [0.00e+00, 0.00e+00, 1.00e+00]])

我使用cv2.solvePnP计算了rvec和tvec：

rvec=np.array([[ 0.14683954],
   [-0.77872218],
   [-0.1339505 ]])
tvec=array([[  27.37206706],
   [  12.89076055],
   [-160.68589399]])

然后我使用cv2.Rodrigues(rvec)[0]

计算了旋转矩阵

rmat=np.array([[ 0.70427006,  0.06582738, -0.70687371],
    [-0.17415072,  0.98128827, -0.08212712],
    [ 0.68824066,  0.18094224,  0.70255583]])

所以矩阵[R | t]是：

Rt=np.concatenate([rmat,translation_vector],axis=-1)

我通过以下方式计算重新投影的地标：

y=(cam_mat.dot(Rt)).dot(obj_pts)

得到了：

y=np.array([[-59293.93250522, -67734.92848781, -71305.43549652,
    -77613.81683607, -61392.72974225, -66330.50940479,
    -71134.46565557, -75647.97266068, -67635.61225272,
    -73018.47592188, -66794.35222396, -74239.47393207,
    -71201.63145976, -71139.03429556],
   [-31594.43367234, -30883.63615633, -30949.7645763 ,
    -31933.73142526, -33378.22414678, -33523.97724932,
    -33612.95005786, -33642.24174657, -38522.96324169,
    -38622.65784207, -43280.73944318, -43418.62858157,
    -44562.89054421, -50542.67218909],
   [  -151.67200329,   -153.63132795,   -155.46253401,
      -161.06772304,   -153.23627339,   -155.37738704,
      -157.84119226,   -160.54736035,   -154.71872396,
      -157.47943345,   -155.60632927,   -159.42470911,
      -156.96583233,   -159.1679986 ]])

那是出乎意料的......

使用numpy实现cv :: projectpoints得到了意想不到的结果

0 个答案: