Why is cv2.projectPoints not behaving as I expect?

Question

I have a 3D world that I am trying to map to a 2D view using cv2.projectPoints, but it is not acting as I expect. My grasp of opencv, numpy and matrix ops is weak, so I must be making a wrong assumption somewhere. This code:

src = np.ones((6, 3))
src[:,1] = 2
src[:,2] = range(6) # source points
rvec = np.array([0,0,0], np.float) # rotation vector
tvec = np.array([0,0,0], np.float) # translation vector
fx = fy = 1.0
cx = cy = 0.0
cameraMatrix = np.array([[fx,0,cx],[0,fy,cy],[0,0,1]])
result = cv2.projectPoints(src, rvec, tvec, cameraMatrix, None)
for n in range(len(src)):
    print src[n], '==>', result[0][n]

generates this output:

[ 1.  2.  0.] ==> [[ 1.  2.]]
[ 1.  2.  1.] ==> [[ 1.  2.]]
[ 1.  2.  2.] ==> [[ 0.5  1. ]]
[ 1.  2.  3.] ==> [[ 0.33333333  0.66666667]]
[ 1.  2.  4.] ==> [[ 0.25  0.5 ]]
[ 1.  2.  5.] ==> [[ 0.2  0.4]]

The x- and y-values are being divided by the z-value?! I thought that as I am applying no transformations with rvec and tvec that the output should match the [x,y] values of src.

Solution

For no operation the rotation matrix should be an identity

1 0 0
0 1 0
0 0 1

edit: It's a very simple function to write yourself.

foreach input 3dpoint xyz
  3dpoint tmp = cameraintrinsic(3x3) * rotationvect(1x3) * xyz(3x1) + cameraintrinsic(3x3)*translation(3,1)
  2dpoint screen = tmp.x/tmp.z, tmp.y / tmp.z

The /tmp.z is because the result is returned in homogeneous coords = essentially 2d coords with a scaling factor.