findHomography + perspectiveTransform --> exceeds target Rect

edit

Hi all,

I develop with python, with numpy and openCV (3.1.0).

I have a SRC array of 2D points, which I wish to warp into a rectangle region.

SRC.T = array([[209 210 239 274 307 337 366 404 427 461 484 489 493]
               [330 309 339 304 310 332 353 311 348 324 337 317 296]])

I've calculated the required homography (H) as following:

H, _ = findHomography(rect_src, rect_dst)

Where:

rect_src = array([[209, 282], [209, 330], [484, 337], [493, 296]])
rect_dst = array([[209, 296], [209, 353], [493, 353], [493, 296]])

I got the following homography matrix (which I've verifying manually it makes sense..):

H = [[  5.81755130e-01  -1.21849288e-01   7.21264687e+01]
     [ -1.46583486e-01   6.99605042e-01   5.90321609e+01]
     [ -3.49962434e-04  -5.83010949e-04   1.00000000e+00]]

I've then warped the SRC points with the above homography matrix, hoping all SRC points will utilize the rect_dst region:

DST = perspectiveTransform(SRC, H).astype(int)

I got the following DST points:

DST.T = array([[209 209 236 267 299 330 363 397 429 462 493 493 493]
               [353 327 363 318 324 351 378 320 370 335 353 324 296]])

Unfortunately, the DST points exceed the rect_dst region, so I didn't get the desired outcome...

Attached you may find a debug image, where SRC points are in blue and DST points are in yellow (please ignore the red and the green points).

Can you please advise me where did I go wrong and how to obtain the desired feature?

Thanks ahead, Shahar

C:\fakepath\img_possible_marks_.jpg