> так я и не понял, написал он свою библиотеку; пользуется OpenCV (пропаченной лично? или не измененной?); что ту непонятного - написал новый алгоритм, с импользованием библиотеки. так вообще всегда делается, нет?
>в чем профит, если с OpenVC можно смастерить такую жe поделку??? или низя?
в том что дурость каждого комментатора можно легко оценить по его комментам к этой новости
есть например вот такие
This is very cool. Having worked on image tracking and recognition software, I know just how difficult that can be (Kinect was an awesome step forward in consumer motion tracking).
Estimating from the coordinates on screen he is processign a 320×256 (or so) pixel image at better than 15 frames per second when it is focussed on his eye.
The finger tracking was done at about 10 frames per second and the face tracking at around 7-8 frames per second.
Picking out his face from a picture was awesome – scale invariance is pretty tough, although we have some good algorithms for that (SIFT and SURF).
Handling changes in lighting is still a weak point (when the car enters the shadow at around 3:18) tracking is lost.
Tracking the panda was awesome, the shape and coloration changed while it moved. Nice!
Liked the line tracking / drawing feature. Though it seems you can defeat the tracking by quickly whisking the object out of the frame (3:35) – and, yes, he did move the pen point outside the frame.
Looked at some more of his videos on YouTube and tracking limitations are more obvious: does a good job tracking the face as a person turns (at least until the back of the head shows up); blocking the face with hands also inhibits the tracking (though, removing glasses did not – not sure if this is because it was just recognizing any face or a specific face – a good test would have been to have someone else walk into the frame); use on cafeteria surveillance footage also exposes limitations – loss of tracking, although no false tracking, took a long time to recover the face once the woman sat down to eat.
а есть такие как у вас
