2

Kinect を使用してモーションをキャプチャし、モーションのパスを保存して、作成した画像をテキストに変換するプログラムを書きたいと思います。

おそらく画像の品質は非常に低いため、OCRが最適な方法であるかどうかはわかりません。

誰かが提案できるシステムはありますか？また、それを達成する方法に関するヒント。乾杯

編集：

Windows Journal は、私が望んでいることとほとんど同じです。

4

2 に答える 2

1

I fail to see the advantage that the Kinect can give you here.

You have to define first what kind of input this system will have: Will it be the RGB image of a character for example or something more complicated like tracked hand position through time?

If you want to use the Kinect for its depth map, then what you are trying to do seems more like gesture recognition (and mapping of each gesture to a character) to me rather than OCR.

Edit for the gesture recognition way:

I haven't implemented anything myself on that, so I can not really give you clear directions or proposals. There exist a couple of papers on the subject (google scholar search) that take advantage of the available depth info rather than working with 2D images. I would guess that the DIY way would be to get as many training samples as possible for each gesture, train your classifier (wikipedia on supervised learning) and then test your new samples.

If you don't feel so adventurous, you could start by testing some code samples that exist:

  • The hand tracking sample from OpenNI -you can see an example here, I think it recognizes some gestures too- .
  • Candescent NUI "Hand and finger tracking with Kinect depth data", I don't think it includes any recognition, but maybe it can help you get better data.
  • The Kinect for Windows Developer Kit could contain something useful, but to tell you the truth I haven't checked it out.

I can see the gesture recognition approach being very tricky though. There has to be a significant variation between the gestures, otherwise, taking into acount the Kinect low resolution and noise issues, even if you get a very large number of training samples, it could still fail. Imagine writing 'i' or 'l' for example. Or the difference between 'o' and 'a' (depending on your handwriting of course). The scale is so small and the moving of hands so trivial that I don't know if a classifier -even a powerful one- could catch that kind of variations.

于 2012-10-12T15:29:55.840 に答える
1

Kinect のようなセンサーが必要で、その API を使用してモーションをキャプチャします。これは、センサーとモーション キャプチャ テクノロジに固有のものです。C# を使用して、サード パーティの API なしでモーション キャプチャ テクノロジを作成することはできないと思います。

于 2012-10-12T00:29:43.040 に答える