I'm looking for an OCR technique (PCA or SVM or anything else) in a peculiar setting. I want to detect the motion of the finger so that if someone writes something in front of the camera in the air, I want to recognize the characters online (meaning as soon as they are written).
There is a paper "Using Mobile Phones to Write in Air". It basically uses the mobile sensors and machine learning to detect words. The Results show that English characters can be identiﬁed with an average accuracy of 91:9%.
There are some open source projects like "hand-gesture-detection" on Google code. I think you can use them as the start for your project.
What you describe is a whole research field, and IMHO a bit out of scope here, but I'll try to provide some pointers anyway.
You should be able to find some review papers on Google scholar that can help you deciding on a technique to use.
Personally I would advise that you divide the task in two: first, detect the path that the finger takes (computer vision), then recognise that path as a character ("handwritten" character recognition).
Now bouncing on the techniques you mention in your question, I want to point out that the choice of the representation and the choice of the features to use is as important (if not more) as the technique itself. Also, PCA is not a classifier, but more of a tool to pre-process your features.
For recognizing hand written letters, you might want to have a look at the detexify website, which is a neat tool for recognizing LaTeX Symbols you write on the screen. On this site, you can also find links to Github with the source code.