Technically speaking, what you are asking to do is impossible. Fortunately, I do imposible things all the time.
I cannot spoon feed you a solution, but I can give you a rough map to follow. First, let's set some boundries to the problem. It cannot be "any light condition". You have to address conditions of too dark (near dark), and too bright (saturation) with a simple error, and give up on those.
If you have unlimited time to accomplish the inspection (on the order of a few seconds), and if your camera supports it, you can perform an analysis of the scene at multiple exposure times, and select the exposure with the best contrast to begin you OCR.
The NI OCR stuff uses gemotric match (I think), so you will need to train each font that you want to use. Each character will need multiple templates. It is possible to build intraining into your app, so that you are prompted to fill in the blank when an unknown character is found.
You can also perform the OCR on multiple exposure images, and do dome logical analysisto determine which result is the most correct.
You can also use a dictionary list if the words will be limited in some predictable way.
If you are going to rely soley on ambient light available, I can tell you that you should give up right now. For machine vision to do its job, it needs good cameras, good lenses, good lighting, and good fixturing. Each time you take one of these away, you reduce your chance of success by 24.25%.
