One or more cascades for detect a object?

edit

Hi. I try to make a hand gestures detector. But better to go slowly... i dont know if is better one cascade for detect one hand in any position, or use a cascade for each position.

To probe this, i make a pen detector. For a single pen position, the cascade works fine, his parameters are i.e. -w 60 and -h 18

If i want to detect the pen in any position. I resize the png's to 60x60 with transparent background, then pass to opencv_createsamples and finally the samples looks good. Now the parameters are -w 60 and -h 60 But now the cascade no detect nothing, not even samples.

I must use one cascade for each position? If i want to detect the pen in diagonal, what -w -h parameters i need?

Thanks and sorry for my bad english.

Edit: whats better for hand gestures detection? Detect the region of the hand and then treat only this region, or treat directly all the given image. The cascade detection process maybe consume a lot time, i think..