If it is not possible to store all of those negative images on your system then you need to use a system with more resources, or you need to choose a different topic. The topic you have chosen requires large large amounts of data because there are so many different things that we need to recognize. For example from where I am in bed, without putting on my glasses (I am fairly short sighted) I can identify 31 different types of object in my view.
One thing to keep in mind is that the information required to train your identification system is not the same as the information you require to classify an image after you have done the training. During the training phase you require complete images in order to extract features from them, but those features can be recorded and analyzed in a second phase that does not need the images anymore.
