机器人视觉——是否一个简单的问题？（英文）

beijinren

对人类来说，90%的信息来自于视觉。就算对一个只有3岁的小孩，也很容易识别一个物体是否苹果。但是对机器人来说，视觉是否是一个很简单的问题呢？ 如下是一个编程人员与一个机器人的对话。 Programmer: I want you to tell me if you see any apples. Robot: Describe this thing you call "apple." Apples are apple-shaped. Look for something that's apple-shaped. What do you mean by "apple-shaped"? Well, sort-of round with a dimple at the top and a little stem. This "stem" is always at the top of an apple? Not necessarily. The apple might be upside-down or sideways. You must look for stems in all directions. What if the stem is hidden behind the apple so I can't see it? And what is a stem anyway? Good point. Forget about stems. Let's start with color. Most apples are red, but then again some are green. I guess a bad apple might even be brown. Are all things red, green or brown apples? No! Apples are other things too. For example, most apples are a little bit shiny. What do you mean by shiny, and how shiny is "a little bit shiny"? They reflect light well, but not as well as a mirror. Great. What is a mirror? And what do you mean by "reflect"? A mirror is something that shows an image of you and what's behind you when you look into it. I don't know what I look like, or what's behind me - I only have one camera. Ok, forget about the mirror. You asked what "reflecting" means. Reflecting is when light bounces off of something. If there's a bright light on the apple, you will see a small white reflection off of it. Got it. So wherever I can find red, green, or brown, and a white spot inside (if it's bright), there's an apple. How can I tell if the light is bright enough that it would make a reflection? Argh! Never mind! Let's try something else - size. Apples are about 3 inches across. What do you mean by "about"? If it's 4 inches across could it be an apple? How about if it's 27 inches across? Let's say all apples are between 3 and 5 inches across. Is a quarter of an apple still an apple? Yes, but let's pretend it's not. What if in front of me there's a red apple sitting on a red table? How can I tell where the apple ends and the table begins? If I can't do that, I couldn't possibly find an apple. Well, you can see the edge of the apple, right? Maybe. How do I do that? The edge is darker than the rest of the apple if there's a shadow, or lighter if there's no shadow and the table is lighter than the edge of the apple, but the apple isn't necessarily a perfectly solid color - I mean there could be dark spots on the apple that aren't edges, and - oh no - there could be photographs of apples on the table which aren't actually apples, or even a wax model of an apple which looks exactly like an apple but isn't, or the apple could have a bite taken out of it, or be really far away so it looks too tiny or too close so it looks really big, or, or... um... Had enough? Yes, I think I get your point.