Robotic Nation Evidence

The Robotic Nation article points out that the main problem with robotic vision right now is the fact that it requires vast amounts of computing power -- power that is not going to be inexpensively available for perhaps 10 to 20 years, but will then be available in abundance. This article is technical, but offers a nice example of why you need so much computing power:

Sorting data in two clock cycles

From the article:

We want to focus on the first step because it sometimes consumes a large part of the overall runtime. In this step, color-depth is being reduced. For traffic signs, only some color information (in other words, red, blue, yellow) or so called noncolor information (black or white) are relevant. Each pixel comes from the image sensor with a color depth of 24 bits (RGB coded). After reduction the color depth is just three bits. This speeds up the application because we have to handle less data. Afterwards, it's helpful to enhance image quality using filter operators.

These operators are partially used on every pixel and their nearest neighbors. The median operator, for example, improves the quality of the image by reducing noise. Using it on the whole image makes it necessary to sort nine data values for each pixel, as shown in Figure 3. That means that nine data values have to be sorted perhaps a million times or more.