Hola, en el libro Pic Robotics John Iovine hay un proyecto con una camara CMU. espero q sepas ingles!
The CMU camera was developed at Carnegie Mellon University
(CMU). The CMU camera uses an SX28 microcontroller interfaced to an
Omnivision OV6620 CMOS camera chip. The SX28 microcontroller does much
of the image processing for us. We communicate with the camera via a stan-
dard RS-232 or TTL serial port.
A few of the CMU camera features are as follows:
Color Robotic Vision System
The robot we will build in this chapter will be capable of looking at and seeing
an object (or target) and following that object. If the object gets too close to the
robot, the robot will back away from it. Choose a bright target that will have
good contrast with the background environment. Colored objects are fine; a
bright red or yellow ball works fine. For my tests I used a 2.5-in square of
orange construction paper taped to a stiff wire.
One of the most difficult areas in robotics today is the creation of an artifi-
cial vision system for a robot to see. Teaching a robot to see is not simply a case
of connecting a video camera to a computer. The electronic representation of an
image created by a video camera must be presented in a way a computer can
“look at” and interpret (“see”) the image.
To gain an understanding of the processes involved, let’s examine how a
computer might look at a simple black-and-white image. We must first define
the resolution of the picture image in pixels. For our discussion let’s assume a
low-resolution picture of 80 143 pixels. At that resolution our computer must
look at 11,440 pixels (80 143 11,440). Each pixel of the picture can be any
tonality of gray between pure black and pure white. We now have to determine
how many different shades our computer can differentiate between pure black
and pure white. If we wanted each pixel to be represented by a single byte (8-
bit number), there would be 256 shades of gray, including pure white (0) and
black (255).
The computer would look at each pixel and assign a number between 0 and
255 depending upon its tonality (grayness). After the computer assigned num-
bers to each 11,440 pixels, it transformed the basic image into a numbered
representation it needs to “look” at the image.
The software that looks at an image and interprets visual features is appro-
priately called image processing. Robotists over the years have gleaned some
techniques for helping computers to see. One technique is called edge detec-
tion. Here the computer looks through the image. You program the computer
