Adding vision to your projects needs not be a difficult task. Whether its machine vision for robot control or the sampling and storage of images for security, CMOS images sensors can offer many advantages over traditional CCD sensors. Just some of the technical advantages of CMOS sensors are,

No Blooming Low power consumption. Ideal for battery operated devices Direct digital Output (Incorporates ADC and associated circuitry) Small size and little support circuitry. Often just a crystal and some decoupling is all that is needed. Simple to design with.

There are many manufacturers making CMOS Image Sensors. Just some of the more notable ones are Photobit, OmniVision, Agilent (formally known as HP), ST who acquired VLSI Vision and Mitsubishi.

There are two different categories of CMOS Sensors based on their output. One type will have a analog signal out encoded in a video format such as PAL, NTSC, S-Video etc which are designed for camera on a chip applications. With these devices you simply supply power and feed the output straight into you AV Equipment. Others will have a digital out, typically a 4/8 or 16 bit data bus. These 'digital' sensors simplify designs, where once a traditional 'analog' camera was feed into a video capture card for conversion to digital. Today, digital data can be pulled straight from the sensor.

The main components to a Digital Video Camera design are

CMOS Image Sensor. The heart of the camera. It produces a digital/analog output representing each pixel. It's support circuitry will normally include a Crystal Oscillator and power supply decoupling. Some sensors may need a resistive bias network of some type. All of these components are normally surface mounted on the back of the PCB and occupies very little real estate. The lens Holder. This will be either a plastic or metal mount which attaches to your PCB and allows a standard size lens to be screwed in. The screw thread facilitates focusing for fixed lens systems. The base of the lens mount may also have a IR (Infra Red) filter. The Lens. This will determine your Field of view among other things. Lenses range from fish-eye to telescopic and need to be purchased to fit the parameters of your sensor and lens holder.

Once you have completed the above, you have yourself a imaging system which constantly spits out a pixel data stream synchronised to a pixel, frame and/or line clocks. Connecting this directly to a microcontroller/processor system will cause headaches. Trying to clock this raw data in will use up great amounts of CPU time, if your uC could do it in the first place. If you drop a pixel because an ISR is doing some thing more privileged, then you have no ability to sample that location again, and thus no method of error correction.

While the frame rate on many devices can be slowed down by using internal divisors, it still doesn't reach an acceptable speed nor allow random access to pixels. Reducing the master clock rate of the device will effect exposure times and other time dependent settings, thus is not an option. Clearly some additional circuitry will need to be designed.

By using a CPLD/FPGA and RAM, you can program the CPLD to dump the data straight into RAM. Your micro could then read this RAM through the PLD which could be memory mapped. If you really want performance (And budget is not a problem), you could use Dual Port RAM. If however you only want to capture one frame, then the PLD could copy one frame into memory and ignore subsequent pixel data until an event such as when your device has read all the data out of RAM. Other options are to use a LVDS (Low Voltage Differential Signalling) serial bus, to relay your data over a few metres or more. At a high enough clock rate, you won't wait all day for a frame.

The other thing you must not forget is how to control the sensor. Most of it's internal parameters are controlled by a serial bus, typically I2C for the majority of sensors. This can either be controlled through a memory mapped Register programmed into your PLD or via an I2C port straight from your uC. All up this makes quite a cheap way to capture video. Ideal for your Embedded Linux Systems.

Photobit have many CMOS Image Sensors. CMOS APS (CMOS active pixel sensor) was first created by a team of JPL engineers lead by Dr Eric Fossum. Dr Fossum is now the Chief Scientist and Chairman of Photobit. Two of Photobit’s more common sensors are the PB-100 and PB-300. Other sizes can be sought from Photobit's Product Matrix

PB-100

• CIF Resolution - 360 x 304 Colour or Monochrome
• 1/5 Inch Optical Format
• 0-39 Frames a Second, 6 Megabyte Data Rate
• Output 8 Bit Serial Digital Video, I2C Control
• 3.3V Supply, 30mA (100mW)
• 28-pin CLCC

PB-300

• VGA Resolution - 640 x 487 Colour or Monochrome
• 1/3 Inch Optical Format
• 0-39 Frames a Second, 24 Megabyte Data Rate
• Output 8 Bit Digital Video Data Bus, I2C Control
• 5V Supply, 60mA (<300mW)
• 44-pin CLCC