|
Logic Beyond its Limits . . .
|
CMOS Digital Image Sensors
|
Logic Beyond its Limits . . .
|
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,
|  |
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
|
|
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
OmniVision not only develops CMOS Image Sensors, but also support circuitry such as the OV-511 Advanced Camera to USB Bridge. OmniVision is one of the more popular manufacturers with devices such as the OV7910 NTSC/PAL Camera on a Chip being used in many small analog camera modules around the world. This would be the recommended starting point if you are starting out designing with CMOS Image Sensors.
|
OmniVision and some third part vendors have evaluation modules for the OmniVision sensors. This allows you to get up to speed with the sensor, incorporating a PCB with de-coupling, a Lens and Lens Holder. The majority of the sensor's signals are broken out to a header which you can use to interface to your own designs. The evaluation modules in small quantities are normally much easier to obtain than the sensors themselves, and are typically cheaper as a result. |
 A picture of the M3188 Evaluation Module with the lens holder removed. The signals can be obtained from the 32 pin header on the top of the module |
DIY Electronics (http://www.kitsrus.com) are just one outlet which sells the third party evaluation boards.
Mitsubishi have broken the pack, to produce smaller resolution sensors. These sensors can typically be used for a range of applications such as finger print sensing, motor detection, gaming, tracing of moving parts etc. Just one application is the new optical mice flooding the market place. They use a low resolution Image Sensor to track movement on a wide variety of surfaces.
Also unique to these sensors is in-built image processing. Both sensors can output edge enhanced or extracted data, making them ideal for tracking on small robots, industrial control etc. The sensors can also process 2D images into 1D. The output of each pixel is by the means of a analog potential, thus this must be fed into an ADC to return digital image data.
M64282FP Artificial Retina LSI
Spectronix have used the ST Sensors in their RoboCam Series. ST also offer a couple of CoProcessors, a STV0657 Digital CoProcessor, a STV0672 USB CoProcessor and a STV0680 DSC (Digital Still Camera) CoProcessor. The DSC CoProcessor offers an RS-232 / USB Interface and on board SDRAM Storage.
|
Choosing a CMOS Image Sensor is the easy bit. Once you have chosen your sensor a suitable lens and lens assembly needs to be found. A PCB mount lens holder will screw to the PCB, covering up your sensor from light. The top of the holder will have a standard thread which will allow you to screw in a lens with the same thread type. The lens holder may also incorporate a Infra-Red Cut Off filter and is typically made of Black ABS Plastic or Black Anodised Diecast Aluminium. The typical thread sizes are detailed below, |
 encased in aluminium |
| Mount Type | Thread | Distance from Back Flange to Image Sensor |
| C | 1 – 32 (1 Inch/32 Threads Per Inch) | 17.526 mm |
| CS | 1 – 32 (1 Inch/32 Threads Per Inch) | 12.5 mm |
| S | M12x0.5 | Not Specified |
| X | M10x0.5 | Not Specified |
C & CS Mount Lens are typically used by the CCTV Market. Being an inch in diameter they take up quite a bit of real estate but give a better result. S and X mount lens are more typically used on the eyeball PC Cameras being only 12mm in Diameter. S Mount (M12x0.5) Lens are the more dominant standard in eyeball / small cameras sacrificing image clarity for size.
The Lens will screw into your PCB mount thus ensure it has the same thread. You will be able to purchase lenses of different focal lengths, made of either glass or plastic. Plastic are cheaper and typically of lower quality than multi element glass lenses. Lens can also come housed in plastic or aluminium. Beware of inferior quality.
There are a number of parameters associated with the lens. The Focal Length expressed in millimetres determines the field of view. A typical human perception is about 40 degrees, thus this is targeted as the normal view. They can range from telescopic at approximately 20 degrees field of view which has a high image magnification to fish-eye at approximately 110 degrees or more. The field of view can be calculated from the focal length and diameter of the sensor, but most vendors will normally specify both the field of view and focus length.
|
The aperture number, f Number or f / # specifies the amount of light which passes through the lens. The lower the number, the more light that will pass through the lens, thus the better performance in low light conditions. Every time the f number doubles, the light is reduced by a factor of 4. A lower f-number requires more precise adjustment of focus, where as a lens with a high f-number will be easier to focus. Eyeball cameras for PC imaging have a typical f-number of 1.8 to 2.0 where 2.0 is standard for most S-Mount Lenses. Pinhole lens, while having the ability to be concealed behind a hole have a high f-number thus are not as effective in low light environments. |
 The inside view of a lens holder showing the IR Filter. B/W sensors do not require the filter. |
 Another example of a Lens Holder with IR Filter. |
The pixel element of a CMOS sensor is normally susceptible to a greater span of the electromagnetic spectrum than the human eye. The sensors ability will range from the deep blues to almost infra-red. To get a more accurate measure of what the human eye is seeing, an IR Cut Off filter is included with most lens mounts or lenses. Colour aliasing and blurring can be a consequence of omitting the IR Cut Filter on a colour imaging system. As the IR cut off filter can either be part of the lens or lens assembly this will need to check on purchasing. The filter present on a lens mount looks like a bit of thick glass mounted to the bottom of the mount. This is normally made up of several layers of optical crystal with a IR Cut Crystal layer. Beware of some lens having a film on the back of the lens as these are easy to scratch. |
Typical lens and lens mount vendors are
Sources of additional information
For the faint hearted or strapped for time (but not for cash) you can get pre-build modules from Spectronix which terminates to a Ribbon Cable and can provide a RAM based memory mapped interface. Spectronix’s RoboCam I (164x124) and RoboCam II (356x292) come in both B&W and Colour Versions based on ST/Visions VV5404/VV6404/VV5300/VV6300 CMOS Images Sensors.
|
 |