Cameras, lenses and mirrors
In the AIRLab you can find different kind of cameras. These are the main groups:
- Analogue cameras. Video output is given as an electrical signal, which needs analogue-to-digital conversion to be used by a computer; this is done by a specific card called frame grabber or video capture card (the latter tend to be the lowest-performance items). Analogue video is outdated for computer vision and robotics applications, due to its cost, low performance and complexity; nowadays digital camera systems (such as all the ones listed below) are always preferred.
- USB cameras. Usually very cheap, they are suitable for low-performance applications (i.e. those where low frame rate is needed and low image quality can be accepted). Their main advantage (along with cost) is the fact that every modern computer has USB ports. The fact that the USB standard includes 5V DC power supply lines helps simplifying camera design and use.
- FireWire cameras. The FireWire (or IEEE1394) bus is generally used for low-end industrial cameras, i.e. devices with technical characteristics much superior to those typical of USB cameras. Industrial cameras usually give to the user a much wider control over the acquisition parameters compared to consumer cameras, and therefore they are usually preferred in robotics; their downside is the higher cost. There are different versions of IEE1394 link (see http://en.wikipedia.org/wiki/Firewire for details), with different bitrates, starting from the 400Mbit/s FireWire 400. Generally they are all considered superior to USB 2.0, even if theoretical bandwidth is lower for FireWire 400. Firewire ports can include power supply lines, but some interfaces (and in particular those on portable computers) omit them. Although the use of FireWire interfaces has expanded in recent years, they are not yet considered a standard feature for motherboards.
- GigE Vision cameras. GigE Vision (or Gigabit Ethernet Vision) is a rather new connection standard for machine vision, based upon the established Ethernet protocol in its Gigabit (i.e. 1000Mbps) version. It is very interesting, as complex multiple-camera systems can be easily built using existing (Gigabit) Ethernet hardware, such as cables and switches. Vision data is acquired simply through a generic Ethernet port, commonly found on motherboards or easily added. However, 100Mbps (or fast Ethernet) ports are not guaranteed to work and can sustain only modest video streams; on the other hand, 1000Mbps ports are now standard on motherboards, so this will not be a problem anymore in a few years. It seems that GigE Vision is becoming the most common interface for low- to medium-performance industrial cameras.
- CameraLink cameras. Cameralink is a high-speed interface expressly developed for high-performance machine vision applications. It is a point-to-point link, i.e. a CameraLink connection is used to connect a single camera to a digital acquisition card (frame grabber). Its diffusion is limited to applications where extreme frame rates and resolutions are needed, because CameraLink gear is very expensive.
The following is a list of the actual cameras available in the AIRLab. For each of them the main specifications are given.
TODO for Giulio: camera list
Industrial cameras usually have interchangeable lenses. This allows for the choice of the lens that is more suitable to the considered application. There are two main standards for industrial camera lenses: C-mount and CS-mount. Both are screw-type mounts. CS-mount is simply a modified C-mount where the distance between the back of the lens and the sensor element (CCD or CMOS) is shorter: therefore a C-mount lens can be mounted on a CS-mount camera if an adapter ring (i.e. a distancing cylinder with suitable threads) is placed between them. It is impossible, though, to use a CS-mount lens on a C-mount camera: if you try you will almost certainly break the sensor, scratch the lens, or both. Just because a lens fits a camera, it doesn't mean it can be actually mounted on it!
Be aware that sensor dimension (i.e. its diagonal, measured in fractions of an inch) is not the same for all cameras. Therefore one of the key specifications for a lens is the maximum sensor dimension supported. If you use a lens with too big a sensor, the edges of the image will be black as they lie outside of circle of the projected image. Also beware of the strange convention used for sensor diagonals, i.e. a fraction in the form A/B" where A and B are integer or non-integer numbers. For instance an 1/2" sensor is smaller than an 1/1.8" one. The variability of sensor dimensions has another side effect: the same lens has a different angle of view if you change the sensor size. Therefore the same lens can behave as a wide-angle with a large sensor and as a telephoto with a small sensor.
An useful guide to lenses (in Italian or English) can be found at http://www.rapitron.it/guidaob.htm.
The following is a list of the actual lenses available in the AIRLab. For each of them the main specifications (and a link to the maker's or vendor's page for full specifications) are given. A '?' means an unknown parameter: if you know its value or experimentally find out it when using the lens (e.g. the maximum sensor size) please update the table before the information is lost again! A 'YES' in the 'Mpixel' column indicates a so-called Megapixel lens, i.e. a high quality, low-distortion lens designed for high-resolution industrial cameras (typically having large sensors); please note that some of these are specifically designed for B/W (i.e. black and white) cameras. The 'how many?' field tells if multiple, identical items are available. All lenses are C-mount.
|focal length||max. aperture||max. sensor size||maker||model||Mpixel||how many?||link to full specifications|
|4.0mm||f2.0||1/2"||Microtron||FV0420||YES (B/W only)||1||http://www.rapitron.it/obmegpxman1.htm|
Much work has been done and is being done at the AIRLab on the topic of omnidirectional (machine) vision (sometimes referred to as omnivision). Omnidirectional vision systems use special hardware to overcome the limitations of conventional vision systems in terms of field of view. The approach to this problem that we generally adopt is the use of conventional cameras in association with convex mirrors, i.e. the capturing of the image reflected by a suitably-shaped mirror with a camera. The possibility of designing mirrors with specific geometric properties gives a very useful means to control the geometric behaviour of the whole camera+mirror system.
TODO for someone who knows better than me (Giulio) ;-) : mirror list