Home Machine vision knowledge center

How to select a machine vision camera interface (USB3 / GigE / 5GigE / 10GigE Vision)?

Machine vision camera Interface whitepaper

This article will provide a deeper inside in what the benefits and downsides are of each type of machine vision camera interface, so you can select the right interface for your machine vision application. It has to be noted that GeT cameras only supplies USB 2.0, USB 3.0, GigE, 5GigE and 10GigE machine vision cameras and the article will mainly be focused on these five industrial machine vision interfaces. However in the table below we have incorporated also other interfaces to make the overview of industrial machine vision camera interfaces complete.

	USB2	USB3	FireWire	GigE	CameraLink	Coaxpress	5GigE	10GigE
Bandwidth (Megabyte/s)	40	400	80	100	Base = 250 Medium = 500 Full = 750	1 lane = 750 2 lanes = 1500 3 lanes = 2250 4 lanes = 3000	500	1000
5MP camera Max framerate	8fps	80fps	16fps	25fps	150fps	600fps	100fps	200fps
Cable length (maximum)	5m	4.6m	10m	100m	7m	100m	100m	100m
Power + data on one cable	Yes	Yes	Yes	Only if PoE	PoCL	Yes	Only if PoE	Only if PoE
Frame-grabber mandatory	No	No	Yes	No	Yes	Yes	No	Yes
Cable costs	Low	Low	Medium	Low	High	Low	Low	Low
Camera costs	Very low	Low	Low	Low	High	High	Medium	High
CPU usage	Medium	Low	Low	Medium	Low	Low	Medium	Medium
Customer acceptance	Declining	Good	End of Life	Good	Declining	Growing	Growing	Growing
Multiple cameras	Fair	Excellent	Excellent	Good	Fair	Excellent	Excellent	Excellent

What is the bandwidth of a machine vision camera interface?

The first topic we will discuss is bandwidth of a machine vision camera interface. Simply said is bandwidth a way to measure the amount of data that can be sent between computers during a specific period. In terms of machine vision, this concerns the amount of image data that can be sent from the machine vision camera to a host controller. Bandwidth is measured in megabytes per second, which means that an USB2.0 machine vision camera can send image data equivalent to 40 megabyte each second.
This also means that an USB 3.0 machine vision camera interface(if used at full capacity) can send image data equivalent to 400 megabyte per second, which is around 10 times more than USB 2.0 machine vision camera interface. When sending large amount of data (large bandwidth usage), it also costs the computer more time to process this data. As a result it takes longer for the computer to reconstruct an image. However, a large bandwidth makes the data transfer faster, allowing the host controller to start calculating and reconstructing images earlier and faster. This is often required in machine vision applications.

Performance factor for machine vision camera interface bandwidth

The maximum bandwidth of your machine vision camera interface is only one of the few factors that affect the final data throughput, resulting in less images per second. Latency, jitter and packet loss can degrade your data throughput, but also the quality of your cable can make less bandwidth available. We advise to read our guides on USB 3.0 or GigE machine vision cameras, where we tackle these issues.
It is very dependent on the machine vision application and the limitations of your system which bandwidth is needed. In general you can say that if you have a 2Megapixel camera, each images is 2 Megabyte. If your requirement specifies 10 images per second, the required bandwidth is 2x10=20megabyte per second. Every machine vision camera interface can cover this bandwidth. However if the requirement is 100 images per second, you need 2x100=200megabyte per second bandwidth and only cameralink, coaxpress, USB 3.0, 5GigE and 10GigE machine vision camera interface remains.

What is the maximum cable length for a machine vision camera interface?

The second point of discussion is the maximum cable length of the machine vision camera interface. These maximum cable lengths indicates which cable length lets you still work with the maximum amount of bandwidth. There are USB 3.0 cables for example which go up to 25 meters, however this number vastly exceeds 4.6 meters machine vision USB 3.0 cables, which results into a big chunk of bandwidth being lost. Cabling varies from the degree of flexibility you desire and also influences how much bandwidth you like to transfer. Typical pricing for cabling are:
-   USB2 cable for machine vision camera cost between 4 and 6 euros
-   USB 3.0 cable for machine vision camera cost between 16 and 20 euros
-   GigE, 5GigE, 10GigE cable for machine vision camera cost between 8 and 10 euros
When the machine vision camera is mounted on a moving platform a highflex cable is required. This highflex cable is capable of bending multiple times per second, every day for many years. GigE Highflex cables are the cheapest machine vision highflex cables. They start at 50euros.

Is a frame grabber required for USB 2.0, USB 3.0 and GigE machine vision cameras?

For the interfaces USB 2.0, USB 3.0, GigE and 5GigE, the cameras do not need a Frame grabber. Only for the machine vision interfaces like 10GigE, camera Link, Firewire B and Coaxpress a framegrabber is mandatory. It has to be noted however, that ‘frame grabber’ has a broad definition and is subjected to interpretation. If a computer has not enough USB 3.0 or GigE ports, we have an interface card to extend the amount of USB 3.0 or GigE interface ports of the computer. This interface card is also sometimes called a framegrabber. All interface cards / frame grabbers we offer are tested with our machine vision cameras to be able to run at full bandwidth.

What is the price for a machine vision camera?

Regarding the five types of machine vision camera interfaces we offer is 10GigE the most expensive and USB2, the cheapest machine vision camera interface. All five are still cheaper than camera Link and Coaxpress cameras. With the competitive pricing model the cheapest models are all available under €100. To put this in perspective, a machine vision frame grabber for coaxpress is around €500 and a simple coaxpress camera is around €1000, added up this is €1500. Making it 10x more expensive than USB 3.0 machine vision camera interface. Therefore we believe that with the USB2, USB 3.0, GigE and 5GigE interface we can cover most mainstream machine vision applications.

How much is the CPU load of the machine vision camera?

Each machine vision camera interface uses CPU power to transfer the images from the interface to the memory of the processor. The amount of CPU power that is required is defined as CPU load. USB 3.0 has a ‘low’ CPU load on the host controller, where USB2 and GigE have ‘medium’ CPU load. USB 3.0 has a low CPU load because it make use of a total different protocol then USB2 and GigE and this protocol needs less resources.
USB 3.0, GigE, 5GigE and 10GigE machine vision camera interface use a Unicast Dual-Simplex data interface, this interface allows the transfer of data in both directions. It works where a host directed procedure sends route packets and notifications explicitly in an allochronic manner. This allows the machine vision camera to send a message to the host controller when it is ready for data transmission. This newer mechanism reduces system and CPU load compared to the polling mechanism in USB2.0. To clarify, the USB2.0 machine vision camera interface makes use of one way data transfer, which results into stacking more data (image packets) on the interface card and putting more work on the CPU.
A 5GigE machine vision camera also uses extra CPU load. The 5GigE camera send images with lossless compression over the 1000mbit ethernet interface to the computer. The computer has to decompress the images, resulting in extra CPU load.
To summarize, USB 3.0 has by far the lowest CPU usage, GigE and 10GigE has a little bit more CPU usage then USB 3.0. 5GigE and USB2 have the highest CPU usage.

Consumer acceptance and Product Life Cycle of machine vision interfaces

To provide you with an ever better overview of the product life cycle then given in above table, we have also drawn the graph below. It shows the stage of the product life cycle where every machine vision interface is in.

Product life cycle machine vision camera interface

Introduction

The first two machine vision interfaces, that are still in the stage of introduction (and development) are the 10 GigE and N-BASE-T (5GigE) products. Both of the machine vision cameras are suited for high speed imaging using the commercial Cat6e ethernet cables that are used with normal GigE cameras. 10 GigE Vision has around 10x the bandwidth of a standard GigE Vision cameras but requires a 10GigE networkcard.

Growth

One of the interfaces that is listed in the growth phase is Coaxpress. This interface is similar to 10 GigE and N-BASE-T as it is suited for high speed imaging. The downside of this interface is that a framegrabber is mandatory and that both the framegrabbers and the coaxpress camera are relative expensive.

Maturity

In the maturity phase you see two machine vision interfaces that GeT cameras sells which are GigE and USB 3.0. The USB 3.0 Vision standard was founded in 2011 and the GigE Vision standard in 2006, both founded by the Automated Imaging Association. We believe both these machine vision interfaces have reach the peak of their maturity and the demand is at its highest momentarily. We expect that GigE will stay in the maturity phase for still at least 5 years and USB 3.0 even longer.

Decline

The last two machine vision interfaces, USB2 and camera Link are in the decline phase. Due to better alternative (USB 3.0 instead of USB2 and Coaxpress instead of cameralink) the demand is and will keep declining over the next years. USB2 has very limited bandwidth and the protocol is not so robust making it suited for only a small range of applications. However, when the machine vision application has very minimal requirements USB2 is the best choice in terms of costs. camera Link has a good bandwidth, but it is an expensive option compared to Coaxpress, USB 3.0 and in the future 10 GigE and N-BASE-T in terms of bandwidth. Camera link machine vision cameras are also expensive and have pricey cables.

Multiple machine vision cameras

When installing multiple machine vision cameras on one host controller it is important to monitor your bandwidth usage. Using this sort of setup usually requires an interface card with multiple inputs. We will only cover multiple machine vision camera setups for USB2, USB 3.0 and GigE.

Multiple USB 3.0 machine vision cameras

Starting with USB 3.0, this machine vision interface has excellent compatibility for a multiple machine vision camera system. You can either use a star network or point-to-point network for USB 3.0. When using a star network, multiple machine vision cameras are connect to a single USB 3.0 hub. The USB 3.0 bandwidth will be divided over the USB 3.0 machine vision cameras connected. This is because all individual USB 3.0 machine vision cameras consume from the single bandwidth of the host connected to the USB 3.0 hub.
The second option, a USB 3.0 point-to-point network, you will connect each USB 3.0 machine vision camera directly to the USB 3.0 host controller. The USB 3.0 interface card can have up to four USB 3.0 host controller inputs, enabling to connect four USB 3.0 machine vision cameras to one interface card. Each USB 3.0 machine vision camera can now use the full bandwidth of the USB 3.0 protocol.

Multiple GigE machine vision cameras

GigE, 5GigE and 10GigE machine vision cameras are also a good option for a multiple machine vision camera system. With high flexibility and inexpensive cabling it is ideal for monitoring conveyor belts and sports & motion analysis with multiple machine vision cameras. The principle is the same as with USB 3.0 cameras. You can have both a star-network using an ethernet switch or a point-to-point network using an interface card with multiple GigE inputs.

Multiple USB2.0 machine vision cameras

The last interface we offer, USB2.0, are the least suited for multiple machine vision camera setups. This does not mean that you cannot connect and use multiple USB2.0 machine vision cameras, but means that it will perform the worst out of the three. Again both network options (star and point-to-point) are possible, just as USB 3.0.