Smart Cameras vs. PC-Based Machine Vision Systems

The question often comes up as to what is the most appropriate approach to take in implementing a machine vision system - using a Smart Camera or using some sort of PC-based approach. There is no question that as the microprocessor, DSPs and FPGAs are getting faster and, therefore, more capable, Smart Cameras are getting smarter. Hence, they are a challenge to more 'traditional' approaches to machine vision. Significantly, however, 'traditional' approaches are also taking advantage of the advances and so, too, are faster and smarter.

'Traditional' approaches more often than not today mean an implementation based on a PC. This could be either using a camera with the capability to interface directly to the PC (IEEE 1394/Firewire, CameraLink, LVDS, USB, etc.), or a system designed based on a frame grabber or other intelligent image processing board or vision engine that plugs into the PC. In this latter case, more conventional analog cameras are used as the input device.

A Smart Camera, on the other hand, is a self-contained unit. It includes the imager as well as the 'intelligence' and related I/O capabilities. Because this format resembles the format of many intelligent sensors, these products are often referred to as 'vision sensors.' More often than not, however, a vision sensor has a limited and fixed performance envelope, while a Smart Camera has more flexibility or tools, inherently capable of being programmed to handle many imaging algorithms and application functions. A PC-based vision system is generally recognized as having the greatest flexibility and, therefore, capable of handling a wider range of applications. One significant difference is that vision sensors/Smart Cameras are essentially single socket units, while PC-based vision systems can generally handle multiple camera inputs.

Another style machine vision system that falls somewhere between the PC-based vision system and a Smart Camera/vision sensor is what some call an 'embedded vision computer.' This type system is essentially a stand-alone box with frame storage and intelligence. It generally has limited flexibility and comes with a number of fixed application-specific routines. These are distinct from Smart Cameras in that the camera is tethered to the unit rather than self-contained. They often have the ability to handle multiple camera arrangements, which can be useful for many applications.

All these systems can be found with high-resolution imagers (nominally 1000 X 1000) and/or color imagers. Interestingly, versions are often competitively priced. Some smart cameras and virtually all PC-based imaging capabilities can handle applications that require line scan cameras as well.

To gain insights into how the suppliers perceive the differences between these products and their applications, input was canvassed from all the suppliers of Smart Cameras/vision sensors, frame grabbers and vision engines. While these lists include over 100 companies, responses were only obtained from 11. These 11, however, are companies that offer representative products from each of the abovementioned classes. It should be understood that there could be some bias based on the specific product class being offered, since not all companies that responded offer products in each of the classes.

The following were kind enough to respond to our questions:

Manish Shelat, Adept Technology, suppliers of PC-based vision systems
Bud Patel, Applied Vision, suppliers of PC-based vision systems
Gus Vargas, Aromat, suppliers of embedded vision computers
Phil Colet, Coreco Imaging, suppliers of frame grabbers
Sal D'Agostino, Computer Recognition Systems, suppliers of vision engines
Phil Heil, DVT, suppliers of Smart Cameras/vision sensors
Yves Joskin, Euresys, suppliers of frame grabbers
Stephane Francois, Leutrek, suppliers of frame grabbers and Smart Cameras
Jason Mulliner, National Instruments, suppliers of frame grabbers and vision engines
Endre Toth, Vision Components, suppliers of Smart Cameras/vision sensors
Vic Wintriss, Wintriss Engineering, suppliers of Smart Cameras

1. What are the advantages/disadvantages of PC-based machine vision versus Smart Camera-based machine vision?

PHIL COLET: 'These are really two different technologies targeted predominately at two different customer segments.  The PC based technology is largely focused on what one would call an OEM customer.  The Smart Camera technology is largely focused on the end user.  Although this is not a hard and fast rule, it does apply to perhaps 90% of the applications out there. For this discussion I will assume that a customer receives two boxes from two vendors.  One is a PC with frame grabber, camera, lighting; the other is a Smart Camera approach.

PC Based Machine vision advantages:

• Flexibility - The PC offers greater flexibility in the number of options that can be selected.  For example one can use a line scan versus an area scan camera with the PC.  One can use third party software packages with the PC approach (Smart Cameras tend to be single source software).
• Power - PC's tend to offer greater power and speed due in large part to the speed of the Intel processors used internally.  This power in turn means that PC's are used to handle the 'tougher' applications in machine vision.

Smart Camera Advantage:

• Cost - Smart Cameras are generally less expensive to purchase and set up than the PC solution since they include the camera, lenses, lighting (sometimes), cabling and processing.
• Simplicity - Software tools available with Smart Cameras are of the point-and-click variety and are easier to use than those available on PC's. Algorithms come pre-packaged and do not need to be developed, thus making the Smart Camera quicker to setup and use.
• Integration - Given their unified packaging, Smart Cameras are easier to integrate into the manufacturing environment.
• Reliability - With fewer moving components (fans, hard drives) and lower temperatures, Smart Cameras are more reliable than PC's.'

YVES JOSKIN: 'Argument #1 - When developing a machine vision solution to an industrial problem, the system integrator usually does not exactly know where the problem analysis will actually take him. It is not unusual that the requirements for processing power or functional capabilities turn out to be poorly foreseen, and the final solution is never simpler than the initial idea.

The PC-based solution offers an immense set of potential resources, in terms of computational or interfacing performance. The PC platform is essentially open, and it became so popular that its cost to performance ratio is unbeatable.

The low-cost argument is particularly true for the desktop PC, but it is sometimes claimed that the mechanical weakness of a mainstream desktop PC is not compatible with the industrial requirements of a serious machine vision application. However, compared to the more expensive industrial PCs, the low-cost desktop PC offers the latest CPUs and associated components, offering the highest performance at the lowest cost.

All in all, when considering all trade-offs to be made in its design, the machine vision developer reaches the conclusion that the PC-based system is the most cost-effective solution. It best suits his need for functional evolution during the design stage, and even after when upgrading the system becomes an issue. Upgrading software is an easy way to improve functionality, and upgrading the PC hardware is an easy to improve the performance.

Argument #2 - Some vision problems belong to a well-defined, special purpose class, identically found at numerous user sites. A good example of such an application consists in decoding Datamatrix codes.

In those cases, the computational and interfacing resources to solve the problem are predictable. It makes a lot of sense to pack the required resources into a single unit, and this is a definition of a Smart Camera. If (and only if) the quantity involved in this special application is significant, the cost can be reduced to a level that makes the product competitive with a PC-based solution. The key is that the scope of the special purpose application is fixed in advance. There are good reasons to believe that there will be no demand for faster performance or added functionality.

In my opinion, Smart Cameras are applicable only for that kind of end-user oriented market. It is true that that Smart Cameras are offered for general-purpose applications, and featuring a set of user-programmable functions. Integrating these programmable Smart Cameras into a system may be a relatively or even very easy task. However, in many cases, the user will sooner or later fall into the scope of Argument #1. He will find a limitation in the Smart Camera, such as lacking speed or unsupported feature. When the problem is simple enough to be solved, it actually belongs to the special-purpose class of application.'

PHIL HEIL: 'The Smart Camera has a processor at each inspection point. This gives a network of Smart Cameras a distinct speed advantage over a single processor system.  Ethernet allows the cameras to be easily managed from a single PC on the factory network and eliminates problems with PC hardware compatibility and operating system bugs.'

ENDRE TOTH: 'The direct connection of the CCD and the processor brings several advantages in accuracy, pixel-identical capture (low or not pixel jitter). You have control over the camera, providing flexibility for the design engineer! In the case of a PC system the processor is removed from the camera. The camera, the frame grabber and the PC come from three different manufacturers, bringing compliance, driver and partially implemented feature set problems. Have you tried to implement the following very simple and obvious machine vision function with a PC system: changing or adjusting the electronic shutter after each frame? There is no difficulty to implement this function in a Smart Camera. You could list many different functions like that, which practically originate from the different architecture topology.

Smart Cameras are compact units, whereas a PC system comes with a lot of 'baggage'. Even the operating system (frequently Windows) has a lot of baggage built into it. The baggage is a disadvantage in control, automation and mission critical industrial applications. You just need to talk to control engineers to find  out. In applications like on an airplane, underwater, mine, hazardous environment and machines like a printing press, Smart Cameras have a definite advantage, and they should be considered when selecting the machine vision system.

In a PC solution you put together your system of components, frame grabber, camera, PC etc. These system components work together through standard interfaces. These interfaces define what and how you can do!

Smart Cameras are 'open embedded' systems. The designer and control engineer are in full control, you can count on and you can be sure what will happen next. The software engineer has to deal with a lot of baggage, when writing the machine vision code for a PC system.  Also, Smart Cameras have fewer parts, less components; they are significantly (many times) more compact units physically than a PC system. Simple maintenance, replace and repair of one part only.

Smart Camera installations include substantially fewer cables. Smart Cameras provide additional flexibility for a system designer. Today a system designer can select a wide range of topology for their design, by selecting from frame grabbers, embedded machine vision systems, Smart Cameras etc. and freely combine them to create the best fit and most cost effective systems. 

Many Smart Camera applications today are networked. In several cases Smart Cameras are connected to PCs, PLCs etc. Smart Cameras perform the repetitive image processing functions providing only the results to the PC. The PC does the record keeping reporting administrative tasks it was originally designed for. Smart Cameras fit well into today's fashionable concepts of distributed computing, distributed control etc.'

VIC WINTRISS: 'Advantages (of Smart Camera): Since computation intensive vision processing is done in the camera, multiple camera systems do not bog down a central CPU.   Even with a single camera, only the results need be passed back to a PC for further processing. usually bytes of data instead of Mbytes of picture.  Only AOI (area of interest) is needed by a PC for analysis.  A dumb camera has to send back an entire picture, and most of it garbage.  Bandwidth channel requirements are much smaller for a Smart Camera system.  PC computation requirements are much smaller for a Smart Camera system.  Also, casting discussion as Smart Camera vs. PC-based approach is not appropriate, because  those are two extremes of the continuum of system configuration possibilities.  There are hybrid systems that employ both Smart Cameras and a PC  It's really a systems design problem to allocate the necessary functionality to the appropriate system element to optimize performance and suitability. Sometimes that will result in a Smart Camera system, sometimes a PC-based system and at other times, a hybrid will be the best choice.'

SALVATORE D'AGOSTINO: 'Expandability, easier to configure other types of I/O, communications, upgrades' are the advantages of PC-based vision systems.'

BUD PATEL: 'PC-based machine vision systems are generally more capable than Smart Camera based systems.  They have more computational horsepower to be able to handle much more sophisticated software algorithms.  The Smart Cameras are great for simple tasks using general edge detection or binary tools; however, they do not have the computational power or memory to handle more sophisticated application specific algorithms.  They will be limited to how fast and how complex the inspection performed will be.  Also in certain applications, Smart Cameras will not be able to handle the throughputs for 100% inspection.  One example of an application where Smart Cameras would not be suitable is inspecting two-piece beverage cans.  It is common for these production lines to operate at speeds greater than 2000 PPM.  This application requires application-specific lighting to properly illuminate this complex part and requires the detection of very small functional defects throughout the can.  Smart Cameras are very good at basic absence/presence type applications.  Typically these low-end systems are used to read barcodes and other simple tasks.   Most Smart Cameras do not include a monitor or user interface.  The burden of the user interface is put on the end user.'

JASON MULLINER: 'By leveraging off-the-shelf commercial technology, PC-based machine vision systems can leverage the high performance of current processor and bus technologies.  This allows for more flexibility because of the open nature of the PC.  Customers can choose their imaging interface whether it be analog, parallel digital, Camera Link, or IEEE 1394.  They can also choose the method of programming.  A customer may want the ease of use of an interactive configurable environment or the flexibility and power of a complete application development environment.'

STEPHANE FRANCOIS summarizes the advantages of a PC-based vision system as 'Flexibility, versatility, scalability.'

MANISH SHELAT (coming from the perspective of robot applications): 'Ease and accuracy of calibration: a Smart Camera based robot system is rudimentary when it comes to calibration.  The robot or mechanism and Smart Camera are calibrated separately.  The Smart Camera calculates part location offsets from a known location and instructs the robot arm to pick up the part at that offset from the initial programmed pick up location.  In contrast, the robot controller will calibrate the vision system and robot in a single coordinate system.  Part locations are then defined in the same six-degree of freedom coordinate space that the robot is programmed in.  This is the proper method of robot and vision calibration.

• Cost of a multiple camera system: Typical PC based vision systems can handle up to four cameras per frame grabber.  In applications that require multiple cameras, the cost of a PC based system should be compared with the cost of multiple Smart Cameras.
• Communications overhead: Smart Cameras communicate with robot controllers via serial interface (RS-232) or Ethernet.  Communication overhead adds a delay in robot to camera communication.  In contrast, in the case of an integrated motion and vision system the communication overhead is miniscule since all hardware and software is on the same platform.
• Seamless integration of motion and vision:  In order to use Smart Cameras the customer has to select and link separate off-the-shelf products.  With off-the-shelf controller-based vision system the customer is investing in a pre-engineered and pre-configured system.
• Size and weight:  Typical analog cameras are approximately 44 x 29 x 71 mm in size and weigh 140 grams. Smart Cameras require additional electronics, which increases their size and weight.   This may be an important selection criterion for applications that have limited camera-mounting space and for arm-mounted cameras where the momentum of the arm or mechanism is critical.
• Power Consumption and Heat Generation:  Smart Camera consumes more power for the additional electronics and this results in higher heat generation.  Heat can deteriorate camera performance over time.' 

GUS VARGOS (commenting from the perspective of a supplier of embedded vision systems) observes:

a) A laptop or PC is not required to program or configure the unit (standard laptop PC cost is $2,000). Keep in mind to setup a vision system with a PC in a manufacturing environment or production site, is quite cumbersome to begin with, imagine dragging around a desktop.

b) Serial Communications (PLC communications). Typically production machines that require a visual inspection system are controlled by small I/O (compact) PLC (Programmable Logic Controllers). Such devices usually provide serial communications ports (RS232) to other smart devices such as Barcode Reader and Machine Vision. Stand-alone vision systems provide direct serial communication to interface directly to the controller (PLC). This allows an effective and fast data transfer of the vision system inspection results, without having to connect through standard I/O.  Which is limited to binary data (Pass/Fail).

c) PC-based Vision Systems require Ethernet Communications, because this type of communication must be used for image transfer from the camera to the PC. The problem with this approach is that initial inspection, setup, and monitoring become slower due to a standard PC update time (refresh time). Not to mention the normal complexity of networking setup and troubleshooting, which requires extensive PC hardware knowledge, and windows hardware compatibility.

On the other hand, stand alone systems provide real-time video feedback directly to the user/operator via built-in video outputs. Which allows easier troubleshooting and setup.'

YVES JOSKIN summarized his answer with the following table:

2.  Does one approach have limitations that the other one does not have?

D'AGOSTINO:  'The limitations [of smart cameras] fall on the fixed configuration of a Smart Camera, the benefit should be cost or space savings.'

JOSKIN:  'The Smart Camera approach has a limited computational power that cannot be exceeded. It also usually has a defined set of functionality.'

MULLINER: 'Smart Cameras currently do not provide scalability.  If a customer is unable to solve their application with a Smart Camera they cannot migrate to a more powerful PC-based system and preserve their software investment.'

COLET:  'Absolutely, but while one approach has a strength (simplicity for example), the other approach has a different opposite strength.  So while PC's are not as simple as Smart