Smart cameras embed processor power

By joining CCD or CMOS imagers to on-board processors, smart cameras mimic a complete machine-vision system.

By Andrew Wilson, Editor

Incorporating charge-coupled-device (CCD) and CMOS imagers coupled to processors such as RISC, DSP, and CPUs, smart and intelligent cameras now offer systems developers complete machine-vision systems in a single package. At approximately $5000, these cameras lower the cost of integration and relieve systems integrators and developers of the burden of choosing individual cameras, frame grabbers, and software.

However, while smart cameras prove cost-effective in numerous applications, they might prove less useful when developers require open, more configurable machine-vision or image-processing systems. Smart cameras are mainly being deployed in applications that require easily definable processing tasks. For example, determining whether bottles are full, labels are correctly placed, or parts are accurately counted on a production line are tasks now being tackled by smart cameras.

Because smart cameras generally represent completely embedded vision systems, developers considering using them need to weigh the same factors as they would in purchasing a complete machine-vision system. These factors include sensor resolution, type of embedded processor, development software, machine-vision application programming interfaces (APIs), programmable I/O, video output, and networking.

If the machine-vision application can be precisely defined, then smart cameras provide a low-cost alternative to PC-based systems. Once deployed, however, they present more complexity to change the configuration or add greater flexibility in terms of I/O, networking, and processing speed. Where applications suggest an open architecture, systems developers may be better off choosing a general-purpose machine-vision system.

Taking a RISC

In the design of smart cameras, vendors are embedding RISC, DSP, and Pentium-class processors to perform general-purpose machine vision. Often coupled to field-programmable gate arrays, these processors also allow machine-vision software to be embedded in the cameras, resulting in highly integrated camera/ processor subsystems. But while the functionality of smart cameras may at first seem similar, the processors used in their designs and the software offered to support them vary widely.

Although the most popular general-purpose processor used in these designs is the Motorola PowerPC, the Intel x86 family is also used (see table on p. 96). These CPUs provide camera designers and systems integrators with a wealth of software-development tools. Camera designers wishing to push performance further have also embedded a number of DSPs, ranging from the SHARC from Analog Devices (Norwood, MA, USA) and the 320 series from Texas Instruments (TI; Dallas, TX, USA) to the TriMedia processor from Philips (Eindhoven, The Netherlands). Although the development tools offered with these processors are company-specific, vendors have also leveraged these types of processors and image-processing APIs into their smart-camera offerings.

Targeting the process-automation market, the iMVS series of cameras from Fastcom Technology (Lausanne, Switzerland) are dubbed machine-vision systems rather than smart cameras. As the highest-resolution member of the series, the iMVS-155 incorporates a 1024 × 1024-pixel CMOS image sensor, an Analog Devices ADSP 21061 SHARC floating-point processor, a 4.5-Mbyte SRAM, RS-232 and RS-485 interfaces, and eight digital I/O lines. To program the camera, developers use C language and program- ming tools from Analog Devices. Together with the company's iMVS run-time library, this system enables the sensor, serial interfaces, digital I/O, and VGA interface to be programmed for a specific application.

Vision systems

Because smart cameras perform the functions of complete machine-vision systems, they are often offered with a number of digital I/O, serial, video, and networking capabilities. Using these capabilities, systems developers can configure smart-camera systems in shop-floor applications that direct the host camera/system to make pass/fail decisions, control lighting, and interface to other peripherals such as programmable logic controllers (PLCs). And, because many of today's factory-automation applications require networking capability, smart cameras are being equipped with Ethernet or Profibus communications.

FIGURE 1. Profibus-enabled VS-710 vision sensor from Siemens is an 80486-based camera featuring a 768 × 580-pixel CCD, digital I/O, a 16-Mbyte DRAM, and SVGA output. Provision software and an OCR/OCV software package allow inspection programs to be created off-line on a PC running Windows and then to be tested and loaded into the VS-710.

Also called a complete image-processing system, the Profibus-enabled VS-710 vision sensor from Siemens (Grand Rapids, MI, USA) is an 80486-based camera featuring a 768 × 580-pixel CCD, digital I/O, a 16-Mbyte DRAM, and an SVGA output (see Fig. 1). To develop image-analysis programs for the camera, the company offers its Provision software and an OCR/OCV software package for character reading and verification. With ProVision, inspection programs can be created off-line on a PC running Windows and can then be tested and loaded onto the VS-710. Using the Profibus, several-VS 710s can be configured with a single PC.

For developers of factory-automation systems considering using smart cameras in quality-control applications, perhaps their most important decision is evaluating the type of software being offered by the smart-camera vendor. These programs range from development systems for embedded processors coupled with machine-vision APIs to full-blown machine-vision software development packages. Depending on the end-user application (and often volume), low-cost smart cameras with limited machine-vision capabilities might prove more viable than expensive ones with sophisticated capability.

High volume, low cost

Smart cameras offered for less than $1000 are useful in high-volume applications such as OCR, where the limited machine-vision tools provided by the vendor are of no consequence. Recognizing this situation, some vendors have embedded their software in rebranded OEM camera products that target markets such as data-matrix readers and OCR/OCV. In doing so, they have invested many man-hours of software-development time in creating end-user-like products that can be sold in volume.

FIGURE 2. Vision Components VC2065 is a 782 × 585-pixel camera that embeds the TI TMS320C6211 processor, a 16-Mbyte DRAM, and a Flash memory. With an SVGA video output and an 8-bit color overlay, the smart camera is supplied with the company's VCLIB image-processing library.

For example, Vision Components (Ettlingen, Germany) supplies a range of smart cameras for such developers. The company's latest product, the VC2065, is a 782 × 585-pixel camera that embeds the TI TMS320C6211 processor, a 16-Mbyte DRAM, and a flash memory (see Fig. 2). With an SVGA video output and an 8-bit color overlay, this smart camera is supplied with the company's VCLIB image-processing library. To understand how to use the library, the company provides on-line access to relevant documentation.

For systems developers facing a complex machine-vision problem, long software-development time is often costly. Therefore, these developers should explore smart cameras with sophisticated PC-like, and often menu-driven, machine-vision software. In the smart-camera market, however, these software packages are manufacturer-specific, and their functions can range widely.

The result of a joint development carried out by Asentics (Siegen, Germany), JAI (Glostrup, Denmark), and Stemmer Imaging (Pucheim, Germany), the ThinkEye TE-100 smart camera is based on Stemmer's Common Vision Blox (CVB) machine-vision library. This software gives developers the option of building an application base on standard edge-detection or blob-analysis tools or implementing their own code. The PowerPC-based camera can be programmed over Ethernet using the CVB Image Manager running on a host PC.

Software benefits

Popular general-purpose machine-vision and image-processing software packages, such as the Matrox Imaging Library (MIL) from Matrox Imaging (Dorval, QC, Canada), Vision Builder/LabView from National Instruments (Austin, TX, USA), and Vision Foundry from Data Translation (Marlboro, MA, USA), are at present not available for embedding in smart cameras. Rather, systems developers using smart cameras must evaluate company-specific APIs and software-development kits before they begin designing their application. Recognizing this, some smart-camera vendors allow developers to freely evaluate their software to determine whether such applications can be accomplished. When successful, systems developers can then deploy the software in smart cameras.

For example, DVT Corp. (Duluth, GA, USA) offers a number of smart cameras, including the Series 500 and Legend Series. To program these smart cameras, developers can download, free-of-charge, the latest version of the company's FrameWork machine-vision software and develop machine-vision platforms on PC-based hosts (see Vision Systems Design, July 2003, p. 5, and Fig. 3). When the application is proven, the software can then be uploaded and deployed across any of the company's products.

FIGURE 3. To program DVT's Legend and Series 500 smart cameras, developers can download the latest version of the company's FrameWork machine-vision software and develop machine-vision platforms on PC-based hosts.

In the near future, increases in performance and density of high-speed processors, DSPs, memories, gate arrays, and peripheral components will drive the cost of smart cameras even lower. Providing more cost-effective platforms for systems developers, smart cameras are already evolving into smart machine-vision systems with imaging, PC functionality, and PLC-like capabilities.

Already, Coreco Imaging ipd Division (Billerica, MA, USA) offers several "vision appliances" targeted to specific machine-vision applications. Although not "smart cameras," per se, the development of camera/ processor tethered-based systems is proof of the potential of the smart-camera market.

Company Info

Analog Devices www.analog.com
Asentics www.asentics.de
Data Translationwww.datx.com
ipd division of Coreco Imaging www.goipd.com
Matrox Imagingwww.matrox.com/imaging
National Instrumentswww.ni.com
Philipswww.philips.com
Stemmer Imagingwww.stemmer-imaging.de
Texas Instrumentswww.ti.com