Smart cameras lower machine-vision costs
With embedded charge-coupled devices (CCDs), processors, and software, the latest smart or intelligent cameras offer systems developers complete machine-vision systems that can be networked or attached to host PCs. These cameras lower the cost of integration and relieve the design burden for systems integrators of choosing individual cameras...
With embedded image processors, smart cameras lower vision-system integration costs and promote network and host PC interfacing.
By Andrew Wilson,Editor
With embedded charge-coupled devices (CCDs), processors, and software, the latest smart or intelligent cameras offer systems developers complete machine-vision systems that can be networked or attached to host PCs. These cameras lower the cost of integration and relieve the design burden for systems integrators of choosing individual cameras, frame grabbers, and software. However, whereas these cameras may prove cost-effective in some applications, they prove less useful when developers require open, easily programmable, machine-vision or image-processing hardware.
Of the dozen or more smart cameras available, not one is offered by such well-known suppliers as Sony, Toshiba, Hitachi, or Basler (see table). While these companies certainly have expertise in camera design, they have, to date, avoided the smart-camera market. They prefer to offer products that can be interfaced to PC-based frame grabbers using broadcast, digital, or high-speed analog interfaces. Choosing this design option allows camera manufacturers to produce volume products that can address a number of application areas in microscopy, machine-vision, medical, and military applications.
In the field of machine vision, certain applications can be easily defined, such as barcode inspection, parts location, gauging, and assembly verification. In these applications, well-known image-processing algorithms, such as image thresholding, normalized gray-scale correlation, and geometric pattern-matching, can perform the desired tasks. The vendors of smart cameras seem to believe that these tasks can be automated by embedding application algorithms to lower the cost of systems development.
For machine-vision system developers, the choice of which smart camera to purchase is limited by the number of vendors producing such devices, operating systems, development tools, embedded processors, and off-the-shelf software. Only four of the available smart cameras were designed in the United States, the others were developed by European companies. Of the US-based developers, none have chosen to incorporate standard x86-compatible processors to perform on-board image processing; instead, they have opted to use digital-signal processing (DSP), reduced-instruction-set-computer (RISC), or neural-network-based processors.
In Europe, the trend toward using x86-based CPUs in smart cameras has allowed such vendors as DS GmbH (Remseck, Germany), Leutrek Vision (Burlington, MA), and Siemens (Norcross, GA) to integrate complete PC-based image processors into their cameras. This approach allows both vendors and users alike to take advantage of the numerous software tools, interfaces, and off-the-shelf software packages available to PC developers. In many cases, PC-based camera vendors also allow their software-development tools to be used off-line. This allows systems developers to create applications on stand-alone PCs, before deploying them in smart cameras.
Recognizing the benefits of unifying PC-based software in an intelligent camera, DS GmbH has combined a PC-based camera and its NeuroCheck industrial image-processing software into a new smart camera called the NeuroCheck Compact (see Fig. 1). Offering embedded software that is identical to NeuroCheck means that systems developers who have in the past deployed NeuroCheck on PC-based systems with separate frame grabbers can now migrate their software to the NeuroCheck Compact camera.
Besides offering serial, parallel, and universal-serial-bus interfaces, the NeuroCheck Compact camera provides built-in Ethernet networking, which enables camera integration into company networks and remote-maintenance through the Internet. If developers want to display the images being captured, they can connect standard PC monitors using the camera's VGA interface. Available with resolutions from 640 x 480 to 1280 x 1080 pixels, the camera comes with a 266-MHz MMX-compatible processor and 256 Mbytes of SDRAM. Its preinstalled software allows systems development to be performed both on- and off-line.
Leutrek Vision also offers a PC-based smart camera dubbed the PentiCam. Based on a combination of notebook PC technology and PMC-size modules, PentiCam is an integrated vision system that provides image capture using a range of standard and nonstandard cameras (see Fig. 2). The camera design is based on separate image and processing modules. To configure this camera/processor combination, developers can choose from a range of PMC-based PC/frame-grabber combinations and 13 different camera heads.
Once configured, the PentiCam supports Windows NT/2000, Windows NT Embedded, Windows 95/98/Me, and Linux software, eliminating the need for dedicated development tools. Running under these operating systems, third-party software packages, such as Halcon and Activ Vision Tools from MVTec Software GmbH (Munich, Germany), Common Vision Blox from Integral Vision (Farmington Hills, MI), and NeuroCheck from DS GmbH, can be accessed using the camera's keyboard interface.
Whereas many European vendors offer open x86-based cameras, US manufacturers have chosen to base their designs around RISC or digital-signal processors. Incorporating such devices limits the US smart cameras to development tools of the processor vendor and third-party software. In many machine-vision applications, however, the nondeterministic nature of operating systems, such as Windows, may limit the predictability of smart-camera operation, and it may prove better to choose a smart camera that incorporates a real-time operating system.
The MachineCam digital area-scan camera from Wintriss Engineering (San Diego, CA) is a good example of such a design (see Fig. 3). Developed around a Motorola PowerPC processor with 64 Mbytes of memory, the VxWorks real-time operating system from Wind River Systems (Alameda, CA), and Ethernet connectivity, the MachineCam camera contains an image sensor that captures up to 30 frames/s at a resolution of 640 x 480 pixels. It also provides RS-232 and RS-422 communications outputs, a 10Base-T Ethernet connection, and NTSC outputs for connection to a monitor.
The Wintriss Engineering interface software allows connection to Windows-based computer systems for camera initialization, inspection setup and testing, and inspection definition controls. In operation, the camera can stand alone in go/no-go inspection applications running custom programs or operate with DT Vision Foundry from Data Translation (Marlboro, MA).
Common among available smart cameras is Ethernet compatibility. Originally pioneered by DVT Corp. (Norcross, GA) in its range of SmartImage sensors, the concept of integrating smart sensors into the factory floor using off-the-shelf networking standards has now been adopted by most machine-vision vendors.
Targeting the factory floor, however, cannot be done with Ethernet technology alone. Indeed, many standards now exist that make such integration nontrivial. These include Modbus, Profibus, DeviceNet, and ControlNet (see Vision Systems Design, May 2001, p. 80).
As a result, DVT has developed a stand-alone, 640 x 480-pixel resolution imaging sensor with on-board image acquisition, processing, and digital I/O capabilities. In addition to providing serial and Ethernet TCP/IP communications, the DVT Series 600 SmartImage sensor incorporates Modbus TCP, Modbus ASCII, and Modbus RTU, along with 12 digital I/O lines and RS-422 communications.
For connectivity to other factory-floor devices, DVT has partnered with such robotic and motion-control companies as Motoman, Fanuc, IAI, Parker Hannifin, Ormec, and ABA to establish connectivity between their devices and the SmartSensor. To program the SmartSensor, developers can use the company's free Framework software development kit. This includes DVT's SoftSensor tools such as pattern finding, blob analysis, measurement, and 1-D and 2-D barcode readers.
Recently, DVT unveiled its Legend Series 500 smart camera (see Fig. 4). Like other cameras from the company, the Series 500 camera is Ethernet-compatible and can measure, count, find features, and compare similarities within images. In addition, the Series 500 reads most 1-D and 2-D barcodes and printed characters using a trainable optical character recognition algorithm.
The latest product from Cognex Corp. (Natick, MA), the Insight 1000, also incorporates Ethernet compatibility in a processor-based vision sensor. The InSight 1000—part of the company's Insight family of sensors—is similar to the InSight 2000 and 3000 vision engines in that it uses an embedded PowerPC. To reduce costs, however, no DSP is installed to perform image-processing functions. Instead, all the supplied software tools, including Cognex's PatFind and spreadsheet-driven software, are run on the PowerPC (see Vision Systems Design, April 2000, p. 47).
In the past, when deploying a smart vision system, systems integrators had to choose specific cameras, frame grabbers, and host computers. Now, with a number of Ethernet-compatible smart cameras available, systems integrators can spend more time developing machine-vision software solutions, lowering the integration time of such systems, and reducing development costs. In the future, distributed imaging systems are expected to use high-speed communications networks to distribute images rapidly and use local or remote hosts running proprietary and/or off-the-shelf software.