Machine Vision Places Demands on Framegrabbers

By: Chris Brais, Imaging Technology

So many framegrabbers are available in the current market that it can be a daunting task to decide which one is best suited for a particular machine-vision application (see Figure 1). The key to the problem is identifying and understanding the requirements placed on a board by such vision applications as inspection and process control. Determining the capabilities necessary for a specific particular application permits an informed choice, and can determine the success or failure of a project (see Questions to Ask Before Starting).

FIGURE 1: Machine-vision systems (top) require framegrabber boards (bottom) with accurate A/D conversion circuitry, control circuits to drive camera timing, and capability for triggering, strobe control, frame-reset, or input signal conditioning.

Demanding applications
Machine-vision applications require different capabilities than multimedia, medical, or even most military applications. In one of the most common vision tasks, the system acquires an image when triggered by a certain event, loads the image into the memory of the host computer, analyzes the data, and makes a decision based on established criteria (see Figure 2). For example, when a ketchup bottle on a conveyor belt passes a triggering sensor, the system acquires an image and sends it to the computer, where a computer-based algorithm assesses the data and determines whether the label is properly placed or the bottle is sufficiently full.

Click here to view Figure 2.

Inspection rate, algorithm details, and physical constraints often determine the speed, resolution, and type of camera and optics used. Although RS170 or CCIR cameras are common, the demand for more detail and faster acquisition rates has been driving machine vision in the direction of higher-resolution, progressive-scan cameras.

To the factors above, add the need for triggered acquisition, frame reset, strobe control, and digital input/output, and the requirements are far beyond the capabilities of any multimedia framegrabber. On the other hand, not every machine-vision system needs the full array of options. Until recently, a framegrabber incorporating any of the features necessary for machine vision would cost over $1000, and a fully-featured framegrabber could be even more expensive. Thus, engineers must tread carefully when choosing a board.

The system determines the requirements
Currently, most machine-vision systems incorporate analog cameras. With few exceptions, vision applications require an analog-to-digital (A/D) converter capable of operating at or above the 10 Mhz rate of RS170. Systems must also incorporate accurate A/D conversion circuitry and either a very good phase-locked-loop or the ability to provide a clock to the camera, driving the camera timing. Without one of these control options, the fidelity of the acquired image is in question, compromising both quantitative and qualitative applications. Multimedia boards do not perform these tasks, though virtually all machine-vision boards do.

As previously discussed, most vision applications require that an image be acquired as a result of an external event, a trigger. A framegrabber should not only provide triggering capability, it should also allow users to coordinate the firing of a strobe, or flash, to light the subject during image acquisition. Depending on the specific application, the user may also need the freedom to adjust or delay the strobe pulse relative to the trigger event.

Inspection applications often require frame-reset capability, which allows camera timing to be interrupted and fixed to the trigger signal at any point. Frame reset guarantees image acquisition immediately upon receipt of the external event. In cameras without reset ability, a delay occurs between the time the trigger signal is issued and the time at which the camera begins to acquire and send data to the framegrabber. To compound the error, the delay is variable, because the trigger signal and camera timing are not synchronized. For interlaced RS170 signals this delay can vary from almost zero to the duration of an entire frame (33 ms).

Input signal conditioning such as gain and offset control are important to minimize effects from camera variability or lighting fluctuations, and to provide the best camera signal to the A/D converter. Also, look-up tables (LUT) at the input can be used to modify data for easier signal processing or better display capabilities.

Multimedia boards are not equipped to handle these tasks. Not even all machine-vision boards have the capability for triggering, strobe control, frame-reset, or input signal conditioning, so users should choose their framegrabbers accordingly.

On-board memory
Many machine-vision applications require that the host computer control several devices and manage several events. In recent years, the technology has been trending toward interrupt-driven software and multitasking. The ideal solution thus involves a framegrabber and software package capable of using interrupts and multiple threads, and compatible with an operating system like Microsoft Windows NT.

To lower manufacturing costs, framegrabber manufactures rarely put memory on the framegrabber itself. Such boards rely on the speed of the PCI bus to transfer the acquired data line-by-line to the host memory (see Figure 3). Although this method works and can in some cases achieve real-time or close to real-time capture rates, it presents several drawbacks in the context of machine-vision.

FIGURE 3: In a system without onboard memory, the framegrabber must transfer the data to the host computer line by line.

Machine-vision systems do not simply acquire an image for display, they perform several tasks at once or in rapid sequence. Tasks include displaying images, reacting to interrupts, processing input and output, and analyzing acquired data. Framegrabbers without on-board memory increase the load on the CPU—during operations, the host CPU is simply busy handling the line by line data transfer and may not be available for the required processing. To achieve real time performance, framegrabbers for machine-vision systems must have on-board memory (see Figure 4).

FIGURE 4: In a system with onboard memory, the data cascades onto the computer, freeing the process up for data calculations.

Software
When considering framegrabber software, designers must first determine how much of the software integration task they're willing to assume. The choice of whether to write a custom algorithm or use an off-the-shelf package will depend upon the uniqueness of the application and the time constraints of the project. In general, using software packages written specifically for machine vision not only speeds the completion of the project, but also results in a faster, more reliable solution.

One of the most common bottlenecks in system implementation is the integration of the host-based software with the framegrabber. To address this difficulty, users must give strong consideration to the maturity of the interface between the framegrabber software and that of the image-processing package under consideration. The software of many framegrabber manufacturers is simplistic and requires that common machine-vision functions be written by the end user.

Basic framegrabber software functionality should include routines for triggering, strobing, frame reset, data transfer to host, and image display, as well as a way to dynamically tweak the camera interface. These capabilities are invaluable when first setting up the system to acquire an image. Once the desired image is acquired and delivered to the host memory, the task of actually solving the machine-vision problem begins.

Access to a powerful image-processing library designed for machine vision, not medical or general-purpose image analysis, can greatly improve performance, decrease development costs, and improve success rates. If you are using an off-the-shelf machine-vision package, be certain that it supports the framegrabber you choose.

There are so many framegrabbers on the market for different intended applications, it is important to understand the real-world demands that machine vision imposes on both the host computer and the embedded framegrabber. Selection of the wrong board often results in project failure. In today's market, once users understand the real demands of machine vision, they can purchase an excellent machine-vision framegrabber at a reasonable cost.

About the author…
Chris Brais is Board Product Manager, Imaging Technology, 55 Middlesex Turnpike, Bedford, MA 01730. Phone: (781) 275-2700; Fax: (781) 275-9590.