Machine Vision

Engineers are adding machine vision systems to a number of industrial applications to reduce costs, increase throughput, and improve customer satisfaction. The reference architecture featured here shows one possible path from cameras and lighting through enterprise-level communication. All machine vision systems include a combination of hardware and software to acquire and process images, usually resulting in a response from a secondary system connected to the inspection system. There are two possible designs for this reference architecture. One includes a low-cost, rugged, embedded solution while the other makes use of the power of a desktop PC to acquire and process images at a higher rate and resolution.

Cameras/accessories - This layer contains all of the hardware required to create the correct environment for the image capture, as well as the camera itself. The necessary pieces vary but usually include a lens, filter, light, and encoder.

I/O connections - Encoders, lights, and trigger signals are usually routed through some sort of digital I/O. The camera also has one of many standard interfaces, including GigE, IEEE 1394, and Camera Link.

Image acquisition/preparation - Image acquisition is the act of pulling the image data from the image sensor (or camera) into a machine vision system for processing. The first step is to prepare the image by extracting different color panes, as well as conduct some filtering on the image.

Image processing - This layer features all of the software algorithms for image processing (including OCR, edge detection, and blob analysis). It also includes the hardware to implement this processing.

System feedback - Once the images are processed, some feedback usually occurs, either to reject a bad part or to log an image of a failed or passed inspection locally.

Enterprise - This includes any feedback to an enterprise-level system, such as an FTP server set up to log images of failed inspections so improvements can be made to the manufacturing process.