Information Chain in Machine Vision Systems
Ensuring high quality in manufacturing often requires visual inspection of products. In the industrial world, machine vision has become a marked asset amid efforts to combat loss of human-intensive work to developing countries. Recent rapid developments in camera technology has made machine vision especially a hot topic. Machine vision is not based solely on camera technology, however. Implementing a smoothly functioning machine vision system requires control of lighting, optics, automation and information interfaces as well as environmental conditions
Information Chain in Machine Vision
The machine vision information chain can be divided into five distinct areas: lighting, optics, image digitizing, image processing and information interfaces to business logic (Fig. 1). In addition feedback from image processing can be applied to control lighting and optics. Unfortunately machine vision systems often fail to work because of failure in one of these subsystems. Even the most sophisticated image processing algorithms can not overcome inadequate lighting. Similarly correct lighting will not guarantee a working result if the processing time does not fulfil the speed requirements. The most important aspect in designing a machine vision system is to control the overall quality of the information chain.
Lighting and optics
Most research in machine vision emphasises computer vision algorithms. Research on lighting and optics in machine vision has been narrower in its focus. However controlling the lighting is undeniably the basis of the entire machine vision system. The quality of the information through the process is based on how an object is seen during the imaging. If features of interest are not visible during the imaging, the result will fail regardless of the methods used later in image processing. Lighting and optics strongly affects the image quality. The imaging parameters and quality measures for machine vision system are contrast, resolution, field-ofview, depth-of-view and image distortions.
Light transports information from the physical world to the camera. The optical system gathers the light and produces the image on the imaging plane. Lighting must produce adequate contrast so that features of interest can be extracted from the image. Shape of the object, surface materials and environment affect what kind of lighting should be applied. Different types of light sources and lighting techniques create various kinds of images of the object (Fig. 2, 3). Some of the commonly used lighting techniques in machine vision systems are dome lighting, coaxial lighting, low-angle darkfield lighting, backlighting and structured lighting. Controlling the polarization of lighting also plays additional role with highly reflective surfaces.
Image digitizing: camera
The selection of the camera has an effect on image digitizing and processing time, image quality, size of the equipment and cost of the overall system.
Both CCD- and CMOS-cameras are used in machine vision systems. Cameras are typically equipped with greyscale- or colour photocells. The size of the cell may vary from an area of 300,000 pixels up to 11 million pixels or more. Some applications require the use of line scan cameras, which are equipped with up to 12,000 pixel wide line array photocells.
In addition to traditional PC-based cameras, machine vision system can also be implemented using smart cameras (Fig. 4). A smart camera is a self-contained system that includes a processor to perform image processing in the camera. Results can be communicated directly to plant automation via camera’s integrated interfaces. During the last decade smart cameras have become the industry standard for many applications due to their simplicity, reliability and price. Nevertheless PC-based machine vision systems are still a valid choice especially for multi-camera and processor-intensive solutions.
Image processing: application
Camera manufacturers and software vendors provide machine vision software development environments and application programming interfaces that include common image processing tools and algorithms like filtering, segmentation, object and feature recognition, measurement, OCR/OCV and code reading tools.
Implementing a machine vision system requires developers to apply provided tools with application specific elements to come up with a solution that corresponds to the system requirements. Developers are required to understand the fundamentals of a wide range of image processing algorithms. Often the selection of algorithms directly affects the accuracy and cycle time of the application.
In most of the machine vision systems the software application can be divided into four areas of procedures: image enhancement, image segmentation, feature extraction and feature analysis.
In a software implementation the same results can often be achieved using very different approaches in algorithm selection. In addition most of the algorithms have many different variations within the group, which may vary on precision and processing time.
Machine Vision Systems can be considered intelligent sensors in automation. As part of a distributed automation system they not only make measurements but also highly refine the results and make decisions independently.
Most common information interfaces between machine vision and other systems are digital I/O, fieldbus and Ethernet. Via I/O- and fieldbus machine vision systems integrate with PLCs or can even run in place of a small PLC to control the actuators directly. Ethernet-interface enables flexible integration with graphical user interfaces, supervisory control and data Acquisition or manufacturing execution system -levels for data miningand quality control purposes.
Case: Food Package Inspection
HK Ruokatalo Oy is one of Finland’s largest food producers. In their Eura plant chicken packaging cell capacity can reach 35 000 units per day. A machine vision system was delivered to ensure correct labeling and uniform packaging quality. The design of the system was conducted with the aim of overall control of the information chain described earlier.
Fluorescent lamps are used in a brightfield lighting solution due to their nature of diffuse illumination. Random glares from plastic wrapping are eliminated by using polarizing filters.
Two smart cameras are used for the machine vision system – one camera for overall package inspection and the other for detailed label inspection. Cameras with standard 640 x 480 area CCD photocells produce adequate resolution for inspection setups.
One camera is programmed to inspect for package edges and position/rotation of the label. The other camera is programmed to check the correctness of the printed barcode and the label template. The correctness of the label template is checked by reading an identification key from the matrix code (Fig. 5).
I/O-interfaces of the cameras are connected to the packaging cell’s PLC for trigger information and inspection results. An Ethernet interface is used to integrate machine vision into supervisory control and data acquisition system. During a change in product the cameras are automatically set for the correct barcode and label settings according to manufacturing execution system. (Fig.6) Statistical information generated from the inspection of defective packages is saved for quality control purposes.