When choosing machine vision software, please keep the following 9 points in mind.
1. Camera Selection
When choosing vision software, the first consideration is determining whether it is compatible with the camera best suited for your application. Low-cost analog cameras are readily available, but typically, applications require more than just VGA resolution; frame rates faster than 30 frames per second are needed, along with overall image quality superior to standard machine vision cameras.
NI's hardware and software are compatible with thousands of cameras, from low-cost USB3Vision to high-speed line scanning and thermal imaging.
2. Hardware scalability
Choosing the right camera is a crucial step in any application; however, camera scalability is another important consideration. As camera technology advances rapidly, you may one day need to upgrade your camera to improve image quality or measure other functions. NIVisionAcquisitionSoftware is driver software that works with all NI image acquisition cards and connects to thousands of cameras through an easy-to-use interface. Even if you change CameraLink camera vendors or switch between over 100 different GigEVision, IEEE1394, and USB3Vision cameras, your software remains unchanged.
National Instruments driver software works with thousands of cameras and all NI hardware platforms, such as PCs and PXI/CompactPCI, compact vision systems, industrial controllers, smart cameras, and CompactRIO controllers. This means you can prototype your application in the lab on a PC with an inexpensive camera and then deploy it to the production floor on a rugged NI vision system without changing the acquisition or image processing code.
3. Software usability
Once you have the image, the next step is to process it. Finding the right tool through trial and error in programming languages, using the selection of current algorithms, can be tedious and inefficient. With that in mind, you need vision software tools to help you get the most out of algorithms.
For many applications, you don't need a programming language to build a complete machine vision system. While not as flexible as programming in C, Visual Basic, or NI LabVIEW, configurable software like NI VisionBuilder AI provides an easy-to-navigate, interactive environment for configuring, benchmarking, and deploying machine vision applications. VisionBuilder AI includes nearly 50 popular machine vision tools such as pattern matching, OCR, DataMatrix readers, and color matching. It can acquire images from any NI-supported camera and communicate with other devices using Ethernet, serial, or common industrial protocols to inspect the results via digital I/O.
While programming vision applications is more complex than programming with VisionBuilder AI, NI makes application development easy and intuitive in LabVIEW, C++, and Visual Basic with NI Vision Assistant. Included in the NI Vision Development Module, Vision Assistant is a prototyping environment that allows you to interactively try out different vision functions to see what works for your application and how long each function takes to run.
Once you've determined how best to meet your application challenges, VisionAssistant generates immediately runnable code with a simple click of a button. You can complete most vision applications before even entering a single line of code. You can run the code generated by VisionAssistant independently or add it to larger industrial control, data acquisition, or motion control systems.
Whether you are a vision novice or an expert vision integrator, VisionAssistant helps you create efficient and reliable vision applications in less time.
4. Algorithm breadth and accuracy
When choosing vision software, it's crucial to ensure that the software tool can correctly and accurately measure important sub-pixel details or object features. If the software is inaccurate and unreliable, then computer speed and camera pixel count are irrelevant. Remember, creating accurate code is easier than making fast code more accurate.
The Vision Development Module and VisionBuilderAI include hundreds of accurate and reliable vision functions. Below are five of the most common machine vision application areas and the most popular algorithms.
Enhance images – use filtering tools to sharpen edges, eliminate noise, or extract frequency information. Use image calibration tools to eliminate non-linearity and perspective errors caused by lens distortion and camera placement. You can also use image calibration tools to apply actual units to measurements, so the tool returns values in micrometers, millimeters, or miles instead of pixels.
Presence check – this is the simplest type of visual inspection. To check if a part or feature exists, you can use any color, pattern matching, or histogram tool. Presence checks always result in yes/no or pass/fail.
Positioning features – Positioning features are crucial when aligning objects or determining precise object placement, serving as the standard for all subsequent checks. Edge detection, grayscale pattern matching, shape matching, geometric matching, and color pattern matching are all tools that can be used for positioning features. These tools reduce object position (X, Y) and rotation angle to one-tenth of a pixel. Geometric matching is unaffected by overlapping objects or objects with varying scales.
Measurement Functions – The most common reason for using vision systems is to perform measurements. Typically, you use edge detection, particle analysis, and geometric functions tools to measure distances, diameters, totals, angles, and areas. Whether calculating the total number of cells under a microscope or the angle between the edges of two brake calipers, these tools always return a number, not a position or pass/fail value.
Part identification – Part identification is crucial for part compliance, tracking, and verification. Direct identification methods include reading barcodes or data codes such as DataMatrix and PDF417. Newer methods use trainable OCR or object classification. Part identification typically produces text or strings rather than measurements or pass/fail determinations.
All Vision development modules and VisionBuilderAI features utilize subpixel precision to insert position, distance, and measurement values, down to one-tenth of a pixel and one-tenth of a degree.
5. Heterogeneous processing
One of the biggest advances in machine vision is processing power. With processor performance doubling every two years and continued focus on parallel processing technologies such as FPGAs, vision system designers can now apply highly sophisticated algorithms to visualize data and create smarter systems.
The performance improvements mean designers can achieve higher data throughput for faster image acquisition, use higher-resolution sensors, and fully utilize some of the latest cameras on the market that offer the highest dynamic range. These performance gains not only help designers acquire images faster, but also process them more quickly. Preprocessing algorithms (such as thresholding and filtering) or processing algorithms (such as pattern matching) can be executed much faster. This ultimately enables designers to make decisions based on visual data faster than ever before.
Unfortunately, one of the biggest challenges in implementing FPGA-based vision systems is overcoming the programming complexity of FPGAs. Vision algorithm development is inherently an iterative process. Designers know in advance that they will have to try different approaches for a given task. Most of the time, they need to determine which approaches don't work and which work best; the "best" approach varies depending on the application. To maximize efficiency, immediate feedback and benchmark information for the algorithm are needed, regardless of the processing platform used. Viewing algorithm results in real-time can save significant time when using iterative exploration methods. However, traditional FPGA development methods can slow down innovation due to the compilation time required between each design change to the algorithm. One way to address this is to use algorithm development tools that help you develop for both CPUs and FPGAs from the same environment without getting bogged down in FPGA compilation time. By combining NIVisionAssistant with the vision development module, designers can develop algorithms that can be deployed on either CPUs or FPGAs. VisionAssistant can also be used to test algorithms before compiling and running them on the target hardware, while providing easy access to throughput and resource utilization information.
6. Integration with other devices
If you've ever worked on a vision application, you'll know that vision is often part of a larger control system. In industrial automation, your vision application might require:
The actuator classifies the products.
The inspection results are communicated to the robot controller, programmable logic controller (PLC), or embedded system.
Save images and data to a web server.
Transmit inspection parameters and results to the local or remote user interface.
Typically, for scientific imaging applications, you must combine vision and motion phases, data acquisition systems, microscopes, specialized optics, and advanced triggering.
As a leading supplier of industrial control, data acquisition, and motion control products, NI designs vision products that can be used with these and other common components. Whether you need to communicate with a PLC via DeviceNet or with a microscope via a serial bus, you can use NI vision products for communication.
7. Price
There are many types of vision packages. Many companies cater to OEM customers by splitting their development libraries and selling algorithms separately. While the cost of each algorithm package may seem low, the total cost of a vision development package is often high. In addition, the licensing costs for each component, as well as application deployment, become complex and expensive.
The Vision Development Module contains all the algorithms you need to tackle the most demanding vision challenges, allowing you to avoid researching, purchasing, and maintaining multiple software packages. Furthermore, deploying applications is very inexpensive – with a single vision deployment license, you can deploy executables using any number of vision algorithms. In addition, the NI CompactVisionSystem includes all the licenses needed to deploy your applications. Therefore, regardless of how much NI CompactVision System you use, you only need to purchase one copy of the VisionDevelopment Module or VisionBuilderAI.
8. Partners and integrators
NI manufactures hardware and software for image processing and machine vision. Because NI does not manufacture lighting, cameras, and optics, the company works closely with other specialists.
Over the past decade, NI's vision products have helped meet thousands of different application challenges, from inspecting automotive parts to assisting cancer research. While NI vision tools are designed for end users, larger applications may require expert vision assistance. To help you with application development, NI collaborates with over 1,000 NI Alliance partners who can help you select the right components or build a complete turnkey solution for you.
9. Technical Support
While NI vision software is designed to be easy to use, getting help when you need it is crucial. NI sells vision software directly to its customers and provides direct product support. Resellers or third parties are never intermediaries between you and qualified NI engineers. When you need an expert, you can contact one of hundreds of application engineers by phone or email.
