Accelerate the development and deployment of machine vision applications

Machine Vision is Critical to Industrial Automation

The manufacturing and industrial world is undergoing tremendous changes, and emerging technologies such as the Industrial Internet of Things (IIoT) are expected to significantly improve factory operation efficiency. As a result, various sensors are being deployed at an unprecedented rate, allowing IIoT applications to sense and gain insight into everything that happens in real-world scenarios such as production processes, industrial automation, and quality control. Although environmental sensors, liquid flow sensors, and pressure sensors can provide information in many aspects, one of the most important human senses that IIoT systems need to fully manage in an ongoing process is vision. Providing computer vision, or machine vision, requires relying on fast image and video processing technology and artificial neural network platforms.
Machine vision is everywhere

When vision is introduced into an industrial manufacturing or assembly process, the potential applications for machine vision become nearly limitless. As a result, machine vision systems are ubiquitous throughout the industrial sector, meeting a wide range of needs. For example, machine vision can detect whether a bottle of shower gel is full or whether a label is directly affixed to the correct location. It can also prompt an actuator to push the bottle into a reject bin if the label is not affixed properly or the bottle is cracked, broken, or deformed. Another example is industrial robots that can automatically assemble complex mechanical parts. Machine vision tasks may include confirming that parts are properly aligned for assembly and then checking that the parts are securely installed before moving on to the next process.

Considerations for Machine Vision Implementation

There are many factors to consider when implementing a machine vision application. First, the development team must determine whether their system requirements can be met with simple image processing techniques, or if the task is more complex and better accomplished using a deep learning neural network.

• Simple image processing techniques include edge detection algorithms, thresholding techniques, and the use of low-pass, band-pass, or high-pass filters on the images captured by the camera. The benefit of using these techniques is that only low to moderate computing resources are required, which means that job throughput is not affected. The techniques listed previously are useful in many manufacturing and process automation scenarios. For example, it can be used to check whether an industrial robot has properly put a cap on a bottle. Machine vision can use edge detection algorithms and high-pass filters to perform this task. If the cap is not on the bottle, the high-pass filter will show dark pixels. Thresholding techniques separate color from the background, so that pills in blister packs can be identified and counted. In addition, machine vision systems can use similar methods to determine whether each pill is the correct size during the manufacturing process.

• If the machine vision task is more complex, such as reading a product's part number, developers can implement an artificial neural network to infer text characters and numbers. In this way, the design work becomes more complicated and requires training a neural network model to quickly, reliably, and correctly recognize letters and numbers.

However, perhaps the most important consideration for designers is the image processing speed and computing task latency that are limited by the processing speed of the production line. To ensure design and implementation flexibility, the machine vision platform should also adopt different image and video protocols and frame rates to make the platform scalable and flexible to support a variety of applications.