Machine Vision: Existing Trends and Challenges

In ecommerce, these solutions are paired to pick items to fulfill customer orders; in manufacturing, products are picked and placed into packing and then inspected for quality control; in the medical field, they are being utilized to pick up instruments on demand and used to help identify potential organ degradation the human eye may not detect; in agriculture, the picking, inspection, and packaging of agriculture are all accomplished with machine vision paired with robotics. The applications machine vision can be used with are almost limitless.

The challenge with machine vision being used to accomplish certain physical tasks is the environment. There are significant variations to all physical objects that, as humans, we have learned to interpret through years of trial and error, but also through the use of additional senses. For example, when we see an object, we know what its texture feels like. We have touched plastic, leather, metal, and paper before. We can identify that it has a smooth or rough surface even through different lighting variations because our brains intuitively utilize our previous experience with these objects to determine the best method of picking them up. When a machine vision application is determining what an object is and its orientation, it is looking for distinguishing features that can be associated with edges and contours. It does this by turning on variable lighting or sensors that are paired with the application, which gives it the best chance of success in determining these features. However, if the lighting is positioned incorrectly, there is dust or water causing the lighting on the object to be distorted, or the item has reflective material, the machine vision application will have difficulty detecting the object features correctly.

These challenges are less of an obstacle when the machine vision application operates in the same lighting and environmental conditions and is used to identify the same object. Contrary to an application that must identify and handle multiple, if not thousands, of different items, like in a supply chain, the success rate declines drastically. In applications that have consistently changing variability, technology leaders must incorporate complex machine learning into their application while spending the appropriate amount of time teaching the application what it is looking at. As future iterations of the application are tested, the system can leverage the attributable parameters to make the best determination on how to fulfill the expected output.

Machine vision will remain in high demand to be used in applications that automate a process. To implement the application correctly, it takes considerable time, focus, and flexibility to implement a system that can perform the expected function. Technology leaders must be willing to embrace machine vision for all its capabilities and shortfalls as technological enhancements continue to add value to machine vision applications.