Computer vision became a huge topic in recent years. It can be applied for a variety of cases: object detection and classification, movement recognition, traffic monitoring, agriculture, etc. The application of computer vision technology for quality assurance in manufacturing is increasingly in demand. Major manufacturers already invest in research of the computer vision solutions for their plants in order to gain competitive advantage using intelligent technology for quality control.
However, although the topic is hotly discussed, most are unaware of the true potential of computer vision applications in the quality assurance process. There exist a number of claims about quality inspection with computer vision that do not (anymore) reflect reality.
Here I would like to present five most common myths about the application of computer vision for visual quality inspection.
1. Computer vision system needs hundreds of images to be trained
Computer vision solutions are based on the same principle: they use a database of images to detect objects through the camera. What used to be the state of the art, hundreds to thousands of images of an object were needed for the system to recognize it correctly. Moreover, so that the system could detect the defects such as scratches, dents, cracks, etc., you had to additionally upload the pictures of the product with such defects. This tedious process can now be avoided thanks to the new approach. The model can now be trained using only five to ten images. Researchers from the University of Waterloo in Ontario even claim that they will develop a technique that will allow the model to be trained with only one image .
Furthermore, for visual quality inspection, pictures of the defective parts are not required anymore. FotoNow’s system generates the examples of the faulty parts using the built-in dataset and an AI-based algorithm.
For the doubters: this approach provides very accurate results. The model, which is trained using only five images, recognizes objects with precision of over 99%.
2. Installation lasts several months
When you think of installing a computer vision solution to your plant, you have complicated calculations and project planning in your mind. First, a consultant must analyze the use case and decide which and how many cameras, what lighting, stabilization and additional hardware need to be installed. The server must be set up and the connection to it must be established. In addition, the third party must ensure the connection of the computer vision technology to the industrial system by specific hardware. The whole process of consulting, installation and integration takes a certain time.
Currently it is possible to minimize the installation time to one day. Flexible software that can be installed on any camera device and does not require additional hardware takes only a few minutes to a maximum of hours to install. FotoNow performs the inspections in the whole field of view of the camera and therefore the position of the camera device does not need to be calculated and defined in advance.
3. An IT-Department is needed to operate the technology
Typically, computer vision solutions for industrial application are thought of as a complex and difficult-to-use technology. In the end, the company benefits only to a limited extent from the use of automated quality control, because new employees with IT skills have to be hired or even an entire department has to be built up to maintain the complex AI solution.
However, it is no longer true that you need an IT expert to install, operate and maintain the solution. The complex processing of files and image inspections run in the background and the user does not have to apply any programming skills. The training of the system and the application can be conducted by any employee. The system is so automated that all you have to do is upload a few images and place the camera device so that the production line is in the camera’s field of view. The interface is designed very intuitively, so the operation of the program can be compared to the use of any app that we use daily for personal purposes.
In addition, the process of integrating computer vision into the industrial environment has been simplified so that it can be done with one click. Similar to how you connect a device via Bluetooth.
4. Only industrial cameras can give the needed level of performance
Very often you can hear, even from AI experts, that the use of industrial cameras is crucial for the application of computer vision for quality assurance. Several industrial cameras with built-in light must be connected by a cable to the production line so that it stops when the product is in front of the camera to take a sharp picture of the part. This is how most people imagine the ideal picture of visual quality inspection with computer vision.
With the latest development of FotoNow’s technology, it is possible to install the computer vision software on any camera device without additional hardware. For a few years now, cameras on smartphones have good resolution, which makes them attractive not only for personal but also for industrial use. With the technology of enhanced resolution, specially developed by FotoNow, smartphones can perform visual quality inspection with the precision of over 99%. The proprietary algorithms of FotoNow allow highly precise inspections with changing lighting, high-frequency vibrations and production line in motion. With this technology, it is possible to perform quality assurance on the whole production line with only one smartphone and without additional hardware. In addition, the physical connection to the production line is unnecessary because the image acquisition happens completely without stopping the line.
5. Computer vision can be applied only on simple production lines
Computer vision is smart, but not smart enough to identify products that look similar but have different configurations. Especially when it comes to assembly verification, it is important to distinguish between different products on the same line. For example, the doors of different car models. Some producers even have personalized products that also need to be configured differently. It is claimed that this task is too difficult for standardized computer vision technology.
This is also no longer true. FotoNow’s computer vision solution can very easily identify the part based on its code. It then can apply the appropriate for this part correct configuration during the image analysis. For this purpose the developers of FotoNow have programmed their own OCR engine. In conjunction with the proprietary OCR, the software can perform quality inspections on the lines with hundreds of different products in real time.
Many manufacturers are reluctant to install an automated visual quality inspection solution to their production plants for these very reasons. They think computer vision is not for them and miss out on so many benefits this technology has to offer. However, computer vision can be much more flexible and widely used than you might think.
 Karen Hao: A radical new technique lets AI learn with practically no data. https://www.technologyreview.com/2020/10/16/1010566/ai-machine-learning-with-tiny-data/