The so-called machine vision (MV) is a combination of different technologies and methods to automatically extract image information, which provides operational guidance and key data for machines to perform given tasks in industrial and non-industrial environments.

Recently, the latest research on the machine vision market by IoT Analytics shows that the advances in cameras, AI and chipsets are promoting the use of machine vision applications. These advances have enhanced the typical advantages of machine vision, such as saving costs, improving competitiveness or improving product quality, and also affected the prospects of 39 machine vision applications identified by IoT Analytics in the research.

The research in 2022 shows that machine vision is expected to experience a sustained and strong investment inflow in the next few years. At the same time, machine vision also has the highest return on investment (ROI) and the fastest amortization time among all industrial 4.0 technologies.

Although machine vision technology has existed for more than 30 years, the recent technological transformation still provides a new impetus for its industrial application.

Three technological advances

Key technology change 1: advanced camera

At present, cameras with a resolution of more than 45 million pixels are not only superior to human eyes in many cases, but also can track moving objects at a very high speed without distortion. However, there are other innovations that may not be obvious to users, one of which is the introduction of "Event-based vision sensors". Similar to the way the optic nerve processes information, the event-based vision sensor captures images only by detecting the brightness change of each pixel. Compared with the traditional frame-based vision sensor, it can also have better effects in darker environment or worse weather conditions.

Key technology transformation 2: better decision-making AI

The transition from rule-based machine vision to AI-based machine vision is influential. Generally, rule-based machine vision is a stricter description, which is only applicable to quantifiable, clear and very specific features, for example, the scratches on the product are horizontal and 30 mm in length. On the contrary, AI-based machine vision can provide accurate results for features that cannot be quantified, and can deal with changes in product appearance and defect types more flexibly. Deep learning is a more complex and powerful subset of AI, and it is also increasingly used in machine vision applications.

Key Technology Transition 3: More Powerful Hardware and AI Chip

The progress of the chip is closely related to the progress of AI. The latest chip is very powerful, suitable for processing images and running AI-based computer vision algorithms. These performance improvements will help shorten the training time of deep learning from a few weeks to several hours. Many smart cameras today are also equipped with powerful AI chips.

Seven upcoming machine vision applications

According to the market research of IoT Analytics, seven machine vision applications are being vigorously promoted due to the latest changes in cameras, AI and chips mentioned above. Among the 39 use cases analyzed, these seven use cases are marked as "particularly interesting".

1、defect detection

Defect detection is a machine vision use case, which is mainly deployed in the process of quality inspection. In the past, non-AI machine vision needed a database containing all possible defect images in order for the system to successfully identify defects. However, today's machine vision technology can identify some defects and detect anomalies without specific images. Take Fujitsu Japan factory as an example, through training AI to repair abnormal areas in thousands of simulated images with defects (such as abnormal shape, size and color), normal images are generated, which reduces the number of hours required for factory inspection of printed circuit boards by 25%. When AI does not detect a specific type of anomaly accurately enough, it can produce more simulation pictures of this type of anomaly, thus accurately improving the weakness of the model.

2、Process/operation optimization

Another use case of machine vision related to manufacturing is process/operation optimization. For example, robots can now complete complex tasks with higher precision and efficiency than humans. As a result, with the help of machine vision, robots or other machines can perform operations in different ways or accomplish things that could not be done before. For example, the new rubber grinding solution developed by Flawn Hof-Institute of Mechanical and Electrical Integration of Design and Engineering (IEM) is to develop a new AI grinding system using mitsubishi electric manipulator, optical laser scanner and control system equipped with AI software, which makes the process of grinding complex rubber-like materials automatic. According to the team, the new method can shorten the time spent in the rubber grinding process by up to 40%.

3、Automatic driving

Machine vision plays a vital role in the process of developing fully-automatic driving cars. There are six levels of automatic driving, from level 0 (fully manual) to level 5 (fully automatic). Nowadays, most commercial vehicles are still providing level 1 or level 2 assisted driving, and only a few can provide level 3. To reach level 4 or level 5, the technology used by vehicles must make a leap, and the very complicated machine vision system and AI computing are part of the leap. Google Waymo One's automatic car calling service is an example of level 4 automatic driving. Every car is equipped with Waymo driver system, which is a complex MV system, consisting of five lidar, four radars, 29 cameras and AI software, which can collect sensor data and calculate the best route in real time. The solution has collected more than 20 million miles of real driving experience data.

4、pallet size labeling

In the field of logistics, one of the upcoming use cases of machine vision is pallet dimensioning. The innovative 3D time-of-flight technology makes it possible to measure the dimensions of loading pallets, eliminates the time spent on manual measurement, and minimizes the potential cost of carriers due to inaccurate dimensions and weights.

5、Attitude/Motion Analysis

Machine vision has also achieved some new applications in the field of medical care. The improvement of camera precision and quality makes it possible to analyze body posture and motion. Now, you can identify the position and direction of bones and joints only by using a camera without additional equipment. Ergonomics, orthopedics and other medical care and gesture interaction can all benefit from this machine vision application. Using the new industrial camera developed by German camera manufacturer IDS, DIERS, a biomedical solution company, developed a solution that can measure the back, spine and pelvis of human body quickly and with high resolution. By using the camera to continuously record the light projected on the back of patients, it can generate an accurate representation of the curvature of spine, thus helping orthopedic surgeons to detect the imbalance or posture defects of muscle system.

6、Automatic checkout

By using the solution based on machine vision, the time required for checkout can be significantly reduced and the automatic checkout experience of retail stores can be improved. Some start-ups have developed a machine vision solution that can scan products without searching bar codes. Because the automatic checkout solution reduced the queuing time, the transaction volume increased by 34%.

7、Identification of pollutants

The identification of pollutants is an important part of food quality assessment, but this process is difficult to be solved by traditional methods because a database containing all possible combinations of pollutants is needed. However, discoloration, foreign bodies and other abnormalities in processed food can be effectively identified through AI. For example, Apeto, a frozen food company, has tested and deployed automated qualitative assessment solutions on more than 20 production lines to ensure that the processed food department can successfully detect all pollutants in raw materials.

Write it at the end

IoT Analytics predicts that the machine vision market will grow by 8% CAGR from 2022 to 2027. Like many technical fields today, it is expected that the software that benefits from the progress of AI will grow fastest. The research of IoT Analytics also shows that about 60% of the identified 313 machine vision suppliers have provided specific machine vision software. It is expected that the above seven use cases will become more common in the next few years, and more other use cases will also appear.

Transfer from: Machine Vision Salon This article is only for technical information exchange. If there is any infringement, please contact to delete it.