Technological upgrading is the core driving force for the development of the machine vision industry

The cost reduction and efficiency increase brought about by technological upgrading is the core driving force for the development of the machine vision industry. From the perspective of the number of global machine IS410SRLYS2A vision-related patents, as of 2019, the cumulative number of global machine vision patents reached 86,000, the year-on-year growth rate of the number of new patents during 2010-2019 continued to maintain positive growth, and the number of new patents during 2017-2019 maintained a growth rate of about 17%. With the gradual increase of patented technology, the continuous improvement of global machine vision technology, and the further expansion of the comparative advantage of machine vision over artificial vision will further open up the development space of the industry.

The system of technical standards has been gradually improved, and the industrialization of core technologies has been accelerated. At present, the standardization organizations in the industry mainly include ISO/IEC JTC 1/SC 24, ITU: SG16, IPC, IEEE, G3 International Machine Vision Standardization Organization AIA, EMVA, JIIA, VDMA and CMVU, G3 has published a number of industry technical standards for common, safe and reliable, Devices, interfaces and testing, acceptance and evaluation, industrial interconnection and other aspects have been clearly related to the technical standards, national standards, industry standards and group standards have also been published. IS410SRLYS2A The step by step unification of the technical standards for the development of the industry is conducive to the continuous growth of domestic machine vision in the benign development track.

Embedded vision system, deep learning, 3D vision and computational imaging are the four technological evolution directions of machine vision. While there is still much room for improvement in the imaging quality of pre-machine vision, embedded vision system and computational imaging mainly solve this problem. Embedded technology can integrate AI modules with deep learning algorithms and image processing functions into industrial cameras. In recent years, the application of embedded machine vision has increased rapidly, and the demand for consumer electronics, automatic driving, life sciences, agriculture and other scenarios has been growing, driving domestic enterprises to increase R&D investment in embedded machine vision. According to the data of the Machine Vision Industry Alliance, the annual compound growth rate of investment in embedded vision system research and development of domestic enterprises between 2018 and 2020 reached 41.4%. Both deep learning and 3D vision are visual analysis technologies that can optimize traditional algorithms and provide richer dimensional information to help machine vision improve the intelligence level of image processing. From 2018 to 2020, the R & D investment in AI-driven solutions and 3D solutions will grow at a CAGR of 40.8% and 49.7%, respectively, and R&D investment will continue to grow at a high rate. In summary, embedded vision system, deep learning, 3D vision, and computational imaging are the four major technological upgrading routes of global machine vision.

IS410SRLYS2A

IS410SRLYS2A Embedded technology provides key value enablement for smart industrial cameras. Intelligent industrial cameras are an important means to achieve edge intelligence in the field of industrial automation. Embedded technology can be used to integrate AI modules of image processing and deep learning algorithms into industrial cameras to achieve edge intelligence. The most important applications include ADAS, industrial automation and security monitoring. Through the integration of AI chips, smart cameras can realize image processing in specific application environments and make logical judgments with embedded artificial intelligence algorithms, providing intelligent solutions without human intervention for automated scenarios.

Deep learning techniques can help increase the universality of traditional approaches and speed up the process of adapting new scenarios. At present, the traditional machine vision technology has advantages in speed and accuracy compared with eye monitoring when the structured scene is relatively fixed and the work content is highly repetitive. However, with the continuous expansion of the downstream application field of machine vision, The traditional method presents the disadvantages of IS410SRLYS2A low generality, hard to copy and high requirement for users in complex and subtle scene processing. Deep learning is to convert the original data features into a higher-level, more abstract feature representation through multi-step feature conversion, and further input into the prediction function to get the final result. In simple terms, deep learning technology can give the flexible intelligence of the human eye to the traditional way in the complex detection environment in a new field.

The research progress of 3D vision at home and abroad is approaching, and the domestic application scenarios are more abundant. Compared with 2D vision, 3D vision can obtain the spatial coordinate information of objects, which is more in line with the upgrading trend of industrial control with increasing degree of refinement and automation, but each of them has its advantages and disadvantages. 3D vision cannot completely replace 2D vision due to cost. At present, standardized 3D vision hardware and software products have entered the market, and the industrial chain has been preliminarily formed. In the research of 3D vision algorithms, domestic and foreign enterprises have a similar starting time, and the research progress is still in the initial stage. At the same time, there are more 3D machine vision application scenarios in China, and new scenes represented by automobiles, electronics, batteries, etc., continue to emerge, and domestic enterprises have a natural and more high-quality environment for 3D technology application research.

Accurate imaging is a prerequisite for the success of machine vision, and computational imaging technology can break through the functional and cost limitations of traditional photoelectric imaging. How to balance the cost of the solution and the complexity of the optical path to maximize the imaging effect is the core problem that each machine vision manufacturer needs to solve first. Computational imaging technology is to add a “calculation” process to the traditional “WYSIWYG” optical imaging method, that is, by establishing a modulated model between the target scene and the observed image, and then using different computational methods to reconstruct the enhanced imaging process, which can break through the problems of traditional photoelectric imaging devices, functions and cost limitations. Meet the internal needs of higher resolution, greater depth of field, and more dimensions of photoelectric imaging. At the same time, computational imaging makes it possible to super-diffraction imaging, lens-free imaging, large-field high-resolution imaging and clear imaging through scattering media, which helps to promote the high performance, micro and intelligent imaging equipment.