Why is the development of machine vision so fast, the reason is actually these

This title reads a bit, but it does reflect the fact that today, in addition to human eyes, there are more and more "eyes of the machine" that help us observe, record and analyze the world around us, let us actually The visual ability is greatly expanded and enhanced. At present, in China, the surveillance cameras installed by the public security system have exceeded 20 million units, and the monthly data traffic reached 7500 PB. Taking into account the surveillance cameras installed in homes and individuals, we love the visual sensing devices such as reversing images and driving recorders on the car, as well as GoPro's consumer-grade digital visuals that pull the wind... Unconsciously, we have In the ocean of a machine vision.

As the name suggests, the core value of machine vision is to replace the human eye and the human brain to perceive and analyze image and video information. Although machine vision is not a new concept, it has achieved rapid development in the past 10 years and still maintains considerable acceleration for the foreseeable future. A development of machine vision has benefited from three factors.

The first factor is Moore's Law. From a hardware perspective, a machine vision system consists of two core devices: a CMOS image sensor (camera) and a processor. Both of these can be manufactured through standard CMOS semiconductor processes, so they are put on the "magic" of Moore's Law, in the seemingly endless pursuit of "three lows and one high" (low power, low cost) The small size, high performance) is rushing all the way, constantly reducing the cost of machine vision. Think about the resolution of an entry-level mobile phone camera today, which may be comparable to that of a mid- to high-end SLR camera image a few years ago. This is probably the most intuitive way for users to use Moore's Law.

Moore's Law is also driving improvements in processor performance, making it perfectly suited for complex image processing calculations. In the choice of processor hardware architecture, today's machine vision developers have a variety of choices: you can choose a DSP optimized for image processing; you can also choose ARM + GPU or other image coprocessor platform; and based on ARM + Heterogeneous processing architecture for FPGA programmable logic (such as the Xilinx Zynq 7000) is available. Even the mainstream ARM general-purpose processor platform, combined with optimized software algorithms, can also be used in many machine vision applications. It is only a matter of time before the user wants to get a more cost-effective machine vision processor.

The first two factors promote the rapid development of machine vision is increasingly rich algorithms and software resources. It can be said that the hardware "kidnapped" by Moore's Law reduces the threshold for the use of machine vision, but if you want to make the machine work like "human eyes + human brain", even higher and more efficient, you must have software. Cooperate. In the last century, the algorithm and software for machine vision was definitely a brain-burning job. There are not a few Ph.D in the company that dare to open. This situation has changed since 2000 - that year Intel released OpenCV, an open source cross-platform computer vision library based on BSD licenses. Developers can easily implement many images and images through a series of C/C++ functions. A general algorithm for visual processing. Since then, based on the ever-changing OpenCV library, functions and algorithms optimized for different machine vision applications have evolved and are easier to port and run in embedded processors, resulting in a complete machine vision software. ecosystem. At the same time, many commercial software development tools have begun to integrate visual processing functions, making machine vision application development more accessible.

Figure 1. The Blackfin Embedded Visual Learning Development Kit from Avnet, including complete hardware and software resources, helps machine vision developers get started quickly.

It can be said that the maturity of the hardware and software ecosystem has created a rapid expansion of the machine vision layout in the past decade. And then, becoming more "smart" is the core appeal of machine vision development. In this process, the third element will play a vital role, it is "artificial intelligence." Using the core technology of artificial intelligence such as deep learning, machine vision will gain the ability to self-learn evolution, continuous iterative enhancement, and the more "smart" it is.

The combination of artificial intelligence and machine vision, the more classic approach is to transfer the collected data to the cloud, training a strong "brain" with the ability of data analysis and self-learning evolution in the cloud. At the same time, some people are considering today, relying on the ability of increasingly powerful machine vision terminals, training the terminal directly, allowing the deep learning algorithm to land on the terminal products to obtain better real-time, accuracy and reliability. It also avoids the privacy and security issues that may exist in the cloud. No matter what kind of thinking, its successful application will undoubtedly have a profound impact on the future of machine vision.

In short, in the world of machines, the work of the human eye—at least part of the work—has become a simple and repetitive “work of strength.” Under the combined effect of many elements, machine vision has replaced or even surpassed the process of human vision. Can't stop.