In response to the rapid development of the robot industry and industrial automation, machine vision technology and related applications that combine camera and image processing to replace or reinforce manpower continue to heat up, and have been widely deployed in the field of industrial inspection and manufacturing. However, for everyday life and other areas beyond traditional applications, machine vision still has unlimited potential and business opportunities.
Over the past few years, this knowledge-oriented industry has become increasingly complex with the components and modules of machine vision systems; on the other hand, market applications related to machine vision systems continue to expand beyond new applications beyond industrial manufacturing, especially It is now that as hardware size shrinks and embedded systems continue to show kinetic energy, emerging technologies such as upcoming nanotechnology, advanced sensors, machine-to-machine (M2M) communication systems, and the Internet of Things (IoT) will further Drive machine vision applications to include consumer electronics, wearable devices, automotive advanced driver assistance systems (ADAS), and intelligent surveillance that are closer to the realm of the masses.
According to a survey released by market research firm MarketsandMarkets last year, the global market for machine vision systems and components in 2012 exceeded $3 billion, and is expected to grow at a compound annual growth rate (CAGR) of 8.2% between 2013 and 2018. It reached $5 billion by 2018. The new business opportunities brought by the global machine vision market have become the focus of industry players.
Richard Kingston, vice president of investor relations and corporate communications at CEVA, revealed that the company has licensed computer vision DSPs to eight companies, including three OEMs in the mobile space. And Tom Wilson, vice president of business development at CogniVue in Canada, is also optimistic that "there will be very strong growth opportunities in the wearable device and automotive sectors."
The global machine vision market is expected to surpass the $5 billion mark by 2018.
Visual processing to 3D spanning
With the official release of Project Tango by Google, 3D machine vision has become more popular. Kingston said, "The 3D processing in the consumer sector is primarily aimed at 3D imaging, natural user interface (NUI) and 3D vision applications such as PCs, notebooks, tablets, smartphones and other consumer devices."
The main reason for the industry to pursue 3D vision is to solve the inherent limitations of 2D machine vision, and to more effectively achieve segmentation (separation of close-range and distant view), illumination (for face recognition), relative position (objects in the scene), etc. The ability to make more applications use 3D spatial information to simplify and improve the precision and reliability of the vision system.
However, whether it is a 3D sensor (such as a TOF camera) or a stereo sensor implemented with two 2D image sensors, the processing capability is also higher. Wilson pointed out that "stereo matching (using the input of two image sensors) requires a difference mapping to generate a 3D depth of field map. This is a very difficult computer vision problem, and the academic community is actively researching optimized stereo recognition algorithms." Since each method that implements 3D sensing has a performance tradeoff, CogniVue is currently developing a new algorithm that is expected to calculate its disparity map for low-cost 3D sensors.
Handling large amounts of real-time data requires intensive computing power. It is very difficult to achieve stable 3D sensing mapping, especially for low power devices. To this end, he emphasized that "CogniVue's APEX Image Recognition Processing (ICP) technology can play a key role in 3D vision applications where power is limited." Vice President, Investor Relations and Corporate Communications, CEVA
Designing a hardware that can effectively perform different visual algorithms is a daunting challenge for system designers. When the system vendor selects an image/video processing solution, it can choose to focus on the CPU, or uninstall some of the image processing work to the GPU, or add hardware logic for image processing. In applications that require 3D processing, the GPU has enabled a portion of the system to perform multiple computer vision algorithms, helping to share the workload for the general purpose CPU.
"The i.MX6 has powerful GPU computing power. Its 3D engine GC2000 includes 4 rendering cores, which can provide up to 30GFLOPS of computing power and support OpenCL 1.1 EP," Freescale Microcontroller Division Asia Pacific Market Li Xingyu, marketing and business development manager, said, "In addition, i.MX6Q also has a dedicated 2D engine (1Gpixel/s) and a vector graphics processing engine."
For the field of image capture, 2D camera sensors or other optical sensing technologies are often used to analyze and calculate 3D data. In addition to time-of-flight (TOF) 3D imaging (continuously transmitting light pulses for the target, calculating the target distance based on the time between the light pulse and the sensor receiving the reflected light), it is widely used in industrial manufacturing inspection. The 3D laser triangulation method differs in that the 3D laser sensor uses point-by-point scanning, while the TOF camera simultaneously obtains depth information of the entire image.
Compared to the 3D imaging method of the laser triangle, National Instruments National Instruments (NI) technology marketing engineer Huang Xiangyu introduced that NI LabVIEW can provide binocular stereo vision for 3D vision applications. Engineers can install two cameras at different angles of the object. Then, use the calibration technique to adjust the pixel information between the two cameras and retrieve the data, and perform mathematical analysis through the LabVIEW 3D tool library. Designing 3D vision applications through a seamlessly integrated graphics and development environment, software and hardware, simplifies the work of engineers.
The perfect combination of IP and processor
In order to integrate embedded vision systems in increasingly sophisticated machines or devices, more advanced CPUs are needed for intensive computing to handle large amounts of data. Huang Xiangyu emphasizes that this will continue to be powerful enough for CPU performance, power consumption and resources. Visual analysis operations present challenges.
CongmiVue mentioned on the official website, “In order to meet the ever-increasing application requirements, the processor architecture has continued to evolve over different stages over the past few decades. CPUs for desktops and servers in the 1980s; DSPs in the 1990s Accelerated audio codec and wireless/wired voice/data encoding and decoding requirements; 2000 GPUs achieved more advanced performance and parallelism for 2D and 3D imaging; today, embedded vision processing requires completely different processors Architecture: ICP."
Tom Wilson also pointed out that "a new generation of vision applications requires more than 100 times the embedded visual performance / power consumption, in order to meet the performance and power requirements of these applications, we must upgrade the same power consumption higher than the traditional processing architecture 100 - 400 times performance," so CogniVue emphasized that its APEX ICP technology can achieve this performance requirement.
Together with the APEX ICP core, APEX programming tools and an APEX-CV embedded vision library, it can cover a wide range of vision applications. For example, feature detection and matching are available for wearable (enhanced reality) and automotive (optical flow and motion tracking). He added: "In addition to common computer vision features, CognVue offers more advanced products for specific applications, such as FaceVue for face recognition, MoTIonVue for motion monitoring in surveillance applications, and FrontVue for car lane departure warning. And SideVue for car blind spot monitoring."
For intensive computing needs, CEVA's CEVA-MM3 series, including the CEVA-MM3101, uses a very efficient and powerful vector engine to achieve the massive parallelism required for computer vision. In addition, the integrated power conditioning unit (PSU) enables dynamic voltage regulation within the processor to assist in the 'permanent line' application type.
In addition to CogniVue and CEVA, manufacturers working on machine vision IP include Mobileye, Tensilica (now part of Cadence IP) and ImaginaTIon Technologies. ImaginaTIon's synthesizable ISP IP core based on 'Raptor' was launched in the first quarter of 2014, making the IP camp increasingly competitive.
A number of semiconductor manufacturers have launched their dedicated vision processors through collaboration/authorization with professional IP vendors, including Freescale, Texas Instruments and STMicroelectronics. ST and Mobileye jointly developed the EyeQ3 image processor for pedestrian detection; Freescale introduced the image recognition processor family SCP2200 based on CogiVue core; Israel Inuitive developed a 3D vision processor based on dual CEVA-MM3101 engine to realize 3D depth of field; Xilinx leverages MTLec's HALCON and Silicon Software's VisualApplets development platform to create an end-to-end Smarter Vision development environment for the Zynq-7000 All Programmable SoC.
Among TI's wide range of processor products, KeyStone's multi-core processors feature 5.6GHz ARM and 9.6GHz DSP processing power, and have lower power consumption than multi-chip solutions for camera applications in machine vision; The company's Jacito 6 series of automotive processors can also be used for machine vision functions such as pedestrian identification and anti-collision warnings in ADAS.
For Xiaomi Glass,Xiaomi Mi A2 Oca Glass,Touch Screen Front Glass For Xiaomi,Xiaomi 11 Front Glass
Dongguan Jili Electronic Technology Co., Ltd. , https://www.ocasheet.com