Artificial intelligence tide hit, the market dividend began in both artificial intelligence and intelligent voice, with a set of performance, power and cost-optimized artificial intelligence chip will become the AI 'empowerment' application of the key, but also to create AI burst Today, in addition to smart speakers, machine vision applications are emerging in the security and other industries, commercial applications of machine vision has begun to come true, a batch of face recognition based on machine vision products are being rolled out.
Among the sixteen sessions of the China International Public Security Expo (Aba Fair) recently held, machine vision products targeting face application have become the theme of the exhibition. Face grabbing and recognition technologies and products are extremely hot. Among them, Haikang Fluorite on display fluorite C5S wall-mounted Internet camera, lovingly show the cloud-AI camera, as well as the new Internet, Wan Jiaan, Wang control, micro-krypton display face recognition cameras and other new concern, these products have Depth learning capabilities that enable multi-target detection, 1080P resolution, and 30-frame real-time video, and the artificial intelligence chips that power these products are the first generation of machine vision chips T01 from Junzheng.
T01, Jun is the beginning of machine vision 'from 0 to 1'
Based on the development and application of camera needs, increasing resolution, linkage speed requirements are getting higher and higher, while the front-end chip computing power is obviously insufficient, in this application bottleneck, though after years of development precipitation, but have not seen Low-cost front-end real-time face recognition camera applications.
However, love and fluorite in the near future were released front-end real-time face recognition camera products, and the official debut at this security Expo, the product broke the front-end chip computing power and linkage speed requirements of the development of application bottlenecks, much industry People concerned.
Face-loving camera on display at the 16th Ann Fair
Which brings breakthrough technological innovation is derived from Jun Zheng T01 machine vision chip.
Driven by technological innovation, industrial change, or 'from 0 to 1', is an important part of JUNZZ's corporate vision. Therefore, it has the origin of T01's name.
First look at the basic specifications of T01:
★ XBurst CPU
★ CV Engine V1.0
★ Multi input / output interface type support
★ Host interface
★ Typical 700mW
Core performance, T01 detection specifications are as follows:
★ simultaneous detection of multiple objects, support up to 16 simultaneous detection of objects
★ multi-target detection at the same time, support for more than 30 simultaneous detection of the target
★ multi-resolution configuration support, up to 1080p resolution real-time monitoring
★ Common types of data input and output interface support
★ 30 frames full real-time detection rate
Not difficult to find, in the specifications of the T01, neither ISP nor H.264 / H.265, there are many complicated redundant external interface.Yes, T01's only task is to complete high-quality target detection.
As the first generation of machine vision dedicated chip launched by Junzheng, XBurst CPU and CV Engine are the core modules among them. XBurst CPU naturally inherits Junzheng's traditional CPU technology and performs instruction optimization for deep neural network. CV Engine Is the first generation of machine vision ASIC engine.
T01 and common AP SOC or MCU in the past is completely different, which is the king is in the direction of the machine vision from 0 to 1 the first step, but also Junzheng Roadmap on a new beginning.
Superior performance, clear positioning, with large-scale conditions
T01 multi-object target detection is not limited, can be for all industries, not just limited to security.If you want to give priority to choose a landing direction, it should be from the beginning, focus on people's attributes, expand the industry application.
For face, half-length human and full-size humanity, these three types of attribute goals, the current T01 on a single attribute detection accuracy rate of more than 97%, the detection speed of 30 30 frames per second simultaneous detection in the detection range, with different Of the lens measured normal human face detection can reach the farthest distance of 30 meters, body detection can reach 100 meters or more.
'Support for multi-object hybrid cross-detection, which is T01's first advantage.' Single-object detection, if there is no detection of powerless cross-detection of multiple objects, complement each other between different coverage, significantly reducing the overall undetected Probability.
'The speed with which real-time detection can be done, and the ability to support accurate angle detection are another big advantages of the T01.' 'For face recognition, the ability to grab a good-looking face at an angle is decisive for the face recognition experience T01 provides full real-time grasping ability, and can simultaneously analyze the detection of the perspective of the object, so that products can be smoothly from the consistent stream of video screening a best quality positive face, and then call the cloud to identify the most Excellent results. Based on T01 speed detection advantages, with the human angle detection, with tracking technology, but also greatly deepen and expand the application scenarios, such as the number of statistics, behavioral analysis and so on;
It is because of these unique capabilities of T01, which also provide an innovative foundation for love and fluorite partners.Then how to partner based on T01 innovation? We know that in the electronic product design applications, want to create A new product must invest a lot of manpower and financial resources and time, and based on the previous mature products continue to improve, small run is a more pragmatic approach.And T01 is the combination of this approach to the coprocessor form, Easy to add, easy to integrate, greatly reducing the engineering innovation and product innovation brought about by the engineering complexity and a variety of burdens T01 through collocations with other common camera main control chip rapid formation of the product.In the system structure, T01 more transparency High, just simple hardware circuit transformation, you can easily T01 integrated into the other main chip system.
For example, loving products, typical T01 with Jun Zheng T20, the product structure is as follows:
The program uses Junzheng T20 + T01 two chip combinations, which, T20 as a master intelligent video chip, with professional imaging ISP, low light night vision; SmartH.264 encoding engine, ultra-low bit-stream control; high performance computing Platform, visual intelligence analysis; leading low-power technology. In addition, T01 is mainly for video intelligent high-performance analysis applications to enhance the video processing capabilities, this chip can be used with the Jun-self T series products can also be With other manufacturers of the master chip, is currently on the market unique products.
Thanks to the performance and positioning advantages of T01, T01 can easily be combined with existing products in a modular way.Only through partial modification, the T01 data path can be increased without affecting the maturity and stability of the original product and achieving high Quality of the target detection.In addition, T01's periphery only requires a very simple circuit, you can work together.From the perspective of technical engineering and application cases, T01 parties have matured and have a large amount of conditions on the mass, as Junzheng T01 put into volume production, the current domestic love and fluorite, fresh Internet, Wan Jiaan, Micro Krypton, Haiqing video have been released using T01 face recognition camera, and has been on the market for sale in the product Overseas markets, there are also some well-known overseas manufacturers have mass-produced or are developing intelligent TOT-based IOT cameras, robots, smart conferencing systems and other products.
Open the Blue Ocean Strategy, T01 or Achieve 2018 face recognition burst
Today, the commercialization of machine vision is accelerating and accelerating the industrialization of smart cities, smart stores, smart homes, smart communities, smart workplaces and a new generation of smart robots. These commercialization processes are becoming the eyes of all things Dangdang important interface, the application of a wide range of prospects in the Silicon Valley venture capitalist Peter Tier's book, 'Road One' is a key step to open up the emerging Blue Ocean market, Jun is the T01 chip is also the Blue Horizon Machine Vision An important step in the application, highly competitive.
Why T01 machine vision chips in face recognition front-end chip has a competitive advantage? T01 is used because the algorithm is embedded directly into the processor, with the ASIC way, while reducing costs, power consumption is also the best Balance, and have enough performance.
Algorithm, depth learning is also the focus of AI landing applications, T01 is different from the GPU and FPGA, but also with the deep learning neural network various XPU chips have some differences.Tio typical operating power consumption, including built-in DDR, about 700mW. Much lower than the GPU and FPGA, but also significantly lower than the number of published data XPU.
Compared with other machine vision chip stacking FLOPS performance, T01 pay more attention to the practical function of the results from the T01 actual test point of view, the target detection effect is already very close to the depth of learning, and with angle analysis, real-time detection is superior to the current stage end Deep learning characteristics deep neural network is undoubtedly the future of the industry, but the depth of learning to achieve large resolution and high frame rate, there is still a long way to go, especially in power control T01 with very low Power consumption, lower cost, and good program usability are the products of choice for deep learning at the moment.
In the rapid iteration of artificial intelligence, pure cloud computing and front-end computing alone can not meet the demand for development, and future products must be more balanced with experience and cost, so a joint computing architecture of terminals and the cloud emerges. T01 will target real-time high-end detection at the front-end, and will be based on the identification of large data running in the cloud, the terminal and the cloud to achieve joint computing. 'Focus on their respective functions through the terminal and the cloud will be very conducive to the unity of the AI industry chain At the same time, the architecture of the two-terminal joint computing is technically easy to cut algorithms, cloud services and hardware products, and it is commercially beneficial to bridge the data scarcity and promote integration. So far, T01 has been working with a number of cloud service companies such as Baidu, Artofus Etc.) to provide all kinds of cloud data processing services; the same time, T01 also introduced a large number of equipment product development partners can quickly create a variety of different scenarios based on the scene, with the corresponding cloud services to provide overall scenarios under different scenarios .
In addition to the current introduction of T01 artificial intelligence chip products such as lover's eyes and fluorite, fresh interconnection, Wanjia On, Han Hui, Micro Krypton, Hai Qing Video, British Gas, SBR and so on, T01 still has many potential applications and customers , For example: with the multi-camera linkage, you can achieve personnel positioning control, trace back, face search, etc .; with the ecological electrical linkage, you can achieve home appliances control, light switch, natural gas energy saving, doorbell doorbell; Health data, you can achieve the elderly guardianship, fall detection, SOS alarm; with the application of shops, you can achieve traffic statistics, retail analysis and shopping guide VIP service; with public safety applications, you can deploy a wide range of electronic face sensitive areas Police; with medical big data, you can substantially prevent medical care, medical trouble and traffickers and other applications scenarios.
Overall, King is the first generation of machine vision chip T01 has a human-oriented high-performance multi-target detection capabilities, and lower operating power consumption, the cost of civilization, good integration, is now in Video camera field first landing.With the AI wave swept, T01 is expected next year will be able to expand the scope of application, there will be more products using T01 landing in more areas.