Any Questions?: +86-23-67305242
Chat Now
Company News Industry News

Professional Integrated Biometric Solution Supplier.

How to choose the face recognition access control device temperature measurement module and the motherboard program?

Author: huifan   Time: 2021-08-24

Do you only know the Hessian solution, arm9 solution, Hyman camera? What is the principle of their programs do you know? Let's see together
Human body sensor module
When face access control is used in a low light or no light environment, the human body sensor module is usually required to detect whether someone is close, and then real-time start the supplemental light module for supplemental light and face recognition. Complementary technologies are typically the following.
  • 1. Near-infrared human sensor
The human body induction class switch is also called pyroelectric human body induction switch or infrared intelligent switch. Automation control product is an automatic product based on infrared technology. When a person enters the sensing range, the special sensor detects the change of human infrared spectrum and automatically turns on the load, and will continue to turn on if the person does not leave the sensing range; after the person leaves, the delay automatically turns off the load. Such sensors are widely used in intrusion alarm systems for detecting human activities.
  • 2. Pyroelectric infrared sensor
The center wavelength of infrared radiation from the human body is 9~10um, and the wavelength sensitivity of the detection element is almost constant in the range of 0.2~20um. At the top of the sensor opened a window equipped with a filter lens, the filter can pass the light wavelength range of 7 ~ 10um, just suitable for the detection of human infrared radiation, and other wavelengths of infrared light by the filter to be absorbed, so that the formation of a special infrared sensor for the detection of human radiation.
HFSecurity Facial Recognition Access Control Device
Advantages: itself does not send any type of radiation, device power consumption is very small, good concealment. Low price.
Disadvantages: susceptible to various heat sources, light sources interference; passive infrared penetration is poor, the infrared radiation of the human body is easily blocked, not easily received by the probe; ambient temperature and human body temperature close to the detection and sensitivity significantly decreased, sometimes resulting in a short period of failure.
  • 3. Microwave radar
Radar microwave detection of moving objects with high sensitivity, induction range even small movements can be captured by the sensor in time, such as raising or lowering the arm, turn, bend, etc.. Microwave induction can penetrate glass and thin wooden doors and walls of less than 10cm. It is not affected by environment, temperature, dust, etc. Microwave sensing has no directional difference, and motion in any direction can be effectively sensed.
Microwave sensing module has the advantages of strong resistance to RF interference and is not affected by temperature, humidity, light, airflow, dust, etc.
Summary: near infrared, pyroelectric and microwave radar for human detection have advantages and disadvantages, near infrared and pyroelectric high sensitivity and low cost but easy to outdoor light interference, microwave radar small size detection range is far but power consumption proofreading larger. A comprehensive view of microwave radar is a more mature and reliable human body sensing program.
Complementary light module
The current face recognition fill light module usually uses high brightness LED lights to fill light. LED fill light is already a very mature technology, is a solid state semiconductor device that can convert electricity into visible light, it can directly convert electricity into light. The maturity of LED fill light module is relatively high and the difference is relatively small, choose the right power fill light module can be.
HFSecurity Biometric solution
Main control board
The main control board is the core module of the face access control calculation, the usual CPU of the main control board has the following types of calculation.
  • FPGA architecture
  • Dedicated acceleration chip architecture
  • X86 architecture
  • Arm architecture
  • 1.Arm main control
At present, the mainstream master control chip for face access control is based on Arm as the core of various embedded intelligent hardware device chips, the main chip solutions are Rising Micro, Allwinner, MediaTek, Qualcomm and many other manufacturers. Due to the wide variety of manufacturers chip, the main control chip in accordance with the chip based on the ARM architecture of the CPU instruction set to be divided, in order to facilitate the comparison of the performance of different hardware.
A. Ultra-low power, ultra-low performance: embedded core
Using Arm Cortex-M series MCU, ultra-low power consumption, ultra-long standby, suitable for smart lock, smart cat eye and other product solutions. ARM Cortex-M architecture, according to market demand, is divided into M0 to M4 and M7 levels, Cortex-M0 focuses on the primary 8/16-bit applications, Cortex-M3 focuses on the intermediate 16/32-bit applications, and Cortex-M3 focuses on the intermediate 16/32-bit applications. Cortex-M0 focuses on the early 8/16-bit applications, Cortex-M3 focuses on the mid-range 16/32-bit applications, Cortex-M4 focuses on the high-grade 32-bit and DSC (Digital Signal Control) applications, and Cortex-M7 focuses on the flagship high-grade automated production and IoT applications.
The ultra-low-power, ultra-low-performance embedded core, in the field of face recognition, generally supports only 300,000-1 million pixels of face comparison, and the capacity of the face library requires less than 100 people, and the faces are required to be highly compatible. It belongs to the application state of standard ideal face recognition. The face recognition algorithm is usually a traditional model not a deep learning model.
B. Low power consumption, low performance: A7, A9 cores
The ARM Cortex-A7 MPCore processor is the most efficient application processor developed by ARM to date, significantly extending ARM's low-power leadership in future entry-level smartphones, tablets, and other advanced mobile devices.The Cortex-A7 processor is a power-efficient processor based on the ARMv7-A architecture introduced by ARM, Inc. It has been widely used in low-cost, full-featured entry-level smartphones since 2012.
The processor is fully compatible with programs developed for other Cortex-A family processors and draws on the design of the high-performance Cortex-A15 processor with new technologies including virtualization, Large Physical Address Extension (LPAE) NEON Advanced SIMD and AMBA 4 ACE conformance. The balance between performance and power consumption is also taken into account. A single Cortex-A7 processor on a 28nm process is five times more energy efficient than the ARM Cortex-A8 processor on a 65nm process (used in many popular smartphones between 2010-2012), with 50 percent more performance and only one-fifth the size.
The lower-end CPUs of the Arm Cortex-A series, the A7 and A9 series, typically offer tens of M FLOPS of arithmetic power in a single core, and can be used for face algorithms where the deep learning model has been significantly optimized for comparison applications in small-scale face libraries. the A7 and A9 architectures have the largest number of chips and are widely used, but are limited by chip performance issues that often prevent complex algorithm calculations from being implemented, and can be The A7 and A9 architectures are the most numerous and widely used, but are limited by chip performance issues and often cannot achieve complex algorithm computation, and can be used for fully-fitted face matching application scenarios.
Applicable deep learning models such as Resnet 18 and other low-level deep learning models, the applicable face library capacity is usually below 1000 people.
HFSecurity Face Recognition Access Control
C. Medium performance: A12, A17 and A53, A57 cores
Based on the arm Cortex-A series of mid-range CPUs, such as 32-bit A12, A17 or 64-bit A53, A57, a single core can accomplish hundreds of MFLOPS of computing performance, multi-core parallelism can achieve close to G FLOPS of processing power, able to run some medium-scale deep learning models such as Resnet 34, Resnet 50, can be applied to 5000 or even 10,000 face bank with the type of face comparison analysis, etc.
The Cortex-A17 MPCore processor is currently the highest performance processor in the Cortex-A family.The Cortex-A17 microarchitecture is identical to the Cortex-A12, but with improved external interconnects based on the Cortex-A12 architecture and the introduction of a new coherency bus, AMBA4 ACE (originally AMBA4 AXI), which allows for faster connection to memory controllers. allows for faster connection to memory controllers, resulting in improved performance and power efficiency.
Thanks to this new bus, Cortex-A17 can support full memory coherency operation for multi-core SoCs and can participate in big.LITTLE dual-architecture hybrid solutions, such as two Cortex-A17s paired with two Cortex-A7s. Cortex-A17 will shoulder the burden of Cortex-A9 next generation, compared to Cortex -A9, Cortex-A17 performance increase of up to 60%.
Typical chips such as RK3288, MT6595, etc. Typical devices such as Raspberry Pi, Raspberry Pi Broadcom CM2708 ARM11 @1 GHz(OC): 316.56 MFLOPS.
D. High performance: A72, A73 and A53, A57 hybrid architecture cores
A72 and A73 are ARM's high performance processors, which can achieve a good combination of high performance and low power consumption by pairing with the large and small core architectures of A53 and A57. Cortex-A73 uses the full-size ARMv8-A architecture, which can reach up to 2.8 GHz main frequency and can use 10nm and 14/16nm processes, and according to ARM's official introduction, when A73 uses The A73 is the smallest processor with the smallest core in the ARMv8-A architecture, with an area of 0.65mm per core, and continues to support the big.LITTLE architecture.
A72 and A73 processors provide single-core super G FLOPS of floating point processing power, close to the performance of x86 CPUs, which can be used in server-class ARM devices.
Typical A73 chips such as the Rising Micro RK3399 and Qualcomm Baron 835. These CPUs and combined with GPU can run Resnet50, Resnet101 and other deep learning models, which can be used for tens of thousands of face library comparison and high-performance face detection.
HFSecurity Face Recognition Access Control Device
2. Dedicated acceleration chip main control
  • A. Embedded GPU chip
Deep learning chip leader NVIDIA has launched embedded development modules Jetson TK1, TK2, TX1, TK2 and other modules specifically for device computing. Take the latest TX2 for example, with a typical performance of 1.5 TFPLOPS.
TX2 is upgraded from TX1's Tegra X1 to Tegra Parker processor, which is manufactured by 16nm process, 6-core design, CPU part consists of 2 Denver + 4 A57 cores together, while GPU adopts Pascal architecture with 256 CUDA and 1.5 TeraFLOPS floating point performance, compared to the old Tegra X1's GPU performance is about 50% higher.
The advantages of GPU dedicated chips are strong performance and good software compatibility, but the disadvantages are high cost, high heat generation, and too low share in the embedded market.
B. Intel dedicated acceleration chip
Intel's acquisition of the neural network acceleration chip company Movidius, introduced a feature called Neural Compute Engine, a DNN gas pedal integrated on the chip. TOPS), and theoretical throughput of 4+ TOPS.
The Movidius Myriad X VPU vision processing unit, a system-on-chip (SoC) with a dedicated neural network computing engine, can be used to accelerate deep learning inference at the end of products such as drones, robots, smart cameras, virtual reality, etc. Myriad X is an on-chip integrated hardware module designed to run at high speed, low power and without sacrificing accuracy. Deep learning-based neural networks that allow devices to see, understand and respond to their surroundings in real time, delivering 1TFlops (trillions of times per second) of computational performance and overall performance that can exceed 4TFlops. the chip is also very miniature in size, measuring only 8.8 x 8.1mm in length and width, and consumes less than 1W of power and is manufactured using TSMC's 16nm FFC process.
The advantages of Movidius chip are low cost, strong performance and low power consumption, but it has a complex development environment, poor software compatibility and poor productization. At present, some manufacturers use this type of chip to do face access control development.
C. Heisi NPU program
The dominant domestic video imaging chip manufacturer, Heisi, released the current industry's most powerful visual intelligence processing chip 3559A in 2017, which is the world's flagship performance SOC chip. It adopts 12nm ultra-low power process; supports multi-core multi-CPU; supports 32MP 30 fps encoding; has independent DSP and GPU, supports OpenGL and OpenCL, and can do a lot of work that only PC can do now; with dual-core NNIE neural network computing engine, supports deep learning algorithm, and the arithmetic power reaches an amazing 4T (far beyond NVIDIA's TX1). Support multi-sensor input (up to 8) and support for running stitching algorithms; support for Professional 4KP30 raw video output, etc.
As a face access control products, the main advantage of the Heisi chip is strong performance, good stability, but there are also too high cost, custom development difficulties, low software compatibility and other limitations.
3. X86 CPU main control
X86 main control refers to the use of x86 instruction-based CPU development of face recognition access control equipment, x86 CPU advantage is good software compatibility, the disadvantage is the high cost, equipment volume, product stability is general, so now face access control equipment less and less using x86 architecture.
Applicable to face access control x86 processor is usually: low-power Celeron, ATOM series processors, low-power notebook CPU.
Celeron processor, ATOM processor and low-power notebook CPU is the CPU leader INTEL in order to enter the embedded and mobile computing market developed by the low-power high-performance processor. Celeron processors are Intel's "budget" products, launched in 1998. Intel Atom (official Chinese translation: Intel Atom processor, development code Silverthorne) is a family of ultra-low voltage processors from Intel. The market positioning of the processor lies in smartphones, tablet PCs and low-cost PCs.
Later, as INTEL abandoned the embedded and mobile CPU processor chip field, these two processor product lines are basically no longer updated. Part of the face access control products in order to software compatibility good use of such processor solutions.
Low-power Celeron, ATOM series processors have a floating point capability of about 0.1-0.5TFLOPS.
Low-power notebook series CPU processor floating point performance in the 0.2-0.6TFLOPS or so.
HFSecurity Biometric Solution
4.FPGA master control
The use of FPGA large-size programmable logic for face recognition operation processing is a relatively early technical solution, due to the large size of FPGA products, development difficulties and other reasons, the current in the face access control integration equipment less use this solution.
FPGA (Field-Programmable Gate Array), that is, field-programmable gate array, it is in PAL, GAL, CPLD and other programmable devices based on the further development of the product. It emerged as a semi-custom circuit in the field of Application Specific Integrated Circuit (ASIC), which solves the shortage of custom circuits and overcomes the shortcoming of the limited number of gates of the original programmable devices.
FPGA development is very different from traditional PC and microcontroller development. FPGA is mainly parallel operation and implemented in hardware description language; it is very different from the sequential operation of PC or microcontroller (either von Neumann structure or Harvard structure), which also makes it difficult to start FPGA development. debugging and other aspects.
The main control board is the computing core of the face access control machine, is currently the ARM architecture of the main control board occupies the mainstream product market. According to the computing performance of the main control board, the face access control equipment can be divided into.
Consumer-grade face access control equipment: face bank of 1000 people or less, suitable for a single room or small and medium-sized enterprises face access control management. Suitable CPU involves: ARM's A7, A9, etc.
Enterprise-level face access control device: less than 10,000 people in the face database, suitable for face access control management of a single building or enterprise. Suitable CPUs involved: ARM's A12, A17, A53, A57 and Intel's saiyan or atom series chips, etc.
Industry-level face access control devices: face database capacity of 50,000 people or less, suitable for face comparison management of large parks. Suitable CPUs involved: arm's A72, A73 and special chip acceleration hardware such as Movidius, Hesse NPU, etc.

Fingerprint Identification Device Manufacturer


Related News:
Huifan technology facebook Huifan Technology Likedin Huifan Technology Youtube Huifan Technology Twitter
Tag: Access Control; Face Recognition Solution; Palme Face Temperature Access Control; HFSecurity; EU Green Pass; Health QR Code Scanner