Edge computing algorithm solution based on RK3588+AI: smart park, smart community, smart logistics

 Introduction to RK3588 AI Edge Computing Motherboard Specifications

About this document

This document introduces in detail the shape, size, and technical specifications of the AI ​​edge computing motherboard based on the Rockchip RK3588 chip, as well as detailed hardware interface design reference instructions, so that customers can quickly apply the RK3588 edge computing motherboard to industrial Internet, smart cities, smart security, Smart terminal equipment in the field of artificial intelligence such as smart transportation and smart medical care.

1. 1. product description

This product is an AI edge computing motherboard based on Rockchip RK3588 architecture. RK3588 is a new generation of domestic flagship high-performance 64-bit octa-core processor. It adopts 8nm technology and has high computing power, low power consumption, super multimedia, and rich data interfaces. Features. Equipped with quad-core A76+ quad-core A55 octa-core CPU and ARM  G610MP4  GPU , built-in NPU with 6  TOPs computing power . Support 8K@60fps video decoding , support 8K30fps encoder, the peripheral interface is very rich, with super scalability. RK3588 built-in embedded neural network processor

(NPU), the NPU computing power can reach 6.0TOPs, support PCI-e/USB3.0/RGMIIMP, and can perform video structural recognition and analysis on 32-channel 1080P network cameras.

The edge computing motherboard of the Rockchip RK3588 chip has powerful computing power, rich input and output interfaces and strong scalability. The PCB is designed with 10-layer immersion gold technology, which has excellent electrical characteristics and anti-interference characteristics, stable and reliable operation, and can reach industrial-grade standards. It can be widely used in artificial intelligence terminal fields such as smart cities, smart security, smart medical care, and industrial Internet.

Chip Block Diagram  

1. 1. platform features
      1. Rockchip RK3588 adopts 8nm LP technology, based on quad-core Cortex-A76, quad-core Cortex-A55 large and small core CPU structure 64bit ultra-high-performance processor, the main frequency is up to 2.4GHz, and the powerful RK3588 can be used in various AI application scenarios Bringing extremely excellent performance, compared with RK3399, the CPU performance has been improved by 3 times;
      2. The built-in NPU computing power can reach up to 6.0 TOPs, three-core structure, flexible distribution of computing power. High computing power NPU can support INT4/INT8/INT16/FP16 mixed computing, and its strong compatibility can easily convert network models based on a series of frameworks such as TensorFlow  / MXNet / PyTorch / Caffe  ;
      3. Support 32 channels of 1080P decoding, built-in a variety of powerful embedded hardware engines, support 8K @ 60fps H.265 and VP9 decoders, 8K @ 30fps  H.264 decoders and 4K @ 60fps AV1 decoders; support 8K30fps H.264 and H.265 encoders, high-quality    JPEG encoder / decoder, dedicated image pre-processor and post-processor, compared with RK3399 , GPU performance has increased by 6 times;
      4. Built-in 3D  GPU , compatible with OpenGL  ES1.1/2.0/3.2 , OpenCL  2.2 and Vulkan1.2 . Special with MMU

The 2D hardware engine will maximize display performance and provide a smooth operating experience;

1. 1. 1. Built-in Rockchip self-developed 48M pixel ISP ( image signal processor ) , supports the realization of many algorithm accelerators, such as HDR ,

3A , LSC , 3DNR , 2DNR , sharpening, defogging, fisheye correction, gamma correction, etc.;

1. 1. 1. Support up to 32G super large running memory, faster response, more real-time, with high-performance 4- channel external memory interface

( LPDDR4/LPDDR4X/LPDDR5 ), able to support demanding memory bandwidth;

1. 1. 1. It has HDMI  2.1/MIPI-DSI/DP1.4/VGA multiple video output and HDMI   RX2.0/MIPI-CSI video input interface , supports multiple 8K video output and 4K video input, and can realize up to four screens with different displays;
      2. 4- way standard SATA3.0 interface, which can expand multiple 2.5/3.5 -inch SSD/HDD hard disks at the same time; onboard high-speed M.2  SA- TA3.0  interface, supports M.2  2242 high-speed solid-state hard disk; allows equipment to be easily expanded Large capacity up to TB level;
      3. Onboard standard PCIe3.0 ( 4  Lane ) interface, the data transmission rate of each Lane reaches 8Gbps , and the transmission rate of 4 Lanes can reach up to 32Gbps , high-speed and stable transmission, which can be used to expand standard PCIe3.0 devices; 
      4. With RS485 , RS232 , I2S , I2C , UART , CAN , SPDIF , MIPI  CSI , MIPI  DSI , USB3.0 , USB2.0 , SPI , GPIO and other expansion interfaces, it can cover edge computing, smart medical care, industrial control and other smart terminals interface requirements ;
      5. The operating temperature range can reach -20°C ~ +70°C, and there is an optional version of the car-grade RK3588 chip from -40°C ~ +85°C;

RK3588 AI BOX Appearance  

1. 1. motherboard picture

1. 1. Motherboard Block Diagram

Motherboard size: 146*102mm

1. 1. motherboard interface

RK3588 edge computing motherboard interface description
Function	interface description
Internet function	Support single frequency (2.4GHz)/dual frequency (2.4GHz/5GHz) WIFI, AMPAK, Bilian and other brands are optional
	Support Quectel and other mainstream 5G/4G modules
	Support RJ45 dual Gigabit Ethernet ports
motherboard interface	Support eDP, MIPI, DP display interface, support touch screen TP interface;
	Support binocular MIPI-CSI camera interface, support MIPI DSI camera input
	Support one Headphone, one MIC, one Speaker and one LINE-IN interface
	USB3.0*1，USB2.0*1，Type-C*1
	TF card*1
	DI*2, DO (Relay relay)*2
	RS485*1，RS232*1
	CAN*1
	HDMI IN*1，HDMI OUT*2
	5G/4G MINI-PCIE interface*1
	5G/4G SIM card*1
	SATA port*1
	M.2 port*1
	RTC battery*1
	POWER button*1, RESET button*1, RECOVERY button*1
	LED Status Indicator*4
	Debug Debug serial port*1
	DC IN*2 (DC socket + Phoenix terminal, 12V~65V)

1. 1. Application field
      1. edge computing
      2. digital cockpit
      3. Smart Security
      4. Smart transportation
      5. Smart medical

and other edge computing and artificial intelligence product terminals.

1. 1. Design and Production Advice

1. 1. 1. reference circuit

Since the reference design is updated irregularly, please obtain relevant reference design materials through formal channels.

1. 1. 1. Backplane layout design

It is necessary to carry out corresponding anti-ESD design for each peripheral interface;

1. 1. 1. Thermal treatment

Since the multi-core high-performance processor generates a lot of heat when it is running at full load, please consider system-level heat dissipation when actually using the product.

1. 1. 1. Transportation and Production

Since the modules are matched with different bottom plate characteristics, the specific furnace temperature setting during the production of the paste varies from product to product, please evaluate in detail during production;

Since the motherboard may have experienced different time and environment during the process from the production of the motherboard to the assembly of the finished product, it is necessary to do a good job of dehumidification and drying of the motherboard before attaching the parts;

Since the main board is exposed in all directions and is an electrostatic sensitive device, anti-ESD measures should be in place during the process of picking and moving;

Smart park: create a safe and efficient park environment, and upgrade intelligent operation management

The central management system of industrial parks and other parks is equipped with surveillance cameras, and the data is uploaded to the cloud for subsequent adjustment.

However, a large amount of data accumulates to form a "data island", which increases the workload of personnel.

Through the AI ​​algorithm, a central "brain" can be given to the park to conduct real-time intelligent analysis of massive video data,

Quickly detect various security risks and management problems in the park, and endow the park with the ability to "prevent beforehand".

 

 Smart community: Collaborate with security inspection, fire protection, and social environment to ensure the quality of life in the community

Cooperate with community cameras for round-the-clock monitoring, which can produce such things as garbage overflow, high-altitude parabolic, high-rise outside

Recognition of cracks on walls and facades, recognition of illegal parking of vehicles, recognition of electric vehicles entering elevators, motion recognition, pet recognition

Don't wait for a series of algorithms that meet the needs of fragmented scenarios.

Empower the refined and intelligent management needs of contemporary communities, and cooperate with security inspection, fire protection, social environment, etc.

Column management module to ensure the quality of life in the community.

 Let the light poles have AI algorithms to protect the life and operation of the community around the clock

The AI ​​algorithm empowers the integrated smart light pole solution to identify and detect non-motor vehicles parked indiscriminately and indiscriminately

Pile of materials, exposed garbage, debris left on the road, water on the road, illegal small advertisements, damaged manhole covers, full of garbage

Overflow, damaged road traffic signs, damaged roadways and sidewalks, danger of tree branches falling off, falling lampposts, etc. 20+

debris scene ,

Algorithm deployment was completed within 2 weeks , and the detection accuracy rate was >93% .

 Smart logistics: All-round inspection of package quality and environment, reducing labor management costs

If there is a lack of reasonable planning in the logistics and warehousing management, the phenomenon of warehouse explosion will occur at the peak node, and it will be carried out manually.

The cost of conducting security inspections is extremely high and the efficiency is low.

Using the AI ​​algorithm, it can conduct all-round detection of storage quality, package movement and warehouse environment, and improve the quality of the warehouse.

The degree of refined management of storage and logistics can reduce storage management costs, improve turnover efficiency, and improve service products.

quality.

Refined management of logistics warehouses to fully guarantee the safety of goods

For the valuable financial collateral products (such as steel coils) stored in the warehouse, the AI ​​method is used to refine

Quasi-cross vehicle identification, meeting the high requirements for cargo safety management.

The delivery cycle of customized algorithms is as low as 1 day, the utilization rate of edge hardware is increased by 4 times, and the TCO of procurement is lower than that of competitors.

50% lower , automatic optimization of recognition accuracy >95% .

Smart manufacturing: Improve industrial testing capabilities to ensure production safety and quality

In the production and manufacturing scene, there are irregular operation behaviors of employees, and hidden dangers in production equipment and the environment are difficult to detect

Feeling and other problems.

Through AI algorithm, it can analyze employee behavior data and provide predictive management and maintenance equipment

The safety of employee interaction, timely discovery of potential hidden dangers in equipment and the environment before accidents, and the impact on product quality

Provide intelligent supervision in dimensions such as quantity detection.

 

 

Meet a variety of testing needs and reduce the cost of manual quality inspection

 

Facing the common AI inspection needs of factories, including: short circuit, open circuit, void, depression, protrusion, scratch...

Based on the automated AI training platform, it helps it meet the requirements of various types of AI algorithms for denim defects, hand defects, etc. within two weeks

Requirements for the development and deployment of various types of AI algorithms such as machine film defect inspection and lunch box dirt inspection , and to achieve product detection rates

98.6% , the overkill rate is within 5% , the processing speed is 0.2-0.3 seconds, and the labor cost of customer quality inspection is reduced by 20%.