Introduction to Integrated Video AI Analytics System

Based on computer vision technology, AI empowers various industries. Relying on AI visual analytics technology and robust "edge-to-cloud" computing power, it performs real-time analysis of events such as smoke, fire, and intrusion. In conjunction with a cloud-based early warning platform, it achieves a closed-loop process for event detection, early warning, and handling.

1. Design Architecture
  1. System Architecture

The video intelligent recognition system is divided into four layers from bottom to top: "Perception Layer, Network Layer, Support Layer, and Application Layer." The system's logical architecture is shown in the figure below:

Perception Layer

Connects to front-end sensing devices, such as video surveillance cameras, NVRs, and other IoT sensing devices. It performs real-time monitoring and analysis of critical channels and locations, providing the data foundation for scene analysis. The types of data collected include image streams and video streams.

Network Layer

The network layer connects to the main local area networks (LANs) and dedicated video networks within the factory area.

Support Layer

The support layer provides the main capabilities for the application layer, including video surveillance, intelligent algorithm repositories, other IoT sensing data, and the establishment of alarm models. It provides capability support for future, more intelligent business applications and management needs.

Application Layer

The application layer primarily handles integrated video intelligent early warning business applications, including but not limited to video surveillance, device management, event centers, and various safety and fire hazard applications.

1. 1. Deployment Architecture

Data collection is performed using both new and existing cameras. Video streams are accessed via edge and cloud platforms, and AI intelligent analysis is performed on events. The system also supports transmitting results to third-party platforms for data display. The network architecture diagram of the system is shown below:

System Components:

Existing Surveillance Cameras

Existing surveillance cameras refer to fully utilizing existing video surveillance and video storage infrastructure as data collection endpoints for video AI analysis.

New Smart Cameras

New smart cameras refer to directly deploying surveillance cameras with AI analysis capabilities for newly added or supplemented surveillance blind spots, to actively identify and detect production safety incidents.

AI Intelligent Analysis Server

The AI intelligent analysis server possesses two fundamental capabilities: it detects, analyzes, and recognizes various behaviors in real-time video from existing surveillance cameras, and it also receives and stores alarm events identified by new smart cameras. The intelligent analysis AI all-in-one machine highly integrates video preview, AI algorithm models, and the comprehensive video intelligent AI analysis business system.

The integrated video intelligent AI analysis business system primarily analyzes video data using intelligent analysis algorithms via common protocols such as RTSP, comprehensively displays the analysis results, and provides other business functions. It also supports interfacing alarm data with third-party business systems.

=================RK3588 + Ascend AI | 40 TOPS Computing Box================

● Equipped with quad-core A76 + quad-core A55, CPU clock speed up to 2.4GHz, expandable with 2 Ascend AI acceleration modules, with computing power up to 40 TOPS.

● 2-channel 4K UI, capable of smoothly running complex graphics processing.

● Supports development tools such as MindStudio, MindEdge, and CANN computing architecture.

● Supports Docker, Ubuntu 22.04.

● Expandable with wireless communication modules like 4G/WIFI6/BT.

● Expandable with mSATA SSD.

● Operating Temperature: -20℃ ~ 70℃.

This wide-temperature edge computing box, with built-in 40 TOPS computing power, is capable of decoding up to 80 channels of 1080p video and supporting 8K@30fps HDMI output. It features strong environmental adaptability, ultra-high computing performance, cloud-edge collaboration, large-capacity storage, flexible configuration, wide temperature range support, and easy maintenance and management. It can be widely deployed in edge environments.

Item

Type

Model Parameters

Description

Core

Configu-

ration

Processor

CPU

Quad-core ARM Cortex-A76@2.4GHz

Quad-core ARM Cortex-A55@1.8GHz

GPU

Quad-core ARM Mali-G610 MP4

NPU

6 TOPS@INT8

Memory

LPDDR4X

Default configuration 16 GB

Flash Storage

eMMC

Default configuration 64 GB

Codec Performance

Video Decoding

16 x 1080P@30fps H.264 AVC/MVC

32 x 1080P@30fps H.265/VP9/AVS2

Video Encoding

16 x 1080P@30fps H.264/H.265

Image Codec

Supports high-quality JPEG encoding

Hardware Features

Ethernet Port

(Ethernet)

Ethernet port, supports 10/100/1000M network access

Switch network port, supports 10/100/1000M network access

SFP

For Gigabit optical-electrical signal conversion and data transmission

RS232

Can connect to alarm input devices such as smoke detectors, infrared detectors, access control systems, or alarm output devices such as sirens (Note: RS232 cable length is recommended not to exceed 10m)

RS485

Can connect to alarm input devices such as smoke detectors, infrared detectors, access control systems, or alarm output devices such as sirens

HDMI OUT

HDMI 2.1 interface, can output 4K@120fps/8K@30fps video source to display terminals, compatible with audio output

LINE IN/OUT

1 audio input interface, 1 audio output interface, 3.5mm stereo jack

USB3.0

Can connect devices such as USB flash drives, USB mice, USB keyboards

CAN

Standard CAN 2.0 interface

TF Card Slot

Can insert TF card to expand storage space

SIM Card Slot

For wireless network connection

ADB

For device image flashing (Type-C)

DBG

For wired device debugging (Type-C)

Power Interface

DC 12V

Expandable

Features

Reserved two AI module expansion interfaces

（Dual module specification）

NPU

40 TOPS@INT8

Video Decoding

80 x 1080P@30fps H.264/H.265 video decoding

Video Encoding

40 x 1080P@30fps H.264/H.265 video encoding

WiFi+BT

WIFI6, dual-band 2.4GHz and 5GHz;

Bluetooth BT5.4;

Requires a case with expandable external antenna and antenna.

4G/5G

Expandable 4G/5G wireless communication module, requires antenna

mSATA

Expandable mSATA large-capacity SSD

SATA

Supports optional 1 x 3.5-inch SATA hard drive

SIM Card Slot

Used with 4G/5G wireless communication module

Linux

File System

ext4

Ubuntu22.04 LTS

Software

Develop-

ment

Media Processing

Gstreamer

Compatible with rkmpp plugin, including interfaces for video decoding, video encoding, image encoding, image decoding, display, RGA, etc.

Computing Architecture

AI Frameworks

Supports deep learning frameworks such as TensorFlow, Caffe, Pytorch, Tflite, ONNX, MindSpore

CANN

By providing AscendCL programming interfaces (supporting Python and C++ languages), it enables users to quickly build AI applications and services based on Ascend AI processors.

Basic

Modules

Network Settings

Command Execution

Supports static and DHCP network parameter settings

Running Status

CPU, Memory, Disk

Device Information

Hardware operating frequency, device junction temperature

Log Management

Running status, operation errors, etc.

Time

NTP, manual time synchronization

Upgrade Management

Flashing Upgrade

Supports TF card upgrade