Background

With rapid urban economic development, video-surveillance systems have become an important technical means for government social management and public services, playing an indispensable role in safe-city construction, traffic management, urban management, and emergency management. Video-data applications have also become more widespread, playing a decisive role in power-grid inspection, petroleum exploration, pipeline transport, and other fields.

As power-sector reform deepens and the power internet is promoted, UHV, flexible power supply, and other new technologies are applied on enterprise networks, and power inspection systems are under increasing pressure.

A new inspection and control system relies on UAVs, intelligent robots, video big-data analysis, and other advanced technologies to raise grid safety and reliability.

Liaoning Power Grid aims to raise inspection capability and build a visual intelligent safety-control platform that integrates intelligent monitoring, intelligent video-data analysis, intelligent O&M, and safety control.

Approach

(1) Build a unified video-surveillance cloud: based on cloud computing, cloud storage, IoT, and other technologies, and within the overall framework of the “visual intelligent safety-control platform,” strengthen top-level design and build a unified video-surveillance cloud to meet future growth of monitoring applications and massive video storage, construction, and maintenance needs.

(2) Build an integrated HD video library realizing video-data management, processing, storage, analysis, sharing, and publishing. Through the video-surveillance cloud, strengthen organic integration and efficient use of departmental video information resources and give full play to the overall benefit of video-surveillance systems.

(3) Unify video-surveillance point planning: plan points by point, line, area, and UAV grid, with coordinated construction, laying a foundation for a gridded video prevention-and-control system.

(4) Unify standards and specifications: formulate HD video-surveillance information-access standards and standardize front-end access.

Platform Architecture

Logically, the platform mainly consists of the following modules:

Front-end devices, access-server cluster, storage-server cluster, processing-server cluster, streaming-media server, center server, and clients.

The role of each module is briefly as follows:

1) Front-end devices

Under scheduling by the center server, front-end devices collect information and transmit it to the storage/processing server cluster via RTSP and SDKs. (Front-end devices include network cameras, UAVs, and others.)

2) Access servers

Access servers mainly work with the center server to complete operations that need to act directly on front-end devices.

3) Center server

a) User management: based on the user information table, manage login and permissions, and add, delete, and manage users.

b) Front-end collection-device management: based on camera and user permissions, manage the status of front-end devices and node running status visible to the current user, with real-time updates.

c) Information interaction with users and front-end devices: update front-end information, schedule front-end devices according to customer needs, and distribute tasks to JobKeeper. Process information data to complete platform-wide scheduling.

4) Storage-server cluster

Provide massive storage for historical-data playback and processing.

5) Processing-server cluster

Provide efficient large-scale data processing such as video transcoding and intelligent recognition.

6) Streaming-media server

Provide standard RTSP streaming-media services. Users obtain processed real-time video via the corresponding RTSP address for monitoring and remote access.

7) Clients

Video matrix, video wall, PC clients, mobile terminals, and others.

Detailed Design

1) Front-end devices

This project mainly includes network cameras and UAV aerial-photography data. Video is provided externally via the standard RTSP streaming protocol, with corresponding SDK function interfaces to control PTZ and other operations.

2) Access servers

Integrate various front-end devices, connecting non-standard devices via protocols or SDKs so that multiple front-end cameras can be accessed uniformly and invoked by other platform modules. Poll front-end devices; if an exception occurs, generate an alarm, parse the SDK, and keep control communication between the platform and front-end devices. In this project, Sugon I840-G30 next-generation high-performance servers provide access services; a single device can provide 500-channel video access.

3) Center server

The core control part of the platform, realizing signaling interaction with clients, scheduling cloud-cluster nodes via the JobKeeper cloud-scheduling system to process tasks, and providing unified management and monitoring of the whole platform.

a) User management: based on the user information table, manage login and permissions, and add, delete, and manage users.

b) Front-end collection-device management: based on camera and user permissions, manage the status of front-end devices and node running status visible to the current user, with real-time updates.

c) Information interaction with users and front-end devices: update front-end information, schedule front-end devices according to customer needs, and distribute tasks to JobKeeper. Process information data to complete platform-wide scheduling.

d) Unified scheduling and management of the server cluster, obtaining each machine’s running status. Automatic scheduling and deployment based on running status, with load balancing to raise machine utilization and thus processing efficiency.

e) Resolve redundant processing states in the server cluster, find and fix errors, and ensure unattended, self-growing efficiency.

In this project, Sugon I840-G30 next-generation high-performance servers provide video-data processing and scheduling.

4) Storage-server cluster

Sugon’s next-generation ParaStor cloud-storage system provides a unified storage resource pool for key data, historical-video playback, and related downloads. It also supports network-mounted drive letters to meet storage needs of other platform modules.

The Sugon ParaStor300S distributed storage system stores video, images, and other critical business data. The file system uses N+M:B so that data are not lost when disks or data nodes fail, while usable capacity can be controlled at about 80% of total disk capacity. After a ParaStor disk fails, rebuilding 1 TB takes only half an hour, requiring a rebuild speed of 580 MB/s. Data controllers must also serve external I/O during rebuild, so 4×1 Gbps on a node cannot meet rebuild speed. ParaStor therefore uses a 10 GbE network as the internal storage network. To avoid inability to read or write if a single data network fails, the data network is also redundant: each data controller is configured with two data networks. This better balances disk reliability and quantity cost, maximizing system stability and ensuring data security.

ParaStor cloud storage is a general-purpose storage platform providing unstructured massive-data storage, serving users as a clustered file system and clustered NAS.

ParaStor cloud storage can provide TB/s-class high-speed bandwidth and EB-scale capacity, meeting applications with extremely high capacity and I/O requirements in aircraft, automotive and ship design, biological genome research, materials science, weather forecasting, earthquake monitoring, environmental monitoring and analysis, energy exploration, e-commerce, online games, social and video-sharing websites, animation rendering, and video editing. It is widely used in education, research, manufacturing, enterprise, healthcare, petroleum, broadcasting, and the internet.

5) Processing-server cluster

Roughly divided into access analysis, data processing, and result distribution, it is mainly responsible for processing tasks on ingested video—such as content recognition, real-time transcoding, and recording storage—then sending results to the streaming-media server and storage servers.

Processing servers use a Sugon I840-G25 high-end four-socket GPU server cluster, each server configured with four high-end graphics accelerators to raise video-processing capability.

Solution Advantages

1. High performance-to-price ratio and scalability

The “cloud video platform” is built on Sugon ParaStor300S cloud storage, with low-cost, highly reliable massive storage and data processing and unlimited scalability. When storage demand grows, users only need to add storage nodes. Hot-swap and non-stop dynamic upgrade are supported, substantially lowering investment and upgrade-maintenance cost.

2. Strong video-integration capability

Facing massive heterogeneous front-end video systems, transparent video-access technology realizes unified integration and management, supporting docking with most mainstream surveillance vendors, standardized protocols such as RTSP and GB/T 28181, and SDK development. HD video data and UAV data are well integrated.

3. Highly reliable cloud-storage technology

Sugon ParaStor300S separates control flow and data flow. Multiple storage servers serve data access concurrently, realizing high-concurrency access. Access from different clients is automatically load-balanced across storage servers. System performance grows linearly with node scale; the larger the system, the more obvious the cloud-storage advantages. There is no performance bottleneck, no single point of failure, and hardware faults are automatically shielded.

4. Dynamic video-access technology

Using “dynamic access,” video data are invoked on demand. The platform ingests and processes a video channel only when a user needs to monitor or store that picture; at other times resources are released automatically. Massive video can be aggregated and managed easily, making the most of existing investment and raising processing performance and network-bandwidth utilization.

5. Intelligent video-content recognition

Sugon high-end four-socket servers with four high-end P40 GPU accelerator cards build an intelligent video-data processing platform. Intelligent image retrieval uses advanced image processing combined with pattern recognition to analyze massive video, providing functions such as region intrusion detection, line-crossing detection, face detection, people counting, leftover-object detection, flame detection, line quality, and line-loss hazard recognition, and can support more linkage applications on that basis. Real-time intelligent analysis runs on the backend server side, independent of front-end cameras and firmware. Any camera can be assigned any recognition type, easily realizing multiple real-time analyses on the same video or batch recognition across many cameras of different types and quantities.