Integrating Video and Intelligence for Safe Cities
Community security disruptions have reached unprecedented levels. As the European refugee crisis escalates and international security worsens, Safe City experts are adopting strategies to quickly predict, detect, and defend against security threats.
Today’s Safe City video surveillance resources are not fully shared, and security capabilities remain insufficient. Separate construction of internal and external networks for public security systems prevents internal networks from sharing data with their outside counterparts.
Cloud and Big Data services used for police investigations in internal public security networks are enhanced by video use but, in order to transfer internal video efficiently to external networks, much of the police cloud data and Big Data must also be transferred. Before such transfers can take place, external public security networks must implement End-to-End (E2E) network, system, data, and application security to avoid the following risks:
- Trojan horses from external websites
- Stealing or leaking sensitive data from application systems
- Identity or permissions misuse by different departments sharing applications
- DDoS attacks targeting important application systems
Conventional video surveillance systems are clearly not enough. The image data that video surveillance systems collect and the valuable metadata derived from this data supply evidence for solving cases and serve as an information resource for all municipal services. To ensure the security of recorded data, video surveillance platforms must run on mature, stable frameworks that provide service-guarantee policies.
Quality, Speed, and Accuracy
In current Big Data applications for Safe Cities, video surveillance systems typically fail to meet service requirements, and computing power for Unstructured-to-Structured (U2S) analysis is generally weak. Specifically, the complexity of monitored areas in Safe Cities causes traditional intelligent analysis algorithms to send false alarms without reporting relevant information. Although groundbreaking technologies such as deep learning have already been applied to such systems, many features remain immature. To extract key information from data, police departments are forced to examine video data manually, which increases operations costs and decreases efficiency.
The critical question in Safe City construction is how best to help police departments quickly search through vast amounts of video data for valuable leads and needed evidence to close cases. The three keys to successful searches are image quality, speed, and accuracy, all of which are determined by the efficiency of intelligent analysis algorithms.
To make intelligence ubiquitous and omniscient, high-definition camera platforms must be mounted to afford the widest possible range of motion to avoid blind spots; transmissions must be immediate, secure, and reliable; and cloud analysis must be accurate and produce fast response times. The capability to create fully connected, perceptive Safe City networks requires the support of high-performance hardware platforms.
Network-wide intelligence equips all terminals on a network with intelligent features. This allows front-end cameras — independently or in combination with platforms — to detect targets or recognize features in people, vehicles, and objects, and analyze complex actions.
The migration towards intelligent networks and the development of built-in intelligence can be divided into three levels:
- Simple intelligence, such as the analysis of complex actions
- Complex video intelligence, such as motion analysis, target detection, and the extraction of key frames or target features
- Deep video intelligence, such as the analysis of target features
The industry is moving towards deep intelligence, but the main obstacle is how to layer and coordinate front-end and back-end intelligence.
The core value in security intelligence lies in its ability to alleviate the pain points of industry users. Network-wide intelligence centered on solving cases and maintaining public security enables police to examine cases and make decisions efficiently due to the following features:
- Video clues allow facial and license plate recognition and tracking of people and vehicles.
- Linearly expandable distributed architecture and dynamically adjustable distributed computing provide fast, concurrent video downloads and high video compression.
- Information can be retrieved in seconds from databases with trillions of records.
As 4K and 8K cameras become commonplace, more efficient encoding and decoding algorithms are needed to lower bandwidth and storage resource consumption and decrease investment. Along with these advances comes the linear scalability of systems and associated requirements:
- Distributed resource management encompassing the internal computing, storage, and network resources of management platforms
- A distributed scheduling framework for platform analysis and storage tasks to increase usage rates of related resources
- Structured data preprocessing to unify processing of metadata output by different algorithms
- An image information database to manage the structured, semi-structured (image features), and relevant unstructured data (images or video clips) output by algorithms
- A distributed image search engine to offer unified, distributed search features for standardized metadata with multiple search types, including standard, fuzzy, and feature-based queries
- U2S analysis interface to provide third-party applications with unified task and configuration management for intelligent analysis, notification management for alarms, search management, and access control for image information
Interconnection and Data Sharing
In most Safe City construction, platforms for intelligent analysis remain separated. The lack of an open standard for southbound and northbound interfaces across intelligent platforms prevents product chains from dividing tasks precisely. This restrains computing resource sharing and reduces security system efficiency.
Intelligent analysis platforms must be sufficiently open; southbound interfaces must support different manufacturers’ algorithms through unified, standardized ports; and northbound interfaces must provide application calls and searching on upper levels through unified SDK or standardized ports.
Open data sharing involves creating a structured video analysis U2S networking platform that is multi-class, multi-area, and uses a layered architecture. U2S platform networking includes:
- Metadata description standards in which different types of intelligent devices and algorithms produce standard descriptions of metadata
- A custom, unified interface standard that decouples algorithms from platforms to create common resource pools
- A U2S networking standard between U2S and U2S, or U2S and third-party applications, including standard interfaces for task management control, alarm notifications, and metadata management
Ubiquitous and Omniscient
The continuing growth of Safe City deployments around the world comes with new requirements for large-scale intelligent service applications. From a systems design standpoint, these requirements include upgrading E2E (camera-to-network-to-platform, also known as cloud-to-pipe-to-terminal) system security, U2S analysis, linear scalability, network interconnection, and data-sharing capabilities.
High-definition video, home security, cloud services, and other advances will become deeply integrated with other new technology concepts. By creating ubiquitous, omniscient networks, we can provide intelligent data mining and sharing for all types of commerce in Safe and Smart Cities. Most importantly, with leading new ICT, we can improve communication to ensure public safety.