Trends at the intersection of video and AI

Trends at the intersection of video and AI 

Installation and use of CCTV Cameras for security & surveillance is a no-brainer. Cameras are considered a fundamental commodity for setting up any surveillance infrastructure, but at the same time, 24×7 monitoring of hundreds or thousands of video feeds by operators doesn’t serve the purpose of providing proactive surveillance and quick response to breaches.

2022, where most security cameras deliver HD performance, with more and more models offering 4K resolution with 8K on the horizon. Advanced processing techniques, with and without the use of infrared illuminators, also provide the ability to capture usable images in total darkness; and mobile devices such as drones, dash cams, body cams, and even cell phones have further expanded the boundaries for video surveillance. Additionally, new cameras feature on-board processing and memory to deliver heightened levels of intelligence at the edge.

But video has evolved beyond the capabilities of advanced imaging and performance to include another level:  Artificial Intelligence. Video imaging technology combines with AI, delivers a wealth of new data, not just for traditional physical security applications, but for a much deeper analysis of past, present, and even future events across the enterprise.

This is more than a big development for the physical security industry; it is a monumental paradigm shift that is changing how security system models are envisioned, designed, and deployed. Much of the heightened demand for advanced video analytics is being driven by six prevalent industry trends:

 

1) Purpose-built performance

Several video analytics technologies have become somewhat commoditised “intelligent” solutions over the past few years, including basic motion and object detection that can be found embedded in even the most inexpensive video cameras. New, more powerful, and intelligent video analytics solutions deliver much higher levels of video understanding.

This is accomplished using purpose-built deep learning, employing advanced algorithms and training input capable of extracting the relevant data and information of specific events of interest defined by the user. This capability powers the automation of two important workflows: the real-time monitoring of hundreds or thousands of live cameras, and the lightning-fast post-event search of recorded video. Vintra video analytics, for example, accomplishes this with proprietary analytics technology that defines multi-class algorithms for specific subject detection, classification, tracking, and re-identification and correlation of subjects and events captured in fixed or mobile video from live or recorded sources.

 

2) Increased security with personal privacy protections

The demand for increased security and personal privacy are almost contradictory given the need to accurately identify threatening and/or known individuals, whether due to criminal activity or the need to locate missing persons. But there is still societal pushback on the use of facial recognition technology to accomplish such tasks, largely surrounding the gathering and storage of Personally Identifiable Information (PII).

The good news is that this can be effectively accomplished with great accuracy without facial recognition, using advanced video analytics that analyse an individual’s whole-body signature based on various visual characteristics rather than a face. This innovative approach provides a fast and highly effective means of locating and identifying individuals without impeding the personal privacy of any individuals captured on live or recorded video.

 

3) Creation and utilisation of computer vision

There are a lot of terminologies used to describe AI-driven video analytics, including machine learning (ML) and deep learning (DL). Machine learning employs algorithms to transform data into mathematical models that a computer can interpret and learn from, and then use to decide or predict. Add the deep learning component, and you effectively expand the machine learning model using artificial neural networks which teach a computer to learn by example.

The combination of layering machine learning and deep learning produces what is now defined as computer vision (CV). A subset but more evolved form of machine learning, computer vision is where the work happens with advanced video analytics. It trains computers to interpret and categorize events much the way humans do to derive meaningful insights such as identifying individuals, objects, and behaviours.

 

4) Increased operational efficiencies

Surveillance systems with a dozen or more cameras are manpower-intensive by nature, requiring continuous live or recorded monitoring to detect and investigate potentially harmful or dangerous situations. Intelligent video analytics, which provides real-time detection, analysis, and notification of events to proactively identify abnormalities and potential threats, transform traditional surveillance systems from reactive to proactive sources of actionable intelligence. In addition to helping better protect people, property, and assets, advanced video analytics can increase productivity and proficiency while reducing overhead.

With AI-powered video analytics, security and surveillance are powered by 24/7 technology that doesn’t require sleep, taking breaks, or calling in sick. This allows security operations to redeploy human capital where it is most needed such as alarm response or crime deterrence. It also allows security professionals to quickly and easily scale operations in new and growing environments.

 

5) A return on security investment

The advent of advanced video analytics is slowly but surely also transforming physical security systems from necessary operational expenses into potential sources of revenue with tangible ROI, or as it is better known in the industry, ROSI – Return on Security Investment. New video analytics provide vast amounts of data for business intelligence across the enterprise. Advanced solutions can do this with extreme cost-efficiency by leveraging an organization’s existing investment in video surveillance systems technology.

This easy migration path and a high degree of cost-efficiency are amplified by the ability to selectively apply purpose-built video analytics at specific camera locations for specific applications. Such enterprise-grade software solutions make existing fixed or mobile video security cameras smarter, vastly improving how organizations and governments can automatically detect, monitor, search for and predict events of interest that may impact physical security, health safety, and business operations. For example, slip-and-fall analysis can be used to identify persons down or prevent future incidents, while building/area occupancy data can be used to limit crowds or comply with occupancy and distancing guidelines. In this way, the data gathered is a valuable asset that can deliver cost and safety efficiencies that manual processes cannot.

 

6) Endless applications

The business intelligence applications for advanced video analytics platforms are virtually endless including production and manufacturing, logistics, workforce management, retail merchandising and employee deployment, and more. This also includes mobile applications utilising dashboard and body-worn cameras, drones, and other forms of robotics for agricultural, oil and gas, transportation, and numerous other outdoor and/or remote applications.
An added benefit is the ability to accommodate live video feeds from smartphones and common web browsers, further extending the application versatility of advanced video analytics.

Indian Army to be equipped with AI-based threat assessment software linked to national databases

The Indian Army will soon get help from artificial intelligence to analyse threats during operations, especially in sensitive areas like Jammu and Kashmir. The ministry of defence is planning to equip the army with an AI-based threat assessment software that can be linked to national databases such as UIDAI and MoRTH.

According to an official communication, this software will be used to “track vehicular movement, monitor online presence of individuals, including on social media platforms, and derive linkages and patterns”.

The software will help the army in tracking and predicting incidents and the reason behind acquiring such a software is that operations require a detailed threat assessment. As of now, details of threat assessment are not available with either security forces or law enforcing agencies but only restricted information is available in the form of registers, both at company operating base (COB) levels and with other headquarters.

“Since the data is either non-existent or is recorded in legacy formats (documents and registers), no methods are available with SF for historic correlation of incidents due to which incidents cannot be tracked or predicted,” the official communication states.

The solution: “an AI-based networked solution that will enable storage and analysis of data to track and predict incidents. Details can be linked with national databases such as UIDAI, MoRTH in order to track vehicular movement, as also monitor online presence of individuals including on social media platforms and derive linkages and patterns”.

The software will have the ability to import data available with other agencies. It will enable central collation of details from various patrols with real-time data visibility at all outstations. According to the MoD, this software will be able to integrate legacy unstructured data available in different formats and provide actionable intelligence by combining different databases through the use of AI and data analytics techniques.

The software application is designed to integrate AI and machine learning (ML) aspects with existing known datasets, thereby giving commanders and staff a comprehensive and common picture with regard to operations intelligence, including the capability to carry out data processing in terms of staff checks, query management, use of business intelligence and visualisation to arrive at a course of actions and also assist in ruling out less viable options.

“As the system is being planned keeping in mind the automation requirements of the Indian Army for intelligence generation, the system will be designed for exploitation by AI and ML aspects,” the MoD said. The software will facilitate display and exchange of information over encrypted and secured internet connection and facilitate day-to-day functioning of operations and intelligence gathering and collating at various headquarters, and also facilitate commanders in decision making by incorporating appropriate inputs.

Adequate security measures will be ensured as per the latest guidelines or policies of army cyber group (ACG) system. It will have NLP (natural language processing) models in English, Hindi, Pashto, Arabic, and Urdu to automatically extract intelligence from generated reports, inputs, open source documents and memos.

It will be capable of conducting video and image analytics for raw camera footage from CCTV cameras, social media videos, images among others. The whole idea is to generate actionable intelligence and ability to do dark and deep web monitoring and analysis.