Genesis of Video Analytics
Since the introduction of video over an Internet Protocol (IP) network was first established in 1997, CCTV has, up until recently, been fundamentally known as a ‘dumb system’, providing just live, pre-recorded video playback and export capabilities with very little or no ‘smarts’ behind it. More recently, however, some CCTV systems have been described as IP video surveillance systems due to the fact that such systems now incorporate the use of computer-based intelligence to monitor video streams using analytics. This evolution in technology has changed the nature of CCTV, providing capabilities to end-users far above and beyond what was available only a few short years ago. With such rapid development occurring in only the last few years, one might reasonably ask, what next?
Growth of Video Analytics
According to a recent report from Transparency Market Research, the IP video solutions industry is now one of the fastest growing areas of security, with an annual compound growth rate (ACGR) of 19.1 percent from 2013 to 2019, with new technologies opening up many opportunities for IP camera manufacturers, software companies, distributors and value-added resellers.
Due to the increased growth rate and market expectations, there is a surge of manufacturers etching their way into the market to secure their own corner. This increase in market competition is spurring companies to action, providing new and unique ways of managing a customer’s expectations through smart technologies.
This is made evident by recent technology changes that have provided the industry with higher definition images, improved compression methods, higher frame rates and better low-light performance combined with improved flexibility as a result of each device being IP addressable. There are now almost infinite opportunities with what can be done with video captured via CCTV, essentially turning each image into measurable and sometimes critical data for a customer. The surface has only just been scratched with regard to what information and data can be gathered by the use of video-based analysis and it is expected that in the near future, multiple systems will be able to communicate and make decisions between devices.
That said, such capabilities are still some way off. So, how can video analytics be more effectively applied on IP video today? What problems might this technology solve for a customer and what could it deliver in the future?
To gain insight into these big questions, consider why video analytics has become a growing trend. CCTV has traditionally been about surveillance; using humans to visually process video. Analytics were first introduced in the CCTV field as a means of augmenting a person’s visual processing with pockets of additional digital information. This came in the form of basic motion detection and human tracking. As it became possible to increase the number of transistors on micro-processors (see Moore’s Law), more leeway was given to the amount of CPU cycles the analytic algorithms could consume. Basically, as computers have increased in processing power, they have been able to execute more challenging tasks more accurately and more quickly. As a result, video analytics is no longer supplementing human observation but is fully automating some aspects of surveillance, giving rise to a plethora of new surveillance possibilities.
The most obvious and widely used analytic that is now fully automating parts of surveillance is facial recognition. There is an increasing demand for more efficient and effective security capabilities as intelligence agencies are overwhelmed trying to provide the resources to manage surveillance effectively. Recently, some companies have provided analytics capable of biometric detection of faces with up to 99 percent accuracy and many of these software solutions have direct integration into some of the major video management system (VMS) platforms. This allows authorities to be alerted with video verification in real-time with push-notifications when a suspect is detected. According to John McGiffen, managing director of Intelliscape, there is also the ability to tap into the internet and social media platforms to link people together with a who-knows-who engine, providing critical data to authorities. Facial recognition is no longer an ‘after the fact’ analytic; if deployed correctly, it could play a significant role in preventing terrorist attacks or other such catastrophes before offenders have the opportunity to strike.
That said, highly accurate facial recognition is just one of many forms of video analytics that have been on the rise over the past few years. Some of the more recent video-based analytics that are being used successfully today include licence plate recognition, object left and object removed from an area, line crossing, direction of travel, people and vehicle counting, intruder detection and forensic-based searching. Audio analytics have also recently come into play, with volume detection and even scream detection built into some manufacturer’s cameras.
Video Analytics and Retail
The future of analytics in CCTV is very bright. One area currently experiencing significant growth is the retail and marketing sector. The use of retail analytics such as heat mapping, people counting and dwell times are assisting retailers with demographic information and usable business data that has never been accessed before using camera technology. Marketing departments are benefiting from this information, allowing them to better understand trends and allowing them to significantly improve business processes. The ability to create specific aggregate reports on various entrance ways to determine traffic flows and strategically place vendors in these locations is all made possible through video analytic technologies. Queue management and queue abandonment is another analytic that is becoming more prevalent, with businesses utilising this technology to further improve customer wait times while helping to provide the correct staffing resources at the right times.
Caveats in Video Surveillance
While there has been significant year-on-year growth in the video analytics market, there are still many areas where analytic technology and software can be enhanced.
It is important to understand the pros and cons of where the technology can be applied. For example, alert-style analytics, which are usually contrast based, can still be adversely affected in some instances by scene changes and changes in light. Many manufactures are constantly improving their technologies, but there is a very specific skill set required to set up analytics effectively to reduce the amount of nuisance alarms. However, with this in mind, the solution needs to be built around conditions such as good lighting in these areas to allow analytics to work to their full potential, although there is an abundance of low-light camera technology coming into the market. As this technology improves, the effects of lighting and scene changes will become redundant, with some manufacturers already able to produce colour images in light as low as 0.0045 lux.
In today’s society, people generate data around everything they do; from using credit cards to surfing the internet, interacting and sharing with each other via social media and so on. People’s day-to-day interactions with security systems are no exception. This abundance of data is often referred to as Big Data. Analytics provide the ability to drill down into and analyse sets of related data in a simplified manner. One can only begin to guess at the kinds of information that might be uncovered when it becomes possible to link and mine data relating to potential security issues from across multiple systems.
Video analytics is not just providing bigger and better ways of extracting information from video feed, it has also become a catalyst for the way in which providers think about IP video solutions. The increasing sophistication of systems and the minimisation of costs from end-users are naturally moving analytics toward the edge, as opposed to a centralised approach. What this means in simple terms is that some of the intelligence in the system has moved out of the VMS to reside within the cameras themselves, relieving some of the pressure on the VMS and allowing it to do more, at a faster pace. Edge-based and server-based analytics over an IP network are now becoming a necessity and a growing number of consultants are specifying these solutions as end-users begin demanding more than a simple ‘video feed into a box’. Edge-based analytics have a distinct advantage in terms of intelligence and metadata. As the hardware capabilities of cameras and encoders increases, so too does the sophistication of the analytics programs they run and the quality of metadata produced. This means metadata can be accessed directly, and quickly, as opposed to having a server transcode information. This has the added benefit of dramatically reducing cost as there is no need for third party software and additional hardware, as many camera manufacturers have this functionality built directly into the camera.
Like edge-based analytics, many cameras now utilise edge-based memory storage where the memory is local to the camera. Although this has been around for several years in cameras, it has changed from SDHC, which is a high capacity format with 32GB limitation, to the latest SDXC format which supports up to 2TB on board the camera. Although the SD cards are not quite there yet, some manufacturers can handle up to 256GB at the edge, which will eventually allow for storage of much more than just video.
The Future of Analytics
So what does the future hold and how might video-based analytics evolve over the next five years? There are countless opportunities and problems waiting to be solved by video analytics. As the computing power of IP cameras continues to improve, video analytics will continue to solve more complex problems. Benefits will continue to be seen across multiple sectors, from aiding intelligence agencies to local retail outlets. There are some non-trivial hurdles to overcome in the world of computer vision with regard to lighting changes, but these pale in comparison to the greater picture – the work that is being done with edge devices and integration with the Internet of Things (IoT). Interconnected smart surveillance systems and increasingly autonomous IP solutions will become the norm in the near future. And who knows, perhaps in the distant future there may be fully predictive surveillance systems. Most have heard about the IoT, which is the network of physical devices embedded with software, electronics or sensors which allows these objects to collect and change data between IP addressable devices. There is a new term in the world of video analytics known as the Internet of Recognition (IoR), meaning computer image processing can recognise, through a single frame to many frames of video, capture and analyse any object or thing and intelligently analyse what it is and what it is doing. For example, is the object a skateboard, bike, car or truck? How fast is it going? What other data can be collected? Does it have words, phone numbers or logos and what else is in the scene? Is it a person or a crowd of people? Are they walking, running, climbing, sitting and so on. IoR will really start to bring everything together in the future and, as technology and computer processing evolves, there will be some exciting benefits.
Without doubt, video analytics will be one of the most exciting areas to watch and one of the most explosive areas of growth in the security field in the next five to ten years.