printer friendly version

Analysing video analytics

The end of the year brings us back to the topic of video analytics. We expect so much of technology that we often disappoint ourselves when it fails to perform miracles, and then we also fail to see when it does a good job in certain areas. Video analytics falls into this space, we want Star Trek functionality for a low cost, often ignoring the bits that do work well and are often offered for free.

In this Q&A session, Hi-Tech Security Solutions asked a few industry experts what was happening in the world of video analytics. Are we seeing this segment of the market developing into a profitable industry, or should we expect to see IP cameras including more analytics on board to differentiate the top-of-the-range hardware from the cheaper imports that are quickly matching image technology with the traditional leaders? And, of course, we also wanted to know who was doing what in this growing field.

Our experts for this article are:

• Roy Alves from Axis Communications,

• Gus Brecher from Cathexis Africa,

• Brendon Cowley from C3 Shared Services,

• Laurence Smith, Graphic Image Technologies, and

• Brian Wynberger from Reditron.

Hi-Tech Security Solutions: Are we seeing more high-end camera providers including more analytics on board their cameras to differentiate them from the cheaper imports? Or are we still seeing basic analytics provided for free with the hardware and more advanced systems being sold separately?

Roy Alves from Axis Communications.

Roy Alves, Axis Communications: We are seeing the three trends mentioned at once and then some. High-end camera manufacturers are definitely including more complex analytics to add value either by developing it themselves or by providing an open platform for third parties to do so. We also see other players, for example VMS providers, sourcing and including video analytics as an add-on to their offering. In all cases, the most important thing is that what is included provides real value and is cost effective from a technical as well as a commercial standpoint. One can compare this with, for example, apps for mobile phones.

Gus Brecher from Cathexis Africa.

Gus Brecher, Cathexis: There are some manufacturers offering onboard analytics, but these are few and far between. The main reason for this is that the cameras themselves have limited processing power, and as processing power increases the manufacturers are using this for video-based things like wide dynamic range enhancements, image stabilisation, increased resolutions and frame rates, multi-streaming etc. Also, because of the processing power requirements, when analytics is used on the cameras, you need to sacrifice elsewhere and this can only be in the form of reduced frame rates or resolution.

Brendon Cowley from C3 Shared Services.

Brendon Cowley, C3 Shared Services: More high-end camera providers are adding analytics to their camera portfolio. We see a trend, followed by acquisitions from large VMS manufactures of specialised video analytics companies, where complete solution providers that offer VMS and cameras, acquire high-end analytic companies and add this feature to their cameras.

In addition, we see a trend where high-end camera companies provide an open camera platform enabling third-party analytic companies to embed the analytics into the camera. In this case, the integration is not full and the performance is not optimised.

Brian Wynberger from Reditron.

Brian Wynberger, Reditron: The same analytics is still supplied by high-end manufacturers, but with increased accuracy. Asian suppliers are now including analytics from basic features to high-end cameras with more advanced analytics – basic is still free.

Advanced analytics is becoming more possible at camera level due to advancement in technology, and integrated and sophisticated analytics is still being offered on hardware/server system and sold separately.

Hi-Tech Security Solutions: Is on-board analytics a good approach as opposed to server-based analytics? What are the pros and cons of the two approaches? Are customers buying into edge analytics?

Alves: It is fair to say that the most complex and sophisticated solutions will continue to come from server-based analytics also going forward. What is embedded today was server-based 10 years ago (or less), that is the way technical development goes. One could view server-based solutions as the development platform for future embedded analytics. Customers are definitely buying into edge analytics, which come with the advantage of being infinitely scalable, decentralised and cost effective – requiring no additional hardware and normally operates with very low bandwidth requirements, installation and maintenance costs.

Brecher: For the reasons mentioned above, we are finding that server-based analytics is still the one of choice, unless a camera is being used specifically for analytics. People counting is a good example, where the cameras are situated in a top-down orientation which is not conducive to general monitoring. Where you want to use the camera for its primary purpose, it is still better to do analytics on the server.

Laurence Smith, Graphic Image Technologies.

Laurence Smith, Graphic Image Technologies: While there is a trend where edge-based (on-board) analytics are preferred over server-based analytics, these two can work well in parallel. DVTEL offers both options, where outdoor analytics are recommended with edge-based analytics, while indoor analytics are recommended with server based systems.

Edge-based analytics tend to deliver better performance, as the analytics run on the raw video rather than on encoded video where some information is lost during the encoding process. In addition, having the analytics on the camera and enabling the analytics to have a feedback to control the sensor optimises the video input to the analytic engine. These optimisations enable overall improved probability of detection and lower false alarm rates. In terms of installation requirements, having the analytics on the edge reduces the infrastructure and network requirements, which dramatically reduces the customer’s cost of ownership, and minimises time for return of investment (ROI).

Since developments for the server-based analytics are much faster and computational resources are higher, server-based analytics tend to offer a wider application portfolio. For example, it is ideal for people counting, licence plate recognition (LPR), face recognition and sex and age classification – which is used in retail. Server-based analytics is usually run on a few cameras, hence it is a single point of failure compared to edge-based analytics.

Overall, these two solutions can work side by side, and some cameras will run the edge-based analytics, while others will use server-based analytics.

Wynberger: As technology increases, on-board or edge analytics becomes more useful. Edge analytics can be used together with server-based systems from selected suppliers.

On-board analytics

Pros:

• No need for additional servers – lower cost – easy setup.

Cons:

• Cameras do not have sufficient processing power to run analytics.

• No redundancy.

Server-based analytics

Pros:

• Very scalable – multiple rules on many cameras at the same time.

• Integration to third-party VMS and database management systems.

• Option of redundant servers/failover is possible.

Cons:

• Requires regular maintenance and upgrades.

• Relies on network connection – if this fails, the analytics fail to generate alarm events.

• Setup is more complex.

Hi-Tech Security Solutions: How have analytics changed? Are we seeing more behaviour-based analytics as opposed to the rule-based approach? Is there really a difference in the results obtained?

Alves: It is definitely true that retail analytics is developing from traditional door counting (counting entrances and exits) to more behaviour based functionality, as in monitoring queues, dwell-time, heat mapping and demographic-related applications.

Brecher: There is a trend towards some behaviour-type of analytics, but these are only used in very specific applications. Analytics by nature requires effort in configuration and the more sophisticated it is the more effort is required. Having said this, things like loitering, queue length analysis, direction, speed etc., are being used a lot more in various environments.

Cowley: The main changes in analytics are overall performance and credibility improvements, as well as ease-of-use. Behavioural analytics have been developed but have not yet matured enough to be widely used.

Wynberger: There is increasing interest in behaviour-based analytics packages. Behaviour-based analytics makes the use of triggered events more useful to law enforcement, crime prevention and behavioural trends. The reducing of human involvement in analysing behavioural events is beneficial and allows more areas to be analysed simultaneously.

Hi-Tech Security Solutions: Would it be ­correct to say that the primary focus of analytics today is on alarm verification and perimeter security – because these are delivering the best, or most measurable results?

Alves: I would agree with that comment, analytics have the ability to remove the human aspect out of the video monitoring and in both of these instances it provides a pre-warning. With more cameras being rolled out, often it becomes increasing difficult to reply on operators to pick up specific detail from strategically placed cameras. Also, for example, we find that many customers look to analytics to get the most from an infrastructure perspective. If they are moving from analogue cameras to IP and today have a door-beam counting solution, they may look to see if the new infrastructure can solve this task with a single intelligent Axis IP camera. So, if alarm verification is what they need, they will seek that.

Brecher: This is the case, but there is also a lot of pull from the likes of retail, banking and logistics environments for analytics to provide marketing and operational information. So things like store heat maps to give information on how people move around stores and which areas are busiest, people counting tied to point-of-sale to give conversion rates etc. There is also a requirement for face recognition in several sectors.

Smith: Analytics aimed for alarm verification and perimeter security have proved to be very reliable, efficient and are widely used. In addition, the requirements for this market are obvious and sometimes required by government regulations or insurance companies. Other analytics are also being developed for other markets, such as retail outlets analysing customer movement and attraction.

Wynberger: Yes. Alarm verification via the video image makes the overall use of CCTV more cost effective, as opposed to incorporating intrusion detection devices. The use of tripwire as opposed to video motion detection results in a huge reduction of false alarms and saves on storage requirements. This has a far wider appeal than other forms of analytics.

Hi-Tech Security Solutions: Is the analytics market becoming part of the Big Data trend in which security data is being used for business intelligence? Or does that only happen in specific instances where standard analytics applications are integrated into customised solutions?

Alves: Yes, that is definitely the trend. I would not call it security data though. The data collected stems from the same infrastructure doing the surveillance, but as is the case in retail analytics, it is not the loss prevention team who is interested in, for example, the traffic flow, queue lengths and occupancy in the store, but the sales and marketing team. In fact, retail analytics is a completely different segment.

Brecher: There is a trend towards information from surveillance systems to be used for business intelligence. And while surveillance along with the associated analytics is being used for security, the system really can perform multiple roles in the organisation (operations, marketing, health & safety, training, personnel). Security is only one useful aspect of the system.

Cowley: We see a trend where very big (mostly government) customers raise the need for big data business intelligence. Current requirements are now on an academic research level, and we might see such research develop into products in the future.

Wynberger: As accuracy and prices decrease there will be more interest in using analytics to gain business intelligence. At present, server-based analytics is used or added to a system to achieve this when the benefits justify the expense. Most applications still just use the no-cost analytics as alarm verifications.

Hi-Tech Security Solutions: Are we seeing standards developing for the analytics market, from ONVIF, for example? Will standards be good for the market, especially customers? Will developers be interested in standards as they may lose their market advantage?

Alves: As it stands today we see no standards in analytics. I am sure that developers will engage in discussing/developing a strategy if they saw that it would be beneficial to them. As standardisation tends to drive market development in a positive sense both for developers and customers – there are numerous examples of that (like ONVIF and Bluetooth) – it is probably just a matter of time before it will surface.

Brecher: There are no standards at present. The algorithms are so specific and unique that I don’t see this happening. There are some organisations that qualify the analytics for use against a baseline, but this is more around the functionality and integrity of the algorithms rather than the algorithms themselves.

Smith: The trend of standards is widely used at the VMS and camera level, with the current trend being storage-on-the-edge integration. However, analytic events are not yet fully generic. Analytics interfacing using other peripheral methods, such as IO Open Network Video Interface Forum (ONVIF), emails and File Transfer Protocol (FTP), still allow standard integration. This standardisation will allow customers with existing systems to easily adopt new technology while improving parts of the system, rather than having the need to change the complete system. For example, adding new analytic cameras to existing sites with existing VMS without requiring special integrations between the components.

A few companies have a complete system to go to market with and will most likely not adopt these methods, while companies that offer cameras or VMS will tend to be more open.

Wynberger: No, the sharing of IP (intellectual property) from manufacturers will hamper putting open standards into practice as they do not want to freely give this information to the general market. This will stifle innovation from developers wanting to maintain profitability in an increasingly competitive industry.

The market is still naïve and untrained in even being able to formulate applicable standards. Only the standardisation of alarm metadata across platforms is useful.

Share this article:

Further reading: