Video Analytics: Business Intelligence + Security

| 4 Comments | 0 TrackBacks

Page:   1   2   3   4   5  Next  »

The strongest business case for video analytics ultimately lies in the technology's ability to connect video surveillance to the strategic business objectives of an organization. In an enterprise in which security operations are integrated into larger business processes, analytics can produce an enormous payoff in terms of productivity, business intelligence and compliance.

Take the international clothing retailer Benetton, which is using video analytics to glean intelligence about customer shopping habits, potentially lucrative information. Yet when enterprises do discuss the strategic benefits of analytics beyond security uses, they often come up in the second round of talking points. This is unfortunate because that strategic focus, the market is getting bogged down in a debate about architecture, particularly whether edge-based or server-based analytics are superior.

"This is a big debate about something that shouldn't be debated. It's not up to the manufacturers to decide for users about what makes sense for them," said Ed Troha, managing director-global marketing, ObjectVideo, which supplies analytics chip sets for cameras and software for servers. "Analytics need to be placed where they provide the highest benefit to the user," he said.


"Video analytics is something that can produce benefits and value you cannot imagine--in business intelligence and operational efficiency," said Itsik Kattan, CEO of Agent Video Intelligence (Agent Vi) "Video analytics will be a driver for the surveillance industry, but the market still needs to be educated."

Video analytics consist primarily of software that "watches" the video image for movement, or more precisely, changes in pixel information. The software applies a series of rules-based algorithms to those changes to determine if the change should trigger an alarm. Most analytic systems currently on the market work effectively at identifying and reporting perimeter breaches ("tripwire"), loitering, object removed, object left behind and speeding. Applications such as people counting and license plate recognition have become more reliable, but more sophisticated functions such as facial recognition, despite the play they get from Hollywood and the media, still are not mature.

"A general rule is that if you can see it with the human eye, then you can deliver video analytics that will be able to see it," said Eric Fullerton, global sales and marketing officer for Milestone Systems, a video management system supplier that offers integration with more than 30 analytics vendors. "If it's something you can't see in the frame, you'll have problems with analytics."

Beyond Security Uses

Although the analytics market has been plagued by hype that oversold capabilities and created skeptics, users now are beginning to realize that the same analytics functions that can be used in a security environment can be applied to situations where lengthy and detailed observation--beyond the limits of the human attention span--can add significant business value. Falling prices, along with analytics software that can be added to individual cameras, have put the technology in reach of small and mid-sized users. The only question is how soon enterprises will recognize analytics technology has great value beyond its basic security applications such as perimeter protection and intrusion detection.

John with j&t.jpg"The Holy Grail is business intelligence," said John Whiteman, president of Americas at ioimage (pictured). People-counting software, for example, can detect when checkout lines are growing too long and alert a retail store manager to open more cash registers, Whiteman said.  

An analytic for advanced motion detection can be used at a food processing plant. If products start stacking up on a conveyor belt, an alarm will trigger "before you ruin thousands or hundreds of thousands of dollars of raw material," Fullerton added.

In a more detailed example, Benetton is using analytics software from Netavis in its 11,000-square-foot store in Vienna, Austria, for both theft prevention and business intelligence. A single Unix-based server from Sun Microsystems is supporting 30 cameras and five counting stations, according to Netavis CEO Wolfgang Baumgartner. An add-on Netavis software module, dubbed iCAT, handles people counting in front of the Benetton store and in different areas inside. The analytics application allows store managers to calculate the "conversion rate," that is, the number of customers who respond to a particular purchasing opportunity, such as an in-store promotion or markdown.

Page:   1   2   3   4   5  Next  »

The strongest business case for video analytics ultimately lies in the technology's ability to connect video surveillance to the strategic business objectives of an organization. In an enterprise in which security operations are integrated into larger business processes, analytics can produce an enormous payoff in terms of productivity, business intelligence and compliance.

Take the international clothing retailer Benetton, which is using video analytics to glean intelligence about customer shopping habits, potentially lucrative information. Yet when enterprises do discuss the strategic benefits of analytics beyond security uses, they often come up in the second round of talking points. This is unfortunate because that strategic focus, the market is getting bogged down in a debate about architecture, particularly whether edge-based or server-based analytics are superior.

"This is a big debate about something that shouldn't be debated. It's not up to the manufacturers to decide for users about what makes sense for them," said Ed Troha, managing director-global marketing, ObjectVideo, which supplies analytics chip sets for cameras and software for servers. "Analytics need to be placed where they provide the highest benefit to the user," he said.


"Video analytics is something that can produce benefits and value you cannot imagine--in business intelligence and operational efficiency," said Itsik Kattan, CEO of Agent Video Intelligence (Agent Vi) "Video analytics will be a driver for the surveillance industry, but the market still needs to be educated."

Video analytics consist primarily of software that "watches" the video image for movement, or more precisely, changes in pixel information. The software applies a series of rules-based algorithms to those changes to determine if the change should trigger an alarm. Most analytic systems currently on the market work effectively at identifying and reporting perimeter breaches ("tripwire"), loitering, object removed, object left behind and speeding. Applications such as people counting and license plate recognition have become more reliable, but more sophisticated functions such as facial recognition, despite the play they get from Hollywood and the media, still are not mature.

"A general rule is that if you can see it with the human eye, then you can deliver video analytics that will be able to see it," said Eric Fullerton, global sales and marketing officer for Milestone Systems, a video management system supplier that offers integration with more than 30 analytics vendors. "If it's something you can't see in the frame, you'll have problems with analytics."

Beyond Security Uses

Although the analytics market has been plagued by hype that oversold capabilities and created skeptics, users now are beginning to realize that the same analytics functions that can be used in a security environment can be applied to situations where lengthy and detailed observation--beyond the limits of the human attention span--can add significant business value. Falling prices, along with analytics software that can be added to individual cameras, have put the technology in reach of small and mid-sized users. The only question is how soon enterprises will recognize analytics technology has great value beyond its basic security applications such as perimeter protection and intrusion detection.

John with j&t.jpg"The Holy Grail is business intelligence," said John Whiteman, president of Americas at ioimage (pictured). People-counting software, for example, can detect when checkout lines are growing too long and alert a retail store manager to open more cash registers, Whiteman said.  

An analytic for advanced motion detection can be used at a food processing plant. If products start stacking up on a conveyor belt, an alarm will trigger "before you ruin thousands or hundreds of thousands of dollars of raw material," Fullerton added.

In a more detailed example, Benetton is using analytics software from Netavis in its 11,000-square-foot store in Vienna, Austria, for both theft prevention and business intelligence. A single Unix-based server from Sun Microsystems is supporting 30 cameras and five counting stations, according to Netavis CEO Wolfgang Baumgartner. An add-on Netavis software module, dubbed iCAT, handles people counting in front of the Benetton store and in different areas inside. The analytics application allows store managers to calculate the "conversion rate," that is, the number of customers who respond to a particular purchasing opportunity, such as an in-store promotion or markdown.

<!--nextpage-->

Studies have shown the conversion rate can objectively measure the effectiveness of sales, advertising and other promotions, and be used to fine-tune the ratio of customer traffic to salespersons on the floor. An uptick of a few percentage points in a store's conversion rate can contribute considerably to the bottom line.

Yet Benetton is more of an exception than the rule. Despite the potential information analytics can provide the executive suite, the purchasing decision still lies with physical security departments. "It's still a security product," said Agent Vi's Kattan. "Security personnel want an alert. Security personnel are going to focus on costs and false alarms."

At the same time, IT departments, which might appreciate the strategic dimensions of analytics technology, remain put off by their complexity and remain stung by the unfulfilled promises of the first analytics promotions. .

So while an understanding of analytics can be pivotal in changing senior management's view of physical security from a cost center to a strategic center and, at the same time, provide valuable data for IT-based decision support and enterprise resource planning systems, neither the physical nor the IT side of security seems to be taking up the analytics-as-value message.  

That failure to grasp the business value of analytics  has led to a distracting debate in the marketplace over the architecture of analytics, specifically, the relative benefits of edge and server-based analytics platforms. Yet, in the end, defining choices in terms of these stark alternatives may be counterproductive both for security and business uses

Linking the "where" to the "why"

The first generation of video analytics was largely server-based systems. Then the growth of digital IP networks, open architectures and the declining cost of processing made it possible Ruth Gratzke_SES.jpgto integrate analytic functions in cameras--at the "edge"--making the technology economical for smaller operations. Companies such as Bosch Security Systems, ioimage and SightLogix introduced analytics designed to be integrated with, or attached to, surveillance cameras. Edge analytics sales now dominate with some 80 percent of the total market, estimates Ruth Gratzke, general manager and head of sales at Siemens Building Technologies Security Solutions' Global Center of Competence for Intelligent Video (pictured), although that figure reflects the large number of edge deployments among small and medium-sized organizations. At the high-end, there is far less discrepancy between edge and server solutions.  

"We believe that when compared against edge-based analytics, server-based analytics offer substantial cost benefits and superior feature sets for larger scale applications in the critical infrastructure market (seaports, airports, power plants, etc.), and have the potential to dominate this market segment in the long run," Gratzke wrote in an email.

Niall Jenkins, a policy analyst at IMS Research and author of an analytics market report scheduled for release this month, agrees to an extent. "Everyone thought edge [solutions] would be a winner," he said. "Server might be stronger than thought." Still, he added that no trend was emerging yet.

<!--nextpage-->

Edge solutions have benefited from the latest spate of digital signal processors, including Texas Instruments' DaVinci, Cernium's P-Core and Xilinx's Freescale. There's no question that the closer the software is to the raw image, the better the results will be, said Kattan. But loading up on analytics applications ultimately forces a tradeoff. "Motion detection, people counting, abandoned object - each is a separate algorithm," Kattan said. "The limit is ultimately reached. You can't continue to compromise on performance. We need some processing power at the server."

"Server-based analytics are coming back a bit, but most larger vendors think you'll need both," Jenkins said. Milestone's Fullerton agreed, and said the company's XProtect Analytics Framework is designed to support edge- and server-based solutions. In a follow-up interview, Siemens' Gratzke also said that ultimately, large corporate deployments are likely to mix edge and server platforms.

Processing power is key

Moreso than the edge-server debate, the real decision on an analytics platform may come down to the processing power the user will need--and that may differ by location and the primary use of the visual data being collected.  

At the camera, the more power allocated to analytics, the less there is for image encoding and recoding. At what point do analytics algorithms become complex enough that they need to be performed centrally? Moving price points complicate the question all the more. Will the pressure of Moore's Law keep edge analytics economically viable or will there be a certain point at which server-based is more cost-efficient, bandwidth issues or no? These are questions users, not vendors, must answer.

Currently, most server-based analytics software can handle between 16 and 32 cameras. Individual cameras do not have to be replaced or retrofitted. Edge-based analytics do their processing in the camera or encoder. The benefit is that with edge analytics, only images that generate alarms have to be transmitted in full resolution and frame rate. When the surveillance system is running on an enterprise network, this keeps bandwidth congestion down. Also, because edge analytics have access to the entire uncompressed image that the camera sees, they  have more information to work with. Therefore false alarms are reduced, yet questionable activity is more likely to be captured.

To get the same performance, server-based analytics must be able to work with uncompressed, high-resolution images. In such a set-up, cameras would have to be transmitting video constantly at full frame rate, as well as all the metadata, the necessary overhead code that contains information about the image data. Gratzke said large users can manage these environments. Others, like Doug Marman, chief technology officer and co-founder of VideoIQ, suggest that no matter how big the operation, bandwidth is always an issue.  

Why Not Both?

VideoIQ and Agent Vi and ObjectVideo are positioning themselves as the vanguard of a third generation of analytics that attempt to get the most from the edge, but in some cases, combine edge and server functions in the form of a hybrid solution. Agent Vi and ObjectVideo systems Although the approach of each vendor differs, the general goal is to perform a degree of analytic processing at the camera, where image information is most complete, and then backhaul footage to a central server for heavy algorithmic lifting. VideoIQ consolidates algorithm functions in the camera (for more information, see comments).

IKattan.jpgAgent Vi's Vi-System 3.3 software, for example, extracts information (in the form of a block of pixels) out of the image, but doesn't do the analytics, explained Kattan (pictured). The selected information and metadata is transmitted to a server, where analytics is done. "We're overcoming limitations of processing power [in the camera], but the solution is more than server-based," he said.

This architecture, say its adherents, both preserves the open systems' benefits of edge analytics yet, on the server end, allows more cameras to be centrally managed. "We can process 100 cameras at the server while getting a full algorithm," Kattan claimed. This can pay off in very large installations in major cities, where hundreds or thousands of cameras need to be supported. In these situations, he said, a server-based solution alone "becomes a significant challenge not only from a cost perspective but from a technical-performing perspective."

VideoIQ employs a similar approach. "Everybody's looking at pixels. We look for patterns in the pixels," said Marman. This pattern isolation uses about one-eighth of the processing power.

<!--nextpage-->

Hybrid approaches, say their vendors, also refine and exploit the new network-centric aspects of security by enabling analytics servers to interoperate with physical security information management (PSIM) systems that can combine analytics with other security and enterprise data and plot a situational response. For example, analytics information can be combined with other Web-based applications, such as Google Maps, to present responders with the full breadth of a situation, said Marman.

To be fair, vendors of server-based and edge-based systems say their systems also offer interoperability. Here's where users need to think critically, however. Many server-based systems, including Siemens', have their roots in end-to-end closed systems. While they may be open to the extent that they accommodate various types of cameras and NVRs, users should vet them for back-end compatibility with other situation management systems, not to mention business intelligence systems, if they are thinking in that direction.

There are some exceptions, notably Netavis, which claims to be fully interoperable with different security systems and business applications.

Economy at the edge

Most edge-based systems are open by design. And while vendors of server-based solutions tend to pigeon-hole edge analytics as suitable primarily for smaller deployments, high-end customers have chosen edge architecture, countering such claims. For example, the Port of New York and New Jersey Authority is using ioimage's edge-based analytics with hundreds of cameras..

Bolt On Diagram.jpgEdge-based analytics also can be deployed incrementally. This means investment can be managed, a plus in the current down market, said Bob Banerjee, product marketing manager for IP video products for Bosch Security Systems. Add declining per-camera costs to bandwidth savings, and there is a powerful case for what Banerjee calls "bolt-on analytics" at the edge (click on graphic for larger version).

For smaller enterprises that nonetheless face workplace compliance or security issues, "bolt-on analytics" costs can be measured in hundreds of dollars instead of thousands. "This brings analytics within reach of smaller installations, he said."

Consider a small employer who must enforce a no smoking law. If smoking is illegal on a fire escape, the business owner might want to add intelligence to an analog camera already covering that side of the building. Still, the rule is only broken occasionally, and the owner is no position to monitor the camera feed constantly, so an analytics-based notification system would make sense. "But how many thousands of dollars do you want to spend to catch one worker?" said Banerjee.

Edge-solutions have made analytics much more economical for these situations. While not as powerful as higher-end solutions, they can offer enhanced security mechanisms that were not previously affordable for most to small and medium-sized businesses. To address the processing power issue, Banerjee said Bosch is planning to introduce a new platform at the end of August using two chip sets: one for encoding and recoding and one for analytics. Furthermore, the chips will use field programmable gate arrays to allow an additional degree of customer configuration.

<!--nextpage-->

Simplicity and Performance

Bosch also touts ease of configuration. Most of the time, its Intelligent Video Analysis (IVA) -based on-camera analytics can be configured in two hours minutes, Banerjee said. The analytics algorithms themselves adjust to their environment in one week seconds, he added. The company feels confident enough in its IVA solution to offer a one-month free trial.

"Lots of users think [analytics are] burdensome," said ObjectVideo's Troha. "The big push is to make it easier to implement as a high-value ingredient--the solution only where and when it's needed."  

Implementing analytics might be easier than in the past, but there are tricks to optimization, and users should be prepared to work with a system for several weeks. "The biggest difference in performance comes for the amount of time the devices have to experience changes in the environment," said ioimage's Whiteman Variables can be clouds rolling, trees swaying, shadows and reflections off puddles.

That's a principal reason why performance standards have been difficult to pin down. "There's no acceptable failure rate; no baseline for video quality," said Siemens' Gratzke.

Camera placement, lighting, weather and background can have a substantial effect on how well systems work. An analytics system deployed on network cameras in a one-acre parking lot in a wooded area that experiences regular season changes, such as Concord, N.H., can produce a much higher false alarm rate than the same manufacturer's system, deployed on the same number of cameras in the same configuration in the same-size parking lot in a desert environment such as Tucson, Ariz.
 
Naturally, users want failure rates as low as possible, but external variables, from weather to the choice of transmission platform (e.g., wireless) prevent any vendor from guaranteeing an analytics performance level. At best, they can work to optimize analytics performance for a given installation, Gratzke said.

Beware of 'low-touch' and 'no-touch' systems and vendor claims that their analytics self-learn, advised Kattan. "There's no such thing. Make the process as simple as possible," he added.  "Define at the workstation the acceptable rules. Find the most appropriate camera placement. For motion detection you'll want it one way, for counting people you place it somewhere else."

Perhaps most important, Kattan and other vendors that stress open systems urge users to see analytics as part of an overall integration strategy. In essence, they say, analytics turns cameras into sensors that convey information to a PSIM system where it can be processed and presented as part of a larger event management system, then stored for later forensic analysis. "In a few years, you will not be able to separate video from access control or fire alarms," said Netavis' Baumgartner. The forensic aspects, however, won't stop there. Analytics will continue to pay off as the support other enterprise systems. Despite current skepticism, analytics will ultimately be instrumental in the transition of video surveillance from a security tool to a mission-critical IT component.

Question: To what extent, if any, are you using video analytics? Are they primarily used for security applications or are they contributing to business intelligence or decision support?

No TrackBacks

TrackBack URL: http://www.securitysquared.com/cgi-bin/mt/mt-tb.cgi/79

4 Comments

The loan are important for people, which would like to start their own organization. In fact, that is very comfortable to get a student loan.

Hi Steve -
Very good piece and thank you.

One correction I would offer that is an important one:

In the section titled "Why Not Both?", you say, "Agent Vi and ObjectVideo systems perform a degree of analytic processing at the camera, where image information is most complete, and then backhaul footage to a central server for heavy algorithmic lifting."

I cannot speak for the capabilities of Agent Vi's solution, but in the case of ObjectVideo's analytics on the camera or edge device, this statement is completely false.

One of the advantages our many manufacturering partners enjoy embedding our analytics into their devices is FULL ANALYTICS PROCESSING at the device level, with NO EFFECT ON PERFORMANCE. As a consequence of this, the only footage that needs to be "backhauled" to the server is that which the user cares about - not EVERYTHING.

The notion that there are "limitations with processing power [in the camera]" is being advanced by analytics providers who can't or won't do it. ObjectVideo analytics on the chip use less than 20% of a DSP and less than 8% of an Intel processor for the FULL suite of analytics capabilities on the chip, which allows the most flexibility in determining analytics functions for integrator AND end user.

Thanks,
Ed Troha
Managing Director, Global Marketing
ObjectVideo

Doug,

Thank you for the clarification, which I have incorporated above. You provide further depth into wsys one can deal with the processing power issue, but I concede I did not get into how advances in algorithm functions can do this. You have a good example here.

Steve

Steven,

Another interesting and well written article. I've enjoyed many of your articles. There is some useful information here.

I wanted to offer a few small corrections and some other thoughts. The field of video analytics is a complex one, and there are more differences than most people realize.

First, a small typo: You quoted Kattan and referenced in one spot that he was from VideoIQ. Well, we would be glad to welcome him to talk to us here at VideoIQ, but I believe he currently works for Agent-VI, as you mentioned later.

Later, you quoted Kattan, who said that motion detection, people counting and object removed requires a separate algorithm for each. That might be true for Agent-VI, and I agree it is true for many systems, but it is not true in our case. Whether you are detecting people entering a region of interest, or counting people entering a doorway, or looking to see if an object has been removed, we use the same algorithms for the detection. What changes are the rules, but rules logic takes very little processing power compared to the analytics.

I'm sure this varies, but in our case we have no problem running all of these at the same time in our cameras and encoders at the edge. We can have 10 different rules running, some looking for vehicles, some for people, some for boats, for example, all at the same time in every camera.

As you point out, the question comes down to processing power, but this can be solved two ways: By using a larger processing engine, or by using a more efficient algorithm. In other words, there is a wide range of performance. While one technology might require 8X as much processing power, and therefore needs to run in a central server, another technology can easily accomplish the same results or better right in the camera. This is one of the main reasons to choose between edge vs server, but this varies with each technology.

We have seen no cases in which server based processing is required to get the quality of detection accuracy needed with our technology, although if you were dealing with a 10 megapixel video stream, that might be different. But then, you couldn't afford the bandwidth to bring that video back to a central server.

You mentioned that VideoIQ was using a similar approach to Agent-VI and ObjectVideo, by doing some of the processing at the edge and some in a central server, but as you might guess from what I wrote above, this isn't true. We do all of our processing at the edge, with our iCVR edge based products.

We do have server based analytics that we have sold for years, but our edge based products need no servers for any of the processing. They are complete self-contained systems.

You said that implementing analytics might be easier than in the past, but it still requires tricks to optimize and the installer needs to be prepared to work with a system for several weeks. This is a topic close to our heart, since we've designed a system to solve this problem. We've gone to great lengths to develop an algorithm that needs no manual tuning or calibration, since it is completely self-learning.

Therefore, we wholeheartedly disagree with Kattan who makes the bold statement that there is no such thing as a technology that needs no-touch or low-touch. And if he really did work for VideoIQ, he would know this. Our edge based cameras and encoders have no way to tune or calibrate them, since it all the optmization takes place automatically.

However, this doesn't mean that there aren't a few sites where you might need to adjust what your region of interest is, since as with any technology, there are the few cases that need extra care. But in most cases it is as easy to set up as video motion detection - with the accuracy of true video analytics.

Thanks.

Doug Marman
CTO
VideoIQ

Leave a comment