Demand for video analytics is rocketing as more and more products incorporate the technology. A few bumps on the road remain. A&S examines where this important technology is heading.
According to IMS Research, the video analytic market will grow from US$67.7 million in 2004 to $839.2 million in 2009 at a compound annual growth rate (CAGR) of 65.5 percent. IMS Research's report on video content analysis (VCA) highlights the rapidly growing trend towa rd embedding VCA algorithms into security devices, such as DVRs, video encoders and network cameras. DSP-based analytics enables these devices to analyze live or recorded video streams to detect predefined suspicious activities, events or behavior patterns.
IMS also predicts that in the short term the biggest penetration of VCA will occur within DVRs and video encoders, said Market Research Analyst James McManus. "This is due to the large installed base of analog cameras. Intelligent encoders offer a cost-effective way to convert to a networked system and analyze video at the point of capture."
A number of suppliers, he continued, are already offering network cameras with a full suite of embedded analytics. "We predict that, in coming years, there will be even greater penetration of VCA as network camera manufacturers increasingly embed low-cost, entry-level analytics for advanced motion detection or system health monitoring." Shipments of network cameras with embedded VCA are expected to exceed 800,000 units by 2010.
"There will be a greater emphasis on moving intelligence out to field devices to enable pre-transmission processing and bandwidth preservation, "said McManus. 'Intelligent edge devices eliminate the need to t ransmit al l captured video to centralized servers for analysis. This is a more cost-effective way of implementing VCA than a server-based approach and will lead to far greater uptake of VCA technology."
Janos Kophazi, General Manager of Intellio, reported that the market for megapixel cameras is growing at more than 100 percent per year, and it is estimated that 500,000 such cameras will be sold every year by 2009. "All of our megapixel cameras now have embedded video analytics."
Ariel Frischoff, Vice President of Strategic Alliances at Agent Vi, estimated that the market for video analytics was worth $100 million in 2007, split equally among the U.S., EMEA and Asia. He also noted that ABI Research forecasts that the video surveillance revenue would expand from $13.5 billion in 2006 to $46 billion in 2013. Those figures include cameras, computers, storage, professional services and hardware infrastructure everything that goes into an end-to-end security system.
While there are multiple reports on market size, ranging from $100 million to $500 million, said Dvir Doron, Vice President of Intelligent Video Appliances, ioimage. "We believe that $100 million are at the lower end." The Americas, he said, is 40 percent of the total market, EMEA 40 percent and Asia Pacific 20 percent.
Sean Shankar, Vice President of Marketing at IntelliVision, put the numbers
lower. "The global market for video analytics is $50 million. This is based on our estimates and from discussions with industry analysts. We expect this to grow rapidly to $200 million in three years. The growth will stem from increased embedding of video analytics in cameras and recorders."
The U.S., said Shankar, accounts for 50 percent, Japan 25 percent and the rest of the world (including Europe) 25 percent. "The challenge in giving exact figures is that while much of the analytics may be sold to manufacturers in Asia (Japan, Taiwan, Korea and China), the final products are sold internationally."
It remains a mostly mid to high-end market. "When I was in the DVR industry," said Doron, " the view was that 15 percent of the market is high-end and 65 percent mid-range. Historically, video analytics has been concentrated on the high end, but there has been a tremendous shift over the past two years to mid-range. The high end is now about 20 percent to 25 percent of the market with the mid-range at about 50 percent to 55 percent. There is no low-end market for video analytics."
In different vertical markets, observed Frischoff, there are different competitors. "For general applications, you see ObjectVideo, aimetis, iOmniscient, Vidient, Mate, ioimage, Nice and Bosch. In retail, there is IntelliVid; in traffic, there are Trafficon and CityLog."
Vidient and Cernium, said Frank Pao, President and CEO of Vidient, are more visible in North America (mostly the U.S.), and there are a lot of Israeli players, such as Mate, in the Asia Pacific region. Agent Vi, he said, is more active in the U.S., while in Europe, there are many smaller local companies.
For Kophazi, the fact that there are too many competitors in video analytics makes it hard to mention anyone specifically. " The major competition is coming from major global brands like Siemens, Bosch and Sony," he said. "All three recognize that market trends are toward video analytics, though they have only started development; there is no broad range of applications widely available on the market."
Focusing on Non-Security Applications
The U.S. government is spending primarily on R&D for specific functionalities , said Ed Troha , Director of Global Marketing for ObjectVideo. "We have 33 of the smartest computer vision scientists in the world working on a variety of government projects. Eventually, the functions that they are developing will find their way into commercial applications."
Troha is seeing greater developments in event counting, which accurately counts people or objects in specific areas. 'There are lots of interesting ramifications outside the security model," said Troha. "This could involve safety, say four people congregating in a high-crime area, or it could measure crowd density on subway platforms. When numbers increase to a certain level, the system could send a message to call for more cars at a greater frequency." Outside of physical security, he sees the most traction in retail, banking and gaming. Transportation has seen less traction, though, because the purchasing decision-maker is usually a government or regulatory agency.
Retail applications determine the amount of time that someone stands in front of a product display and on any given day. "This enables store owners to charge suppliers premium prices and have empirical data as proof of their store's 'hot spots,'" Troha said. Systems that enable occupancy data queries and dwell time with conversion to actual purchase have been selling well in the market in the past 12 months.
Shankar repor ts seeing steady growth in transpor tation, retail, heal thcare and satellite -based detection. "These are not necessarily security-related. Wherever cameras have a role to play, video analytics will have applications. Use of video analytics and millimeter cameras to detect hidden objects (shoplifting) in retail is especially important."
"We are in five major areas: airports, railways, roads, shopping centers, and the oil and gas industry," said Rustom Kanga, CEO of iOmniscient Intelligent Surveillance. While the first three are all in the transportation segment, they represent separate businesses. "We also have a certain amount of business looking after large events like football stadiums, exhibition centers and major events like the Commonwealth Games."
In Frischoff's best estimate, security accounts for 30 percent of the demand for video analytics, retail 15 percent, transportation 25 percent, traffic 15 percent and city centers 15 percent. He also sees video analytics becoming a standard in many tenders. "The technology has reached a maturity level that allows a market penetration of 30 percent to 60 percent in new installations, compared with 10 percent to 20 percent two years ago.
The most significant new uses, said Frischoff, are improved operational performance and business intelligence. Retail and banking are two industries that are utilizing analytics for applications beyond security. "Retailers are using video analytics for restocking alerts for high-turnover merchandise, determining how much time customers spend in front of a display and monitoring numbers of people passing through store entrances and exits as well as other areas of interest."
Financial institutions, he continued, use video analytics for counting numbers of people passing through bank entrances and exits and other areas of interest; measuring length of queues at teller stations, ATMs and drive-up windows; and calculating periods of peak activity and duration of transactions to improve customer service.
For Kophazi, major verticals include retail, transportation (license plate recognition) and art galleries (stolen object detection). Other applications, he stressed, will definitely involve retail.
Critical infrastructure, noted Doron, remains an important vertical, government and the military less so. In the mid range, manufactur ing, industr y, education, car dealers, and small and medium-sized business are prominent verticals.
Defining High End
Video analytics comprises many technologies that can be of value to the customer, some new and some old, said Troha. "There are plenty of people who call their products intelligent. While that may be true in the broadest sense, it is not the same as having computer visionbased analytics, as there is no active analysis of the scene (at any time)." In addition, intelligent systems create streams of metadata and have the ability to separate foreground objects from static or moving backgrounds, which in turn enable accurate classification. "You need all three of these criteria for true intelligence," stressed Troha.
In Troha's view, a high-end solution is one that has capabilities and functionalities that justify user investment in video analytics. "We concentrate on working with the end user closely to understand their problems and solve them with the appropriate application of video analytics," Analytic-enabled solutions must have a comprehensive set of features and functions, while providing easy integration with user environments. And, they must be priced right. "While an unlimited budget will get customers whatever they want, that is an unrealistic expectation," said Troha. "Analytic solutions have to be usable, practical and help the end user make better use of people and processes already in place."
To define high- end products, Shankar recommended looking at how long the
company has been in business and the type of products that it sells. "Customer references are an excellent way to determine this as long as they, too, are high-end bus ines ses." He al so sugges ted determining whether the company is active on the embedded as well as server side. "Finally, you may wish to look at the company's track record and product mix. Is it selling the same things over and over again or is it moving forward?"
Doron said the high end entails a large number of cameras, sites and customized applications. "Integration of many different components is involved. Of ten, these solutions are used for critical infrastructure, transportation and military or border security." He questioned, however, whether a project with 10 or 16 cameras in a museum is mid range or low end. "There has to be some level of appreciation for the brand and/or solution and not just price," he said.
"You have to determine whether the system has to operate in difficult conditions like crowds," said Kanga. "Simple systems can detect whether a person has jumped a fence, but what about a left bag or a person falling in a crowded railway station or airport? These sites would require much higher levels of technology." Second, he said, is how robust the system is. "It has to be able to cope with a large number of false alarms caused by, say, light changes, moving leaves, shadows, reflections and water. Ver y few companies can do this." The third is the ability to understand what is happening in a scene rather than merely determining motion.
For Kophazi, high end is definitely a megapixel solution. "This should provide high-resolution images in real time for monitoring and future analysis. You do not get that with MPEG-4, which provides only VGA resolution." VCA embedded on edge devices, he added, lowers bandwidth requirements, increases speed and lessens storage required. It also makes the system more futureproof. "High end is about complete solutions encompassing cameras, recorders, intelligence and software. Another factor is having a robust system. If a server goes down due to technical difficulties, other servers must take over so that there is no loss of images." Finally, system scalability, easy upgrad ability and open architecture are key as they enable easy integration with other camera brands.
Selection criteria, said Frischoff, should include solution scalability, compatibility and ease of integration with existing systems, as well as cost to performance and return on investment, especially in retail and traffic applications. Other variables are ease of use, reliability, accuracy and ability to adapt to different environmental conditions (night, inclement weather).
Technological hurdles result from the fact that video analytic technology is still new. "Development of our own technology," Kanga said, "was first initiated 16 years ago and commercialized only three to four years ago. The technology still has a long way to go."
There are several unresolved problems, he said, such as tracking from nonadjacent cameras. "You could have 10 people walk into an elevator, all leaving on different floors. How do you track each one? If you are tracking the guy in the red shirt, what do you do if he changes in the elevator? While this would not be a problem with overlapping cameras, you do not have that ability here. You would need to look at other characteristics, not just color or dress."
Another is tracking in large crowds where everyone is dressed the same, say, a large business conference where everyone is in a dark business suit. "There is no way to differentiate this at present," said Kanga, "even a human would have difficulty doing so."
"There are obvious technological bottlenecks," Doron said. "We are attempting to broaden the reach of video analytics. We want to create cost-effective devices that combine video, hardware, sof tware and analytics, but getting the prices down to acceptable levels is quite challenging."
Educating the Market
Many people are using intelligent sur veillance, Kanga said. "Many have gotten burned because they do not understand the technology. They may, for example, specify cameras that are not appropriate. They may also specify that the system have left-luggage capabilities. Given that hundreds of companies can provide a system that determines abandoned objects in an empty environment, customers may not realize that they have to specify having this ability within crowds. Lack of clearness or an understanding of specifications means that the system will not work well in practice."
Buyers, Frischoff said, are concerned with ease of use, calibration and rule setting. Issues that typically need to be addressed include per formance (detection level versus false alarm rate). "To ensure accuracy and reduce false alarm rates, enterprises need to add dedicated hardware or change their system designs."
"While the competition is excellent in this high-demand market," Troha said, "buyers are understandably cautious as this is still a relatively new technology. What makes us unique is that we enable manufacturers to create valuable solutions through embedding our technology in devices."
"The market," said Doron, "is transitioning from high end to mid range. We are very involved in making that shift happen, and education is impor tant. The backbone of the mid-range segment is security dealers; many are not aware of what video analytics can do for their businesses. It reminds me of the early stages of IP adoption. There is a lot of talk, but not a lot of people are engaged in actually doing it."
Like the other major players interviewed, Agent Vi, said Frischoff, helps hardware partners understand its technology through training, on-site support, improved integration tools and comprehensive testing procedures.