Video analytics

	Neugent’s LX Power 9000 series of DVRs are embedded with intelligent video solutions.

Experts dig deeper into this emerging technology, stressing what it can and can’t do, and where best to put it.

Today, video analytics comes in various names and forms. The most commonly used terms include intelligent video surveillance (IVS) systems, intelligent video solutions, video content analysis (VCA) and video motion detection (VMD). Other less popular terms are mere mixes-and-matches of the words “intelligent”, “video” and “analysis”. Products and solutions in the market provide different levels of security and include functions not previously offered by surveillance and monitoring systems.

Video analytics has become a buzzword in security industries worldwide. However, adoption remains slow, as potential users perceive the technology as complicated, prohibitively priced and not as effective as touted by its proponents. To date, important questions remain unanswered: What makes up a video analytics system? Is video analytics really intelligent? Is it expensive? Is it necessary for my business? How will it benefit me? What solutions are available today?

Experts from video analytics companies in the Philippines and Taiwan believe that computer vision can help video analytic solutions become truly intelligent. However, promising as it may seem, computer vision has yet to achieve the level of intelligence and accuracy critical to video analytics systems.

Another critical consideration is the choice of the right solution. Most of the time, the most practical video analytics solution is one that is somewhere between pure analog and pure IP.

	Neugent’s car counting module is a type of object detection function that can be applied to road traffic management and parking lot management.

According to Jose Antonio Lacsam, chief technology officer of Neugent Technologies Inc., one of the earlier programs labeled as video analytics enabled computers to watch TV and identify advertisements aired on specific TV shows and channels. “For example, I’m a fastfood chain in the US, say, I’m Jack-in-the-Box, and I want to know how much McDonald’s is spending on noontime shows or primetime shows. Without video analytics, it would take a long time to get that kind of information,” Lacsam related.

“I think the main difference between video analytics before and video analytics today is computing power,” Lacsam continued. He said video analytics installations are now smaller in scale compared with the systems of yesteryears, because older systems had to compensate for the relatively slow speed of processors. Thus, buyers of previous systems had to invest in a larger network of computers and implement a business model in which revenues could cover the cost of the video analytics installation.

“Back then, when we were doing that project on monitoring ads, we were using a Pentium III 500, which was the fastest processor available at that time. It was able to handle a few channels per machine. With the likes of Pentium Core Duo today, processing is 12x more powerful,” Lacsam said. In terms of the economics of the solution, before, one had to spend a lot of money on hardware to solve a particular problem. Today, however, a user only needs to spend a small amount to be able to enjoy the benefits of video analytics.

Since video analytics is a DVR solution, its capabilities hinge on the development of the DVR itself. Dr. Vladimir Mariano, Neugent’s consultant for IVS, explained that older DVRs were designed to detect a limited number of events, such as motion events, forced recording events and sensor events. A motion detection event is triggered by motion within the specified surveillance zone. A sensor event is triggered by an alert from a sensor attached to the DVR such as a smoke detector. In the case of a forced recording event, the trigger is manual, such as a security guard noticing something suspicious in the video and pushing the DVR’s recording button.

Each event is color-coded in the DVR’s user interface to indicate the nature of the events tagged in the DVR and which events are high-priority.

When video analytics was introduced, the functions of the DVR expanded. “With video analytics, we’ve introduced another event, which is the object detection event,” said Mariano, who holds a doctorate in Image Analysis from the University of Pennsylvania. According to him, while motion detection is asimple form of video analytics, the latest video analytics software can detect objects and people, track them in special ways, and trigger an alarm if certain preset parameters are achieved.

Object detection is the process of detecting a new object or a missing object. The object can be a thing or a person. For example, when the system detects a bag left in an airport terminal for a considerable period of time, it records the video and tags it as an object detection event. The system alerts the user or the security personnel on where there might be a possible theft or security threat.

In relation to object detection, video analytics enables object counting, which includes people counting, car counting and counterflow detection.

Unlike in previous setups, where security personnel are holed up in a monitoring room to watch video feeds from several cameras, thus running the risk of missing security threats in certain videos, video analytics saves users’ time and effort, and provides more targeted surveillance.

The question is, can video analytics really provide the high level of security users expect?

	Counterflow detection helps spot early signs of security threat.

According to Mariano, the past 10 years have seen comprehensive research in the area of computer vision. “Computer vision is like the eyes and brain. You have a camera (that can see) and your computer (that does the thinking),” he stated. “Even if you have a faster machine, you still need better or more accurate algorithms or programs to, for instance, count people or detect if a person is there.”

Such research efforts have helped computer programmers and engineers discover more sophisticated applications and more accurate algorithms to support today’s video analytics solutions. The results of research projects in the past decade are, in fact, most evident in video arcades.

However, computer vision has yet to achieve the level of intelligence critical to video analytics systems. “I think one of the limitations of currently available video analytics systems is their accuracy level. Most applications are not 100 percent accurate because computers still lack the visual intelligence that humans possess,” noted Michelle Tecson, sales and marketing executive associate at Neugent.

In people counting, for example, computers currently count blobs—blobs of people coming in, whether two people entering together side by side or one large person with a shopping cart. Both will be counted as one. When a person is standing directly behind another person sitting down, with his back on the sitting person, the computer will recognize them as one instead of two.

“The science is not there yet,” Mariano declared. He said that there have been attempts to pursue advanced research related to visual intelligence, but so far, nobody has achieved a significant breakthrough. “It is still a hot area in computer science, especially now that we have fast processors, cheap cameras and cheaper bandwidth.”

In addition, video analytics can detect motion and count people, but it cannot determine conclusively the intent of the person, if he is merely looking around or has the intention to steal. “Computing intent is easy for us humans to do, even from a distance, but not for computers. So if a vendor or supplier promises you that this DVR can determine whether someone is about to steal something, do not believe,” Mariano warned.

Tecson also cited other relative limitations of some video analytics solutions, such as compatibility and lack of integration. Neugent’s LX Power DVRs have built-in video analytics to address the integration issue.

	Bosch’s AutoDome modular camera system is built around a system of intelligent and interchangeable modules, allowing users to easily update dome camera functionality.

Other applications are in retail and market research. Retail stores use video analytics to identify hotspots within a store. The term “hotspot” was coined by weather channels that show typhoon areas in color codes. In the retail arena, hotspots are places within the store that are most frequented by customers. By accumulating data on customers’ movements within the store over a period of time, the storeowner can create a color-coded map identifying hotspots. Based on the identified hotspots, the storeowner can plan the future layout of the store or the positioning of merchandise to boost visibility of items and increase profits.

Video analytics in market research and advertising is a slightly more sophisticated system, but the benefits are real and tangible. Mariano observed that in some applications, face recognition technology is involved. “It’s a fairly new DVR application. Some companies in the US, such as Brickstream and Shoppertrak, are installing DVR solutions to analyze customer behavior in such places as banks, retail stores, grocery stores or gasoline stations,” he said.

Mariano cited one project his former company had with McDonald’s in Philadelphia in the US, which involved counting cars going through drive-thru lanes, and detecting and classifying the gender of drivers based on their facial characteristics. They discovered that the DVR solution was 70 percent accurate, which was not bad for a machine. If companies and market research agencies have this kind of information accumulated over time, then they have a very valuable tool for profiling customers, and adapting sales pitches, marketing campaigns and product promos to customer profiles.

Advertising agencies have also introduced some video analytic solutions that measure the impact of an ad to a passerby or bystander. It detects a face (if it’s male or female) as well as the face’s movement, whether it is looking at the ad on the wall and for how long, or looking away from it.

Video analytic solutions are widely used in Europe, North America and Asia, specifically in the UK, Germany, France, the US, Canada and Japan. They are installed in business establishments, houses, schools, entertainment clubs and law enforcement agencies. The biggest users are the government, manufacturing, banking and finance, transportation and education sectors.

Robert Bosch Taiwan Co. Ltd’s (Bosch) intelligent platform for security applications is widely used in international airports, nuclear power plants and transportation. David Chen, Bosch’s business unit manager, said that territory safety and coast defense are two of the potential markets in the near future.

GE Security’s wide range of commercial and enterprise solutions for access control and video surveillance provides integrated security systems for entire campuses.

For some industries, video analytics applications go beyond security. GE Security’s wired and wireless residential security systems not only guard occupants from intruders but also enable early detection of smoke, carbon monoxide and other hazardous elements. These systems allow remote monitoring and tracking through wireless devices such as mobile phones.

Territory safety has always been the most important defense against attacks. With the recent improvements in technology, analytic solutions are also covering biological threat detection. Security providers are now working on detecting invisible assassins like bacteria, viruses and toxins, according to Susy Liang, GE Security’s director of sales and marketing for Northeast Asia and Taiwan.

For high-risk facilities such as nuclear power plants, major airports, military installations and government buildings, GE Security provides advanced systems for detecting explosives and narcotics. The human and object detection software offered by the company can improve the effectiveness of video surveillance technologies.

GE Security’s airport access control systems feature extensive application integration, combining several functions such as access control, intrusion zone control, elevator control and enhanced I/O control. The system also enables T/A data collection and reporting, mode/event/function scheduling, alarm/activity and device status monitoring, and full system reporting.

The improved algorithm in Bosch’s third-generation system, Intelligent Video Analysis (IVA) 3.0, adapts to difficult conditions such as changes in environment, including rain, snow, clouds and leaves blowing in the wind. The background learning algorithm is continuously developed in-house. The device is able to detect multiple line crossing—from single line up to three lines—combined in a logical row.

The IVA technology is being developed to support more specific functions, such as crime prevention and face recognition of wanted criminals, said Chen.

Developers like Bosch have embedded Intelligent Video Motion Detection (IVMD) to front-end devices such as IP cameras and video servers. Bosch’s Chen said that all their IP cameras are equipped with IVMD. According to Taiwan makers, they have been developing IVS software for both back-end and front-end devices, given the fact that DVRs’ false motion detection rate is quite high. At present, about 80 percent of makers install IVS software in back-end devices for monetary reasons. But installing video analytics at the back-end necessitates related peripherals to support large amounts of network bandwidth and digital storage. Thus, Bosch believes front-end analytics has greater value and time savings in the long-term.

Bosch burns the IVMD software into the DSP and embeds it on video servers or IP cameras. It also offers a licensed option for the IVMD software. End users can choose IP cameras with or without IVMD function and upgrade to the latest version without dismounting them.

Bosch’s second-generation IVMD system, model IVMD 2.0, is a video motion detector for indoor or outdoor use. Its configurable filters and advanced tracking function help improve reliability and reduce workload. The model is capable of detecting moving objects while suppressing unwanted alarms from spurious sources in the image. Functions built into IVMD 2.0 cameras and encoders can detect idle and removed objects, as well as loitering people.

On the other hand, Neugent’s Lacsam believes the right solution still depends on the user’s requirements and investment capabilities. On-the-edge or front-end solutions are equally expensive. He said that the primary drivers or criteria for determining a suitable solution are cost (especially the type and cost of the camera) and network bandwidth.

“IP cameras usually require large bandwidth. If the user does not have an existing infrastructure for their network, then they may have to invest more money on setting this up,” Lacsam noted. If the distance between the office and the target location of the camera is far, network cables will take up a huge chunk of the investment in the video analytics system. He said that last year, the cost of networks—both wired and wireless—was still very high. But if network cost goes down in the future, front-end installations may prove viable.

“There will always be economies of scale. If you centralize the solution into one box, then you benefit from economies of scale,” Lacsam continued. He explained that an on-the-edge solution where you have a single camera for every surveillance zone and a dedicated processor is more expensive than a back-end solution where one processor handles inputs from 16 cameras. But the speedy processing of such analytics-embedded cameras will save time and cost, and require less investment in peripheral devices.

Lacsam also reiterated that it is still unwise, if not impossible, to go all-IP in video analytics because of the cost involved. Up to a certain level, users can install purely IP systems. However, he doubts whether all-IP solutions will be cost-efficient now or in the near future.

“In video analytics, you may start with an analog solution and then another company will propose an IP solution. Most of the time, you will find that the practical solution is a solution somewhere between. I’m not sure what the experts mean by ‘hybrid’ solutions, but if they’re referring to the combination of pure analog and pure digital IP solutions, then I agree with the hybrid approach,” Lacsam stated.

Until now, Taiwan-based companies prefer analog CCTV systems than IP camera systems. Because of cost concerns, they are not keen on changing their entire system. Bosch’s Chen suggests two basic solutions. One is to change the CCTV cameras to IP cameras, and the other is to add video servers for the systems. He urges users to shift to IP cameras with embedded IVS. Bosch’s IVS software is designed with functions/features that support end users’ missions and are also compatible with end user’s surveillance systems.

Lacsam, however, believes that if cost were not an issue, and current network and communication infrastructures were ready, video analytics on-the-edge solutions would make more sense from a network infrastructure standpoint.

Despite the current hype on video analytics, end users still have much to learn about the technology and its benefits. Neugent’s global accounts manager Flory Gonzales said that it is often difficult to reconcile a particular DVR solution being requested with the actual needs of the customer. She revealed that when customers stipulate the things they want, it turns out that they in fact need a different kind of product. They require very sophisticated solutions for very simple problems that can be solved by other types of security systems. “I guess one of our goals as a video analytics solution provider is to educate our end users on the limitations of the system and the real-life situations where you can use a DVR solution. At this point, I think they have some ideas of what video analytics is but do not really know how to use it or how to maximize its benefits,” Lacsam said.

GE Security is using a pro-active approach in delivering customer-focused solutions. The company helps customers build their business and increase profitability. It conducts an in-house Voice of Customer survey and offers its GE toolkit. In so doing, the company educates end users on the benefits of the technology.

Lacsam added that it might take at least a couple more years for the market to catch up. “I’m hoping that by next year, we will start to see a more mature market in which technology and people’s expectations meet and match,” he said.