|Video analytics allow users to automate the use of video and extract more value
Almost anyone in the video analytics market will admit that the capabilities of the technology were initially oversold. The first generation of analytics simply didn’t work as promised, and an undercurrent of distrust of the technology in general still haunts the market. Ironically, video analytics capabilities have now matured into a robust and dependable option for a variety of applications. Suppliers are eager to get the word out.
“Since video analytics were introduced to the security market, we have learned enough about customer needs and the technology to offer fully robust solutions that can provide significant value to the end customer,” says Maor Mishkin, director, Video Analytics Product Champion, DVTEL. DVTEL’s focus today is on the analytics that provide the most value such as people counting, loitering and directional, he adds.
Processing power has increased to allow analytics to be pushed out to the edge, increasing use cases and reliability, adds Mishkin.
DVTEL’s ioimage video analytics line is a comprehensive portfolio for outdoor perimeter protection. DVTEL’s HD cameras feature built-in analytics, enabling both edge-based and server-based flexibility. Serving a range of vertical markets, ioimage cameras help to increase the probability of detection and reduce the false alarm rate. Site View, a web-based remote live viewing and playback feature, enables operators to respond quickly in real-time and also investigate incidents after they happen. Ioimage IP analytics are available over H.264 video for the company’s HD and thermal cameras, or as a server-based video analytics solution (SVA) through DVTEL’s Latitude NVMS.
DVTEL’s video analytics portfolio includes server-based and edge-based analytics to maximise the advantages of either approach, depending on the application. The smart cameras work independently or as a complement to perimeter intrusion detection sensors and other technologies.
Since 2005, ioimage’s team has fine-tuned the advanced algorithm-based technology, which has increased market acceptance. The company has invested significantly in R&D and expanded system reliability and flexibility.
Driving value for the user
Generally speaking, video analytics allow users to automate the use of video and extract more value. Without applying analytics, surveillance systems tend to create large amounts of video that isn’t doing anyone any good. It is captured and archived using terabytes of expensive storage, and then it’s deleted after a time, having provided no value. Agent Vi has identified three types of applications for video analytics that can drive value for the user:
-- Real-time event detection, allowing the user to transform the video system into a proactive detection system rather than a passive viewing system.
-- Forensic investigations, cutting down on the costs of investigations by making it easier to search video based on video analytics and other parameters.
-- Business analytics, providing data that would otherwise be impossible to collect, such as people-counting, and in-store behaviour analysis in the retail environment. The only alternative would be to position a person at a store entrance or throughout the store to observe activity and count customers with a “clicker.”
“These days, most of the customers are comfortable that this is actually going to work,” says Zvika Ashani, chief technology officer (CTO), Agent Video Intelligence (Agent Vi). “There’s still an education process. People have heard of video analytics, but they don’t really understand or know what the analytics are. It’s not what they saw on TV. Once their expectations are well defined, they can get a lot of benefit out of it.”
|DVTEL’s focus is on the analytics that provide the most value such as people counting, loitering and directional
Meeting diverse operational needs
Verint offers multiple analytics for refining video into “actionable intelligence.” For situations that may be simple or complex, Verint Video Analytics offer a range of capabilities for security, surveillance and business applications. Security analytics are focused on opportunities to identify target individuals or vehicles, based on unique identifiers or simply discerning which video has people or vehicles within the frame of view. Surveillance analytics help in identifying crowding, loitering, or objects left behind. Business analytics seek to support pattern identification for retail merchandising and checkout excellence. Most customers are interested in some mix of these analytics to fit their diverse environments and operational needs.
“The first wave of analytics met the harsh realities of customer environments and eroded confidence in all areas of the security advisory community, most notably among security consultants,” says Joshua Phillips, director of product marketing, video and situation intelligence solutions, Verint Systems Inc. Accuracy, processing load, and application guidelines have improved greatly since analytics first burst on the scene over 10 years ago, and it’s time for evolution to take its course, he adds.
Video analytics can help a customer move toward a system that is easier or less expensive to operate, says Phillips. A video analytic detecting an anomaly can be programmed to set in motion a security response that has been pre-defined, such as a guard or officer dispatch to the specific area. Without the analytic, the security operations centre is relying on the operator to have the appropriate camera view open at the right time, visually discern the breach, and be able to take action to initiate the response. If the problem is clearly understood, and the analytic rules are applied, the customer in this case gets the result they want – expedited response. This simple example may require an additional means of verifying the accuracy of the analytic detection, says Phillips. The customer may have other sensors available through other systems, or cameras, which if triggered could be paired with the video analytic to create an enhanced alarm, he says.
Analytics in a complex world
iOmniscient specialises in video and other analytics in practical context situations – such as a crowded scene. For example, it’s easy to find a bag left behind in an empty room, but much more complex to identify that bag if there are 1,000 people in the scene. In face recognition applications, iOmniscient only requires 22 pixels between the eyes to identify a person (while some competitors require up to 300 pixels). Therefore, using iOmniscient, a standard camera can recognise people 50 meters away in an uncontrolled environment, says Dr. Rustom Kanga, CEO of iOmniscient.
"Most customers don’t know
what to ask for; they get things
that don’t work. Where we have
well-educated customers, we
are the most successful. We
spend time educating them
about the technology in depth"
“It’s a complex world, and we specialise in complex environments,” says Kanga. Many customers try to implement inexpensive systems that are “simple and trivial” to address situations that are complex, which is why many systems fail, he adds.
Kanga points to the Boston bombing incident as an example of a situation where iOmniscient’s system would have been helpful. The system would have been able to identify the bag left behind (containing the explosives) amid the crowded scene. It also could have employed face recognition to identify the person who left the bomb, and it could have sent information in real-time to a nearby first responder.
An automated response capability is a new development of the iOmniscient system. It locates the nearest police car or other first responder and sends an alert. The capability can reduce response time on street accidents from 25 minutes to under 5 minutes, Kanga says. The feature is also useful for applications when there is no central control room; if there are five security officers, the system can identify and notify the officer nearest the scene.
Kanga attributes early problems among video analytics companies to the short attention spans of venture capitalists looking to make a quick profit on the emerging technology. “It took 10 years to develop the technology, so they didn’t have time to wait for it,” Kanga comments. “We raised our own money and built it. The technologies have taken a long time to build, but they are robust and are implemented in many places.”
Video analytics keeps improving, based on improving algorithms, says Kanga. One improvement in the iOmniscient system is better people counting accuracy – now 99 percent versus the previous 95 percent.
A requirement for successful implementation of video analytics is that customers ask clearly for what they want, says Kanga. If a requirement is to “find a bag,” a lower-cost system might be proposed. However, if the proposal is to “find the bag in a crowded scene if the view of the bag is obscured 50 percent of the time,” lower-cost systems would immediately be ruled out. “Customers are learning to ask for what they actually require,” says Kanga. “Most customers don’t know what to ask for; they get things that don’t work. Where we have well-educated customers, we are the most successful. We spend time educating them about the technology in depth. They become good customers. When implemented properly, the systems give you good results.”
iOmniscient also combines video analytics with sound analysis and smell analysis. Sound analysis might include gun shots or people shouting. An example of smell analysis is a recent project in Asia, where the customer wanted to detect the threat of an electrical fire before it starts. The smell sensor alerts to the scent of plastic getting hot. (It’s actually a third-party sensor that enables analysis of a “chemical signature;” iOmniscient has adapted it.)
An example of the benefit of multiple types of analysis can be seen in the case of a person falling down. A video analytics system can alert on the incident, and typically a security person would hurry to the scene to help the fallen person. However, if a sound analysis also indicates a gunshot at the same time the person falls down, the response would be very different.