By some counts, there are 38 companies pitching video analytics technology or solutions for security applications. Many of these companies use return on investment (ROI) as one of the big selling points. The idea is to let software determine what video is of interest so that security guards can be more productive, reduced or even eliminated.
I sought to find evidence of these claims: can investment in video analytics really provide a financial payback? What is the ROI and does this stuff really work?
There has been a great deal of hype and promise around video analytics since the introduction of DVRs. Manufacturers have promised analytics as means of differentiating their products. Start-up companies like ObjectVideo, Vidient and many others were formed around the idea. The idea of applying face recognition to spot terrorists passing through airports and public places received much publicity in the weeks following the Sept. 11 attacks, only to fizzle as the hype seemed to exceed the reality. Hollywood fostered the myth with its depictions of almost magical video analytics in movies like Ocean's Eleven and on the NBC TV show Las Vegas .
I found three specific video analytic applications that have a proven ROI and payback on the investment. In each of these successful cases, the video analytics technology was one component of an overall solution to a specific problem.
What is Video Analytics?
Video analytics (or intelligent video) are computer algorithms that process digital video to identify specific objects and object movement. The technology has roots in computer vision applications that have been deployed in manufacturing environments to identify production defects. Analytics software for security applications is commonly deployed on commercial servers that accept video feeds that are digitized and then processed to separate objects of interest from the image background. Some manufactures now embed analytics directly into cameras (Sony, IOImage) and video servers (Verint).
Regardless of the physical implementation, the analytics must be integrated with event management software to generate alerts, video pop-ups and to trigger video recording. The analytics typically require a high degree of configuration using software tools. The system must be carefully programmed to identify objects of a specific size, motion characteristic and type (for example, human vs. vehicle). “Trip-wire” applications will generate an alert when the specified object crosses a defined line or threshold in the camera image.
Analytics for Perimeter Security
Several car dealers in New York are using a remote video monitoring service in place of on-site security guards for a cost savings of $4,500 per guard. With the installation of a 24/7 remote video monitoring service from Visentry, dealers have virtually eliminated theft, increased their sales and improved the visibility of their business operations.
Car dealers have problems with theft and vandalism. According to Sean Timmons, general manager of DCH Millburn Audi in New Jersey , he could not leave any inventory in an unfenced area for more than three days without the theft of Xeon headlights from the automobiles – a hot commodity. Prior to the installation of Visentry, DCH hired on-site security and would physically move their car inventory into fenced areas each evening.
With Visentry's service, DCH has been able eliminate the on-site security service and leave inventory in an unfenced area with access to the street. According to Timmons, giving customers the ability browse the car lot after closing hours is essential to good business. Since installing the Visentry system and allowing customers to browse inventory after hours, Timmons says that he has experienced record sales without any theft. Timmons noted that the police love the system and are quicker to respond to alarms with video verification that can describe the suspect, his automobile and perhaps the direction of his escape.
Visentry's solution uses video analytics from VideoIQ to identify people and cars that cross configured perimeter boundaries. Analog cameras feed an on-site VideoIQ server. Standard DSL lines transport video and real-time alerts from the dealer site to Visentry's central monitoring station where monitoring personal are attentive to the alarms and the associated video. When Visentry personnel are alerted to suspicious behavior, they warn the suspect with an audio announcement that is delivered over the DSL connection. Local police are notified if the warnings are ignored.
Nearby Paragon Honda in Queens , N.Y. , eliminated two of its security guards for a monthly savings of $9,000. Paragon pays $800-$1,000 per month for the surveillance service at each site. The installed cost of the overall system, including cameras, analytics and video recorder was about $45,000 per site. In addition to the security benefit, the Paragon ownership uses Web-based access to the system on a daily basis to remotely view activity at each of its four locations.
According to Yona Wieder, Visentry CEO, the company could not practically provide its service without the use of video analytics to identify perimeter violations. The analytics allow Visentry's monitoring personnel to effectively view a large number of cameras across multiple sites.
Prior to installing a VideoIQ system, one of Wieder's customers used a fence sensor and another used photoelectric motion detectors. In each case, the previous alarm monitoring company “quit” because they could not deal with the false alarms. “ When we installed our system at the dealership we tried to use part of the fence (sensor) in conjunction with our cameras,” Wieder explains, “but we stopped using the fence sensors because wind was causing too many alarms .”
Adds Doug Marman, CTO of VideoIQ: “[The system] knows how to separate out changes in the background. It learns the scene and knows to ignore water fountains, moving escalators, and all forms of weather, such as rain, snow and lightning. [The system] recognizes all of those as changes in the background, so that it can focus on detecting people, vehicles or boats.”
Analytics for “Tail Gate” Detection
San Francisco International Airport (SFO) has deployed Vidient's SmartCatch system to replace more than 15 security guards. Like all U.S. airports, SFO prevents unauthorized access to secure areas containing baggage and aircraft. A common problem around these areas is the practice of “tail-gating” or “piggy-backing” -- when an authorized person gains access to an area with a badge and is then followed by a second person who does not swipe a card before the door closes. The situation usually involves lazy employees who are not following the established access control procedures. This practice is not only an obvious security threat, but the TSA imposes hefty fines on the airports for violations.
The Vidient analytics alert a central monitoring station when two or more people enter a secure area with a single access authentication. The analytics identify the number of people passing through the area, and the access control system indicates how many authentications occur. When violations occur, the central station is provided with video pop-ups and audible alarms.
The airport has eliminated more than 15 contracted security guards from these secure doors, according to Michael McCarron, director of community affairs for SFO. The staff reduction results in an annual savings of more than $1.2 million.
McCarron emphasized that the analytics are one component of the airport's evolving security system. The Vidient system has improved over time, as the analytics are being perfected -- resulting in fewer false alarms.
Analytics for Parking Lot Fraud
The City of Tampa 's Parking Division has increased revenues by more than 143 percent in a three-year period using the AutoVu system from Genetec.
Previously, parking division officers identified delinquent parking ticket holders and bootable vehicles through manual inspection of license plates covering more than 16,000 city parking spaces in garages, lots and on-street parking. With the installation of the system, a parking division vehicle is equipped with cameras and mobile PC running the AutoVu Mobile License Plate Recognition Software (LPR) software. As the vehicle is driven through parking facilities and streets, the system detects and reads the license plates of vehicles that are parked in parallel, at 45- and 90-degree angles. Once the plates are identified, the system checks for a match within an onboard database of delinquent motorists that is updated by the Tampa Parking Division's central server. When a match is detected, the system generates an alarm notifying the officer that a bootable vehicle has been detected.
Keys to Success
The three successful examples have several common factors:
* Solution-oriented: In each case, the analytics are one component of an overall system to solve a particular problem. The analytics are embedded with alarm/event management systems, remote video monitoring, recording and database systems to provide an overall solution.
* Camera Placement: Effective camera selection and placement is critical to provide the appropriate optics to reduce false positives and improve reliability.
* Configuration and customization: In all of the examples, the manufacturer was involved in tuning the configuration and algorithms to improve reliability. Make sure you have a capable integrator and the support of the product's manufacturer to properly configure and tune the system.
Tom Galvin is a network video specialist for NetVideo Consulting ( www.netvideoconsulting.com ) . NetVideo Consulting provides product evaluations, training programs and software tools to enable successful networked video solutions. Galvin recently published a competitive study of open-architecture video management software products.