Is Video Analytics Ready for the Mainstream?

When video analytics started making an appearance in the market, promises of extraordinary capabilities had industry professional expectations high, but the reality is they simply did not deliver. As with any new platform, there were problems and early adopters still tell war stories about the failure of first-generation systems, even as underlying technologies have enhanced the capabilities of later-generation products. I think we all agree that marketing hype often outpaces real-world success. But fortunately in this case, todays analytics solutions are delivering proven performance at affordable prices.

In its short history, the potential of video analytics has progressed from use in limited, niche applications to its current potential as a mainstream tool within the reach of any video end user. Video analytics capabilities have evolved from being almost exclusively a software add-on to large Video Management Software driven systems to being embedded in products such as cameras and recording devices. This shift in the deployment of video analytics is changing the landscape for systems design.

Smarter cameras at the edge

Video motion detection was a forerunner of video content analysis, or video analytics, and has been around for decades. Video cameras today can detect motion easily and dependably, and, in fact, this feature is usually taken for granted and not considered real video analytics. But video motion detection performs the same primary function as more advanced video analytics, which is to analyze the content of a video image and provide an alarm. The only difference is that the video analytics inside smarter video cameras can effectively analyze a more diverse range of data.

The processing power being built into new network cameras makes more advanced analytics possible, and also plays a key role in the improvement of overall image performance. For example, in-camera processing is used to manage the dynamic range of images by correcting the balance of light and dark areas within a scene, and can also stabilize images to offset wind vibration, among other useful features. Smart cameras can also create privacy zones within images and block the viewing area of specific areas in a scene such as, say, the windows of an apartment complex.

Relative to video analytics, advanced in-camera processing provides these capabilities:

Object left behind. The camera can alarm if an object appears and remains in a scene for longer than a pre-selected period of time.

Object moved. An object in a scene can be identified and specified, and the camera will alarm if the object is moved.

Virtual triplines. A directional line is specified and an alarm triggers in the event something crosses the line. Alternatively, an area could be specified and the alarm would indicate something has entered the area.

Object tracking. As an object or person moves thorough a scene, an alarm can indicate if the direction has changed.

People tracking. By identifying people in a scene, video analytics can help prevent loitering or tailgating (a person follows someone else through a gate or door without presenting a credential).

People counting. The camera can identify the number of individuals or other objects such as cars in a scene and count them automatically.

Scene change detection. In the event a camera is tampered with or its field of view obstructed, intelligent cameras can sense the change and generate an alarm.

Identifying faces. The camera identifies a face and isolates it from the rest of the video frame. These captured faces could be compared to face images stored in a centralized database or even on a digital video recorder to provide face recognition.

It is no longer necessary to specify specialized or expensive cameras to get these capabilities. They are increasingly being included as standard features of the new smarter, higher-resolution network cameras. It is also not difficult for integrators of IP-based systems to implement these capabilities in existing or new systems. The technologies are easy-to-program, robust, proven and do not add appreciably to system costs. Adding video analytics at the camera level also does not require additional server capacity.

Another system advantage of leveraging in-camera video analytics is the ability to filter when video signals actually transmit over a network. Too much video on a network, especially if the network is being shared by other enterprise uses, can be a bandwidth and storage challenge. This is especially true given the migration to megapixel video cameras and larger image sizes. Use of an in-camera analytics alarm to trigger when video needs to go over the network, or what video is recorded, can provide an attractive option to better manage network resources.

Analyzing video at the network edge also ensures that all the information in the video image is available to be analyzed, which is not the case for example with compressed video traveling across a network. Use of video analytics at the camera level is also ultimately scalable. You only have to use as much as you need; for instance, you could start small and add additional smart cameras specifically where you need them over time.

Centralized video analytics

As useful as in-camera analytics can be, it may not be the best approach for every application. By their nature, the analytics inside cameras are limited in terms of programmability and customization. Not every video analytics application can be reduced to whether a line is crossed or whether an object is moved. For more complex or application-specific needs, there are a variety of third-party video analytics systems that are typically integrated as part of a VMS system.

These analytics systems are more flexible and can be programmed for a wider variety of uses and application environments. They can detect a range of security threats in multiple locations, even if the threats are different at each site. They can also be used with more complex scenarios, for example, to incorporate a greater number of user specified parameters when deciding whether to issue an alarm. And as good as the in-camera chips are, software-based systems used with a server have much more processing power. Greater processing expands the capabilities of these systems, although the greater functionality has a price, both a software cost and a need for additional hardware. In terms of upgrading the system, it is much easier to upgrade a centralized software-based system than to manage the upgrade of possibly hundreds of cameras located at the network edge. Maintenance and system management also tend to be simpler with a centralized approach.

Deploying analytics spanned across multiple cameras is also a benefit of centralized analytics. For instance, a centralized analytics system could check for similar or related activities across a group of cameras to analyze system-wide patterns that might prompt an alarm. Software to read license plate numbers is mature and effective, and higher-resolution video images enables the software to do an even better job. Another advantage of a centralized system is the availability of a database to capture and compare multiple face images or license numbers simultaneously in high traffic applications like toll plazas or stadiums.

Important variables in server-based systems include compatibility with cameras and other system components and the importance of image quality. Higher-resolution video cameras provide better images and more information for analytics systems. Systems receiving higher quality data are consequently more effective.

Server based systems have also matured and expanded their functionality. False alarms, a problem with earlier systems, can be better managed using analytics. Many of the systems are user-friendly and intuitive to operate. There is also now a large installed base of video analytics systems, so users have an opportunity to hear from their peers how these systems have improved since the early trials and tribulations.

Cameras never blink

It has been decades since Sandia National Laboratories tested the attentiveness of individuals tasked with watching video monitors for several hours a day. Results of the Sandia test for the U.S. Department of Energy estimated the attention span of anyone watching an inactive video screen to be about 20 minutes. That is a scary number considering how many video screens there are that are presumably being watched.

Video analytics functionality now provides an answer to the attention-span dilemma. They can effectively analyze video data and alert an operator only when there is something of interest to see. Technology will never replace the need for human interaction and response, but video analytics has developed into a useful tool to make security personnel more effective overall.