Exploring IBM Video Analytics (Smart Surveillance System)By: John Honovich, Published on Jul 06, 2008
IBM offers a unique approach to video analytics. I highly recommend understanding it and its trade offs. While I think most users, outside of the most demanding, will prefer less expensive and simpler solutions, IBM's approach reveals some of the key challenges that video analytic solutions shall be grappling with over the next decade.
At first glance, IBM's Smart Surveillance Solution might appear similar to other commercial offerings. Their video analytics solution provides the fundamentals you might expect. You can select from a number of different type of analytics, you can get alerts based on those analytics and you can search for matches based on the information the analytics generate.
I see two critical differences in IBM's approach that need to be understood and appreciated:
- IBM is building a complete video analytic overlay to video management systems.
- IBM is aiming to solve the most challenging and demanding video analytic problems.
The standard approach in monitoring and using video analytics is to use the same management system and user interface for analytics as you do for your video surveillance recording. This is common both for vendors who perform video analytics at the edge (ObjectVideo and ioimage [link no longer available] are two prominent examples) and for vendors who perform it at the server (for instance, i3DVR [link no longer available] and 3VR).
Managing and monitoring video analytics from a singular management system provides a number of benefits. One, it keeps costs low as you reuse your existing video management system. Second, it simplifies user operation as the operator can accomplish all relevant tasks inside of a single application.
The most important limitation of using a single system to manage video analytics and video recording is that it constrains optimizing your solution for video analytics. Indeed, most video recording solutions were not built to handle analytics and have bolted on simple to modest support for video analytics.
By contrast, IBM has built a separate management system for video analytics. The Smart Surveillance System is dedicated to manage video analytics and provides an optimized user interface for that purpose. The IBM system does not handle watching live video or other common video tasks. A video management system (such asGenetec or Milestone) is used for that purpose. When the IBM System's analytics find an event of interest, IBM requests video from the video management system for display.
This allows IBM to focus on optimizing video analytic management. The architecture they use is the same one used by dozens of the largest corporate web applications in the world. This is ideal for searching and analyzing millions of data points from huge amounts of cameras. Such ability is in stark contrast to today's leading IP video surveillance system which are designed for a different purpose - that of handling hundreds or thousands of video streams.
IBM's architecture provides unmatched flexibility and scalability to handle complex combinations of analytics across very large sets of cameras.
Demanding Video Analytic Problems
IBM's commitment to demanding and challenging video analytics seems to be stronger than any other provider in the marketplace. Part of this is due to the roots of IBM's video analytics solution. This product grew out of a long term research project within IBM called PeopleVision [link no longer available]. This project experimented with numerous and esoteric analytics. As such, they have a deep pool of code and expertise to draw on for demanding video analytics. IBM's architecture reinforces and supports the use of demanding analytics. Because it is designed to deal with a variety of analytics from a variety of cameras in a flexible and scalable manner, it is easier for IBM to pursue such demanding analytics.
However, this differentiation is not simply because IBM is smarter. Many of the leading and most well known video analytic providers in the market have explicitly decided to limit the complexity and sophistication of analytics provided. Take ObjectVideo for example. ObjectVideo, by their own acknowledgment, is focused on making analytics easier to use, more widely available and cheaper to deploy. This pattern repeats itself with all the big video management vendors who are more concerned with making analytics work with their existing products than exploring uncharted territory.
For these reasons, I expect IBM to maintain a leadership position in testing, using and deploying novel analytics.
Reviewing the Competitive Landscape
To determine if IBM is the correct choice for your project, you need to evaluate what the alternatives and risks are. I see 3 major factors that need to be considered:
- Simpler but less expensive mainstream video analytic solutions
- Risks of complex IBM video analytics working poorly
- Potential for basic mainstream video analytic solutions to improve
Simpler But Less Expensive Alternatives
Today, integrating video analytics and video management systems has become fairly common. It is not perfect nor is it limitless but almost all the leading DVR/NVR vendors have some form of third party support. Many video analytics providers have optimized their solutions to work in industry leading cameras. Some of the DVR/NVR providers have actually built analytics inside of their appliances. In either event, the cost and process to deliver a video analytic system is fairly low.
By contrast, IBM's solution requires a number of complexities:
- Currently IBM's analytics require a dedicated appliance. You cannot run them on cameras today nor can you do so inside the DVR/NVR. Adding such appliance are generally far more expensive than the alternatives. If you have numerous sites (like a retailer), you could be adding in hundreds of these appliances. While IBM does not share prices publicly, this could easily add a few thousand dollars in cost per store/location. Even when IBM adds support for cameras, they may have issues performing the really sophisticated analytics they want to do inside a camera.
- The IBM video analytic system requires the design, configuration and implementation of an enterprise management server. This will inevitably require expensive software licenses and on-site consultants. Compared to mainstream video analytics, solutions this will easily add $100k plus in costs. Now, for very large projects, this cost may be trivial but it is a barrier to entry to start a new system.
- IBM's advanced search and analysis tools are not integrated into an NVR/DVR. It is likely that this will continue because of the customization and sophistication of the analytics that IBM wants to offer. The big practical factor is that it leaves users with two user interfaces which can increase complexity in the use and operation of systems. Users frequently prefer and often require a singular user interface or common operating picture.
While IBM offers a lot of potential beyond mainstream video analytics, genuine costs and trade offs do exist. You will need to determine that IBM's added abilities are more than enough to offset the cost and complexity that the solution adds.
Risks of Complex Analytics Not Working
Because numerous alternatives exist for video analytics and IBM's architecture is more expensive, IBM must differentiate itself by offering more powerful analytics that cannot be matched by competitors. I am not sure ultimately what those solutions are. IBM's publicly available information is shallow but from demos and industry articles, it appears solutions will including searching for similar license plates, objects, people, etc over large sets of cameras.
The challenge here is: How Well Can They Make it Work?
This is not a slight against IBM, this is a recognition of the complexity of the problem video analytics represent in today's world. Making video analytics work well in a variety of environmental conditions for a single camera is hard enough. Making it work well across numerous cameras and for more subtle characteristics is very challenging.
When you perform video analytics across large camera sets and large time frames, small errors can be enormously magnified. Often people like to talk about finding the little girl lost in the mall (see a video demo by another vendor of this). Because of differences in angles, lighting and the inherent imperfection of the analytic itself, such searches can return vast number of false matches and miss any number of correct matches.
I clearly do not know how well IBM can make their analytics work. I would not be surprised if IBM's engineers are still trying to figure this out themselves, not because they do not know what they are doing but because some of these aspects are research issues that have never been fully deployed in production. As such, customers should be very careful to understand the risks involved.
Potential for Mainstream Analytic Systems to Improve
The final factor that should be considered is that mainstream video analytic systems can move upwards over time. While there are certainly limitations and constrains in how well these systems can get, as analytics improve, they will certainly look to employ those improvements. This will continuously put pressure on IBM to push the limit of analytics to establish differentiation.
Unless you have extremely pressing security demands, you likely can upgrade and enhance your mainstream video management system over time. While I certainly think you will be sacrificing some benefits that IBM can offer, given the cost and complexity differential, this may be good enough for now.
IBM's offering is truly innovative and aims at solving crucial security problems at an unmatched level of sophistication. For this, IBM deserves respect and recognition. Nevertheless, I think customers will find it difficult to choose IBM as simpler, cheaper solutions will be attractive on the low end and the risk of analytic performance will constrain on the high end. For very large deployments with complex security needs, IBM should be seriously considered. In all cases, caution should be used in evaluating this choice.