Top 3 Problems Limiting the Use and Growth of Video Analytics

While video analytics holds great promise, people are still asking about the viability of using analytics in the real world. Indeed, as stories of video analytic problems have spread, concerns about the risks of video analytics now seem higher than a few years ago when the novelty of the technology spurred wide excitement.

This article surveys the main problems limiting the use and growth of video analytics. It is meant to help security managers and integrators gain a better sense of the core issues involved.

Top 3 Problems:

1. Eliminating False Alerts
2. System Maintenance Too Difficult
3. Cost of System Too High

Eliminating False Alerts

Since the goal of video analytics is to eliminate human involvement, eliminating false alerts is necessary to accomplish this. Each false alerts not only requires a human assessment, it increases emotional and organizational frustration with the system.

Most are familiar with burglar alarm false alarms and the frustration these causes. On average, burglar alarm false alarm per house or business are fairly rare. If you have 1 or 2 per month, that is fairly high. Many people do not experience false alarms of their burglar system for months.

By contrast, many video analytic systems can generate dozens of false alarms per day. This creates a far greater issue than anything one is accustomed to with burglar alarms. Plus, with such alarms happening many times throughout the day, it can become an operational burden.

Now, not all video analytics systems generate lots of false alarms but many do. These issues have been the number one issue limitation of the integrators and end-users that I know using and trying video analytics.

System Maintenance Too Difficult

System maintenance is a often overlooked and somewhat hidden issue in video analytics.

Over a period of weeks or months, a video analytic system's false alerts can start rising considerably due to changes in the environment, weather and the position of the sun. This can suddenly and surprisingly cause major problems with the system.

Not only is the increase in false alerts a problem, the risk now that the system could unexpectedly break in the future creates a significant problem in trust. If your perimeter surveillance one day stops functioning properly, you now have a serious flaw in your overall security plan.

This has been a cause of a number of video analytic system failures. The systems, already purchased, simply get put to the side becoming a very expensive testament to not buying or referring one's colleagues to video analytics.

This being said, not all video analytic systems exhibit this behavior but you would be prudent to carefully check references to verify that existing systems have been operating for a long period of time without any major degradation.

Cost of System Too High

While you can find inexpensive video analytic systems today, these system tend to exhibit problems 1 and 2, high false alerts and poor system maintenance. Indeed, in my experience, video analytic systems that are either free or only cost $100-$200 more generally have significant operational problems.

One common feature of systems that work is that the complete price for hardware and software is usually $500 or more per channel for the analytics. Now just because a video analytic systems is expensive obviously does not mean it is good. However, there are necessary costs in building a systems that is robust and works well in the real world.

The cost of video analytic systems comes in making them robust to real world conditions that we all take for granted. The developer needs to make the video analytic system “intelligent” enough to handle differences in lighting, depth, position of the sun, weather, etc. Doing this involves building more complex or sophisticated programs. Such programs almost always require significantly more computing hardware to execute and significant more capital investment in writing, testing and optimizing the program. All of these clearly increase costs.

The challenge is that it is basically impossible to see this from marketing demonstrations because from a demo all systems invariably look exactly alike. This of course has the vicious effect of encouraging people to choose cheaper systems that are more likely to generate high false alerts and be unmaintainable.

If you select a system that works, the cost per camera can make it difficult to justify the expense. Indeed, so much of the first generation video analytic deployments, came from government grant money, essentially making the cost secondary or not relevant. Nevertheless, for video analytics to grow in the private sector, they will not only need to work they will need to generate financial return.

When video analytics allow for guard reduction or reduce high value frequent losses, it is easy to justify and you see companies having success here (in terms of publicly documented cases, IoImage is the leader here). For other cases, where humans are not being eliminated, the individual loss is small or the occurrence of loss is low, the cost can be a major barrier.

Conclusion

Though I anticipate video analytics successes to increase, I believe such success will be constrained to applications where the loss characteristics and/or the human reduction costs are high. While analytics will certainly become cheaper, such cost decreases will take time and in the interim, it is these high value applications where analytics can gain a foothold of success.