San Francisco recently released one of the most detailed reports on municipal surveillance:

http://www.sfgate.com/cgi-bin/article.cgi?f=/c/a/2009/01/11/MNLU156OAF.DTL

It points out serious weaknesses in most systems that are easily correctable. Video analytics provides the answer to many of the problems they experienced.

Perhaps the most significant statement from the report is:

“We find no evidence of an impact of (the cameras) on violent crime,” the report stated. “Violent incidents do not decline in areas near the cameras relative to areas further away (and) we observe no decline in violent crimes occurring in public places.” (Quoted from the San Francisco Chronicle article above.)

This was not true for other crimes. For example, larceny targeting people, vehicles and homes dropped 24 percent near cameras. The report surmises that violent crimes did not see the same reduction because they are “often committed outside the bounds of rationality.” In other words, knowing that cameras are present does not deter these acts of passion.

But this misses the real problem. The following comment from the Chronicle points to the biggest weakness of the San Francisco system:

In addition, the cameras have not had a “feedback cycle” – criminals have not seen immediate consequences to lawbreaking under the cameras’ gaze.

San Francisco has a particular restriction that you won’t see in many cities: They are not allowed to monitor live video. They can only retrieve recorded video after a specific incident. This prevents any ability to stop crime through response. Thus, video becomes nothing more than a deterrent, and is used for gathering evidence after the offense.

This is like having police officers standing by watching an event unfold, but not being allowed to help until after the murder or theft takes place. The purpose of this restriction is to protect privacy. This reveals one of the biggest issues that will need to be addressed in the coming years, and I’ll explore in a separate blog post:

Most crime happens under the cover of privacy. Therefore, there is an inherent conflict between privacy concerns and the ability to reduce crime.

The next revealing part of the San Francisco study jumps out of this quote from the Chronicle:

“Officers and others note that despite poor image quality, (the camera) footage has been useful in criminal investigations … more often footage is helpful in establishing a sequence of events for a crime or placing witnesses at a scene,” researchers wrote.

Poor image quality. This was one of the big criticisms. But there were other serious shortfalls:

  • No dedicated manager for the program
  • No training of the officers for how to find or view the recorded video
  • A lack of storage space produced video that was lacking clarity
  • Recording only retained one week’s worth of video, when a full month was needed

The cameras used in San Francisco recorded only 3-4 frames per second due to storage limitations. They paid $700,000 to install high resolution cameras, but the video was so choppy that they often could not read license plates and missed important details of the crime. And with only a week’s worth of storage, there was often no video to retrieve when they needed it.

Their cost estimate to increase the storage for faster frame rates and one month’s worth of recording: $3M, which they can’t afford at this time. That’s over four times the cost of the cameras installed.

This sums up the problems with most video surveillance systems:

  1. They only record: No active monitoring and no immediate response
  2. They typically store video at CIF resolution (352 X 240 pixels) and 3-5 frames per second
  3. They often don’t retain storage long enough

Why hamstring the system this way? Especially after you pay good money to install high quality cameras?

The reasons are simple. It is too expensive to actively monitor video using security officers to watch monitors. And the cost for bandwidth and storage for full motion video at full D1 resolution (704 X 480 pixels) for a month is too high. This is why most systems are compromised.

Video analytics offers an answer to all of these problems.

It is easy for video analytics to watch all the cameras all of the time. They can automatically alert a security officer if a breach is occurring, or warn when a crowd is developing and a potential problem might be brewing. With audio through the cameras, you can respond immediately to prevent crimes before they happen.

The report raised this very observation, saying that it would help to use “smart cameras that are capable of sounding an alarm if a gun is brandished, a fence is jumped, or a person falls down.” (It isn’t easy for analytics systems to detect guns, but fence jumping and trip-and-fall detection are common.)

As for the cost of storage, the problem becomes obvious. The cost of storing 5 frames per second (fps) at CIF resolution for a month, using MPEG-4 video, costs an end-user about $100 per camera. Raising this to 15 fps at D1 resolution increases the end user’s cost and bandwidth by 9X. If you jump from 1 week’s worth of storage to a month, you’ve increased it by 4X again.

With analytics, you can significantly reduce these storage costs by using what we at VideoIQ call Content Aware Storage. That simply means that the analytics can see when something important is happening and the system then records high quality, full motion video. When nothing is happening it records at traditional low quality, low frame rates.

Video analytics is not just for detection and pro-active response. It can make the whole video system smarter and more effective. And it solves the biggest issues that are compromising the effectiveness of surveillance systems today.