The Evolution of Video Analytics: Past Failures to Accurate Crime Preventing Tool

by Gerald Narciso
Jun 12, 2014

analytics 1

A decade ago, the expectations of video analytics in the surveillance industry were all over the map. Some people thought that it would solve all their monitoring problems, while others believed it could analyze scenes the same way that a human could. The “CSI effect” that derived from sci-fi TV shows and movies only amplified the expectations and exaggerated the capabilities of video analytics.

Dr. Mahesh Saptharishi has witnessed first-hand the highs and lows of the video analytics industry. In 2007, he co-founded VideoIQ, an integrated video analytics company that focused on analytics for cameras and edge appliances. Prior to that, he started Broad Reach Security which specialized in server-based analytics. He has spent much of the past decade developing technology that would not only correct the shortcomings of previous video analytics solutions like false alarms and ineffective algorithms, but would also position video analytics as a useful and reliable tool for the surveillance industry.

Now with Avigilon – following the company’s acquisition of VideoIQ in January 2014 – Dr. Saptharishi is responsible for integrating innovative analytics technology solutions into Avigilon’s product line. Recently, Avigilon released the Rialto Analytics Appliance, which provides real-time event detection and verification, enabling users to proactively identify any unusual activity that occurs in a monitored scene and take immediate action as necessary. The technology brings accuracy and credibility to a video analytics industry which has historically failed to provide a consistent and dependable solution.

Dr. Saptharishi recently talked to Connected about the current state of the video analytics industry, current trends, and the future.

What were the expectations of analytic software that most developers weren't able to deliver on in the past?

Dr. Saptharishi: Well actually, the major one was accuracy. People expected that the system would provide almost zero false alarms, but at the same time not miss any real event. And the fact of the matter was that analytics at the time produced a fair number of false positives. It wasn't something where you could connect an analytic system to auto-dial police officers to come in and respond to an event.  And the second mismatch was the type of useful information that analytics was perceived to provide. People wanted to be able to use analytics for everything from security applications to business intelligence to retail applications. The problem was the quality of the analytics wasn't high enough and the detail with which it could pick metadata out from the video wasn't rich enough for them to do the types of things that they wanted.

How does the video analytics technology you developed at VideoIQ, and is now used at Avigilon, provide value for the end user?

Dr. Saptharishi: We have a very focused goal with analytics in the security space - to take video surveillance from being strictly a forensic tool and turning it into more of a preventative tool. A lot of end-users only view their video footage for after-the-fact evidence or investigation. We have now created a solution that is more of a preventative tool that can be paired with real-time monitoring. For example, if the analytics software catches something out of the ordinary in a live video scene, like a person or vehicle, it can alert a guard remotely or at a monitoring station that somebody's actually intruding on the property. It gives them the capability to intervene and to send law enforcement to that scene and prevent a bad outcome from happening in the first place. At a high level this is what our technology is all about.

What kind of trends have you seen in video analytics over the last couple of years?

Dr. Saptharishi: Video analytics is becoming a mass market tool. Installing the software used to be a fairly difficult, cumbersome and expensive setup and maintenance process. As a result, analytics was something that was limited to only high-end enterprise users or users in the critical infrastructure space. One of the things that we wanted to do was to build video analytics technology in such a way that an installer could install it just like they would a normal surveillance camera or an NVR. Now, it is as easy as plug and play, and you get the level of performance that you would expect right out of the box, with no continued maintenance issues. What this did was it brought analytics from the realm of high-end applications to more commercial applications.

We are now seeing a lot of commercial customers who cannot afford to have their own security operations center starting to rely on third-party central monitoring stations. Historically, the third-party companies that monitored burglar alarms had no experience actually monitoring video as part of their services. Our analytics are accurate enough and cost-effective enough for these central monitoring stations to now monitor effectively.

Another trend we are seeing is analytics working at a higher level with HD video cameras. In the past, the industry as a whole focused on standard resolution. Even if you had a 29 MP camera, the analytics would still operate at standard resolution. This is a significant problem that I see on the whole in the industry because when an installer installs a 5 MP camera, they expect a certain coverage field of view. And when they map analytics through it, all of a sudden you don't get the benefit of that higher resolution because you're still down sampling that video to a much lower resolution. Our capability to now operate at a high-definition level for analytics really complements the video quality that Avigilon HD cameras provide, whether it’s a 5 MP camera or a 29 MP camera. This is a trend that I think is going to start heating up pretty soon because we see a lot of customers being very disappointed at the fact that you've got all of these analytics vendors out there who claim to be able to do all kind of things with video, but at the same time can't take advantage of the high-resolution that the cameras are now providing at a fairly reasonable cost.

Facial Recognition is a hot topic in video analytics. How far away are we from an effective and accurate facial recognition solution?

Dr. Saptharishi: I think facial recognition technology has come a long way. Given the right lighting conditions and camera resolution, facial recognition can be fairly accurate with a cooperative or semi-cooperative subject. As the resolution of surveillance cameras is constantly improving, there will be less noise and more detail in the image making it easier to derive information. Sensor technology is also contributing to acquiring richer information about a person. It has a video image, but it also can judge depth. It can judge your own body motion and things like that. How people walk, how people move actually are very unique to each individual, so you could combine facial recognition information with their body movements to uniquely identify them. And when you have two mutually exclusive pieces of validation that uniquely identifies you, the accuracy now goes up without the restriction of having a cooperative subject.

The other thing that body movement and facial geometry also tells you is roughly the age and gender of the person. I think in five years, you will see algorithm technology improve to get you that biometric identification that you previously have not had. Another quick example that I'll give you is iris recognition-- identifying a person based upon their iris signature required a person to be truly a cooperative subject. We can track the face of the person over time and get a good enough resolution on the person's iris so now you can not only use facial recognition, but you can also use iris recognition as another biometric that really could identify the individual.

It sounds like the video analytics industry is moving forward in baby steps. Is the industry being more cautious not to make the same mistakes they made ten years ago by overpromising on the capabilities of video analytics technology?

Dr. Saptharishi: Absolutely, I think that this is coming from both the industry side and the customer side as well. The industry is being very careful about how they manage customer expectations, but customers are also very skeptical now. They want more proof that it actually is going to do its job. They're going to want to test it, and that's why organizations like the NIST (National Institute for Standards and Technology) are actively evaluating technologies and providing consumer report style performance reviews of the accuracy of these various technologies. On the other hand, managers are being a lot more cautious. They don't want to overpromise like they did 10 years ago.

Stay Connected. Sign up and receive exclusive industry articles and other value-added content from Avigilon.

We welcome the re-use, republication, and distribution of "Connected" content. Please credit us with the following information: Used with the permission of   

Category: Innovation


For Media Relations

Please email or call 604-629-5182


ACC Version Last version of ACC tested with camera. This also implies support for later versions of ACC unless specifically listed otherwise.
Audio Input Receive audio feed from camera.
Audio Output Send audio to speaker attached to camera.
Autodiscovery Automatic discovery of camera IP address when connected within a LAN environment.
Compression Type Describes the encoding types supported for the camera.
Connection Type Describes the type of Device Driver used. Native refers to the Manufacturer's specific device driver.
Dewarping In-Client dewarping of fisheye or panoramic cameras.
Digital Input Receive Digital or Relay inputs from camera.
Digital output Trigger digital or relay outputs physically connected to a camera.
Motion Quick display of whether Motion Recording is available on for the camera.
Motion Configuration Configuration of motion detection within the ACC Client.
Motion Recording Support for motion-based recording.
PTZ Quick display of whether PTZ functionality is available for camera.
PTZ Control Basic PTZ Movement.
PTZ Patterns/Tours Ability to create and trigger either PTZ Patterns, or PTZ Tours, depending on camera support.
PTZ Presets Create and trigger PTZ Preset positions.
Unit Type Type of camera.
Verified By Organization which tested camera and reported capabilities.
Verified Firmware Specific firmware version tested.
Manufacturer Blah
Model DS-2DE2103
Connection Type ONVIF
Unit Type IP PTZ camera
Compression Types H.264

  • ACC Version
  • Model DS-2DE2103
  • Connection Type ONVIF
  • Manufacturer Blah
  • Manufacturer Blah
  • Manufacturer Blah
  • Manufacturer Blah
  • Manufacturer Blah
  • Manufacturer Blah
  • Manufacturer Blah
  • Manufacturer Blah
  • Manufacturer Blah

Verified by:

Download Test Report