Video Analytic Architecture ConsiderationsAuthor: IPVM Team, Published on Oct 08, 2011
Video analytics may be implemented in a number of fundamentally different ways. Many ask: Which one is best? Each has important tradeoffs. Moreover, in our experience, some are far more likely to be accurate as well as simpler to deploy.
Below, we rank and analyze five fundamental approaches (alphabetically):
- Add-On camera analytics
- Analytics appliance / encoder
- Cloud analytics
- Embedded camera analytics
- Embedded DVR / NVR / VMS analytics
- Server based analytics
We also look at recent advancements in video analytics and what may be available in the future.
Let's start with a review of each approach:
Add-On Camera Analytics
What is it? Analytics software is loaded onto an IP camera. Video is analyzed within the camera and events/alarms are sent to VMS systems for display / monitoring.
- Flexibility to pick and choose which cameras one wants to run analytics on.
- Freedom to choose preferred camera manufacturer rather than being locked into a specific camera vendor.
- Only a handful of manufacturers currently support this, most notably Axis. Vicon, and Hikvision Series 4 cameras are also supported by some vendors.
- Analytics are typically not optimized for the specific camera model.
- Must ensure that the analytic output can be integrated with one's VMS
Note: people often talk about bandwidth savings being a benefit. We disagree. While it may save bandwidth, this is generally not a key benefit. The reality is almost all users stream all of their cameras anyway so streaming for cameras using video analytics does not make a big difference.
What companies are examples of this? Agent Vi, Jemez Technology.
What is it? Combines concepts of server based analytics and embedded camera analytics. A dedicated hardware device running embedded software analyzes an image stream from a camera.
- Typically smaller physical footprint and lower power than server-based analytics.
- Easier setup/maintenance due to embedded design.
- Can add embedded-style analytics to cameras that do not have the resources/ability to run on-board analytics.
- Higher cost compared to Add-on or Embedded analytics, due to being dedicated hardware.
- Often have limited VMS integrations.
- Can be more difficult to troubleshoot due to mix of camera vendor and analytics vendor.
What companies are examples of this? Avigilon (Rialto), VCA (Bridge).
What is it? Video is sent to a cloud service provider for analysis.
- Camera can be relatively dumb, just sending a 'regular' video stream out.
- Cloud provides great scalability to process video without having to add capabilities or hardware on-site.
- Cloud providers can learn from lots of video sent to them to improve their analytics dynamically.
- Requires bandwidth to send video upstream to the cloud, which is often limited / expensive, especially if many cameras from a site are being sent
- Can be expensive, requiring service monthly fees
- Limited number of companies are offering this
In the future (looking toward 2020), this is likely to become a more common option, with more providers.
Embedded Camera Analytics
What is it? One manufacturer develops and manufacturers a camera with their own embedded video analytics.
- Performance: Manufacturer can optimize performance to specific camera settings, features and issues.
- Simplicity to setup: No need to install analytic software or new hardware.
- Simplicity to troubleshoot/service: Since a single vendor is behind this, easier to get clear answers about what is wrong.
- Flexible to add cameras to existing VMS systems.
- Limited to a handful of companies that offer this.
- Cannot re-use existing 'regular' cameras.
- A few major manufacturers call their embedded motion detection "analytics" which is misleading and should be avoided.
How to tell motion detection from analytics?
Motion detection-only systems will have very few parameters for rules, generally a region/zone and a sensitivity threshold. More advanced motion detection systems may go as far as implementing a simple tripline-style logic.
True analytics will have behavior-based rules, such as loitering, direction of travel, or object present within region. True analytics will also adjust for anticipated object size within an area of the scene, understanding that a further away object will be smaller than a nearby one, and adjust sensitivity accordingly. Most true analytics systems will have object classes, meaning that rules can be applied only to people, or only to vehicles.
What companies are examples of this? For real video analytics embedded in cameras - Avigilon, Axis, Bosch, FLIR/ioimage/DVTel, Sightlogix.
For motion detection marketed as video analytics - Sony and Samsung among others.
Embedded DVR/VMS Analytics
What is it? Analytics software is pre-loaded on a DVR/NVR appliance or within VMS software. Administrators simply license and/or turn on analytics for their desired channel.
- Simple setup: No servers to setup, no additional software to install and no need to integrate with separate recorder
- Integrated display: Because the analytics are part of the recorder solution, typically the recorder's client software nicely integrates alarms, events and/or searching inside.
- Accuracy issues: Because the analytic software does not know the quality or characteristics of the cameras streaming, it cannot optimize its performance for camera specific issues.
- Using with IP cameras: Heavy processing overhead if used with H.264 streams which significantly limited number of total cameras used.
- Performance constraints: Especially if run on DVR/NVR appliances, the analytics may have significant restrictions in total performance or cameras supported.
- Restricts VMS choice: Typically, users pick VMSes or recorders for numerous reasons and usually have a platform in place. Difficult to switch just to get analytics.
What companies are examples of this? Aimetis, 3VR.
Server Based Analytics
What is it? Analytics software is loaded on a PC/server that is separate from the recorder/VMS being used. Video is streamed in to the server and events/alarms are displayed on the software's client or sent back to a VMS or PSIM client.
- Server side analytics do not depend on the resources available inside cameras and recorders (which is often limited).
- Developers can easily add this to an existing system regardless of the cameras or recorders support of analytics.
- Requires adding another piece of hardware.
- Increases network bandwidth - the server pulls an additional stream from the camera
- Costly: Because this requires a new vendor and new hardware is added, this tends to be the most costly approach.
- Accuracy issues: Because the analytic software does not know the quality or characteristics of the cameras streaming, it cannot optimize performance for camera specific issues.
- Complexity: Because a new server needs to be set and integration with a VMS or PSIM needs to be added, this can be quite time consuming to set up.
- Potentially Limited Support: Displaying alerts/information and meta-data tagged video may not be available for all VMS platforms
What companies are examples of this? Giant Gray, Iomniscient, IPS, PureTech
Rating Video Analytic Architectures
Our tests have shown embedded camera analytics to offer the most reliable overall performance, with analytics appliances offering similar performance when used with quality cameras.
We rank video analytics architectures in the following order, keeping in mind this is a rule of thumb and not absolute:
- Embedded Analytics - Simpler setup, lower cost, generally strong performance but beware camera VMD pretending to be analytics.
- Analytics Appliances - More flexibility of cameras, but higher costs.
- Add-On Analytics - Generally strong performance, but overall limited choices.
- DVR/VMS Analytics - Simple setup, directly integrated with VMS. Often lower performance.
- Server-Based Analytics - Greatest design flexibility. Likely to be highest cost, complex setup and can present integration risks with VMS.
Note: cloud was omitted from this ranking, as it is too early and limited to generally say how good or bad it is.
While this ranking shows general trends, given the state of video analytics today, the top priority for most applications is reliable performance. One cannot take for granted that video analytics works. A prudent user would be better off assuming that most do not work well enough for production use.
When evaluating systems, users (or manufacturers) will often focus primarily on maximum range. While range is important, total coverage area should also be evaluated (eg: an analytic camera that achieves 500' range, but with only an 80' wide horizontal field of view may be less useful than an analytic camera with a 300' range, but a 200' wide field of view). Users should also consider maximum practical achievable range in their environment, in many cases it is difficult to achieve >500' range consistently if the area is prone to rain, snow, fog, dust storms, or other weather.
Additionally, false alarm rejection should be considered. If achieving a 500' detection range comes at the cost of a high number of false alarms, the system may end up as unreliable overall.
Recent Changes And Advancements
Video analytics for security applications have seen relatively little advancement over the last several years.
Many systems still process video at less-than-megapixel resolution, and the few systems that handle higher resolution images max out around 2MP (note: while some analytics cameras and software work on 4K or higher resolutions, they simply down sample the image, this is apparent by the fact that stated coverage area does not increase above a given resolution).
Object detection classes remain relatively limited, systems often classify all humans as "person" and have relatively limited vehicle classes ("car/small vehicle", "truck/large vehicle"). Other kinds of object classes, such as "boat", "bicycle", "airplane", "animal", "forklift", "bus", etc. are rare.
Advancements have primarily come in the form of ease of deployment and configuration, with companies like AgentVi and DvTel/IOI promoting auto-calibration or easier setup procedures.
Camera hardware has advanced, with cameras having much-more onboard processing power than before, which will make embedded or add-on analytics even more prevalent.
Over the next 5 years, we expect availability and reliability of video analytics for security applications to improve significantly. For example, projects like Google Advanced Image Search are advancing image processing and object-tagging algorithms. Self-driving vehicle developments are also providing renewed interest in real-time video processing and object identification. And, last but not least, lower cost cameras with more on-board processing power will make video analytics more viable.
Most Recent Industry Reports
The world's leading video surveillance information source, IPVM provides the best reporting, testing and training for 10,000+ members globally. Dedicated to independent and objective information, we uniquely refuse any and all advertisements, sponsorship and consulting from manufacturers.