There is much skepticism about the working of video analytics. However, a number of enterprises understand the limitations of analytics and have benefited from carefully deploying them to meet their security objectives. These benefits include timely responses to security breaches, efficient use of security personnel and prevention of security incidents. The alternative to analytics is a scenario including security directors monitoring live video streams from multiple cameras- which is proven to be ineffective and inefficient. There is an ongoing debate concerning the best architecture to deploy analytics - at the edge (camera), at the server or distributed across the edge and server. The simplest answer would be to choose an architecture that best meets an enterprises priority needs and objectives. It is interesting to note that a similar debate on where storage must be located-- at the edge, server or distributed across the edge and server is beginning to generate interest.

Key Factors that Impact Architecture

Security objectives, practical aspects, reliability and costs impact the architecture that businesses select. These aspects influence the computing, storage and bandwidth-requirement which, in turn, impacts the overall architecture.  The security objectives also determine the type of video analytics to be installed.

 

 Typical security objectives include:

• Quick response to events as they happen and identification of objects of interest.  For example, sending a security officer when someone trespasses and recognition of the number plate of a car that has been parked in a no parking area.
• Providing evidence only for what happened.  For example, proving that a person stole a particular asset. A sterile zone can be marked around the asset of interest and recording can be triggered once anybody enters the sterile zone.
• Proving that something did not happen.  For example, proving that somebody did not slip and fall in the establishment’s premises.  This means that you have to record video on a continuous basis.  In such a scenario, analytics will be used to provide necessary alerts and generate metadata to quickly isolate the video of interest.
• Providing evidence that is acceptable in a court of law.  This is determined by the resolution and frame rate of the recorded video. 
• Need to view a single video feed or multiple video feeds simultaneously on different monitors.  Security directors will require access to recorded video from a control and command center.

 Practical aspects include:

• The need to protect investments in deployed solutions.  For example, if an establishment already has invested in analog cameras and DVR, it is more appropriate to run the analytics on the server.  Alternate approaches of replacing the analog camera and DVRs with smart IP cameras is more expensive.
• Time taken to detect a security breach.  In cases where a security breach can be easily identified, the need to store recorded video for more than 30 days before overwriting it is low. 
• The nature of the asset that needs to be monitored or protected.  For example, a bank may place multiple cameras in each of its branches/ATMs across different locations in a city.
• Difficulty in running and maintaining a cable/wireline network.  For example, when surveillance needs to be provided to a large campus with many outdoor cameras, it is easier to have the cameras on a wireless network.  The wireless network bandwidth available is much lower than the bandwidth available on the LAN and hence impacts the architecture.

While all security systems must operate at very high levels of reliability, the amount of downtime that can be tolerated will impact the architecture decision as well:

• How much of downtime in the server or network be tolerated?
• Can the business live with loss of stored videos due to hard disk failures?
• Can the business tolerate video loss while video is being recorded?

Budgetary constraints toward meeting the security objectives of a business are a crucial element in determining the architecture:

• Costs for one time installation and commissioning
• Costs for maintaining the systems on an ongoing basis

Typical Architectures for Deployment

Once the security objectives, reliability aspects and budget limitations are well understood an informed decision on the architecture can be made.

 

The major architectures for deployment of analytics and storage are:

• Running analytics completely on the server and storing of videos on the server
• Running analytics completely on the edge and storing of videos on the server
• Running analytics completely on the edge and storing of videos on the edge
• Distribute running of analytics and storage of videos across the edge and the server

 The pros and cons of each architecture :

Architecture	Pros	Cons
Analytics & Storage on the Server	·      Sophisticated analytics can be installed ·      Flexibility offered in terms of reassigning video analytics to a different camera ·      Situation awareness is improved ·      Easy to upgrade analytics or increase storage capacity at any time ·      A number of companies provide analytics on the server side.  Hence, end customers have greater choice on the analytics that can be used ·      Most video management software vendors provide easy access to the stored/recorded video on the server	·         Bandwidth requirements are high ·         Videos from multiple cameras is lost in case of server/network outage ·         Additional processing power is required to decode the video streams at the server before analytics is performed ·         Bandwidth, storage and processing costs are high ·         Recurring cost of power and cooling for server/storage is significant
Analytics on the Edge & Storage on the Server	·      Cost of analytics is bundled into the price of the edge device ·      Bandwidth requirements will be lower, especially if continuous recording is not required ·      Analytics can be run on the raw data, resulting in higher accuracy	·         Analytics limited by the processing power available on the edge device ·         Videos from multiple cameras is lost in case of server/network outage ·         Recurring cost of power and cooling for server and storage
Analytics & Storage at the Edge	·      Cost of analytics and storage is bundled into the price of the edge device ·      Recurring costs are much lower as there are no power and cooling requirements ·      Analytics can be run on the raw data, resulting in higher accuracy ·      Bandwidth requirements is low Minimal impact in case of network/storage outage	·         Analytics limited by the processing power available on the edge device ·         Recorded videos limited by the capacity of storage on the edge device ·         Very few vendors provide solutions with analytics and storage on the edge devices ·         Support for access to stored/recorded video on the edge device through the video management software is poor
Analytics & Storage Distributed Across Edge & Server	·         Analytics can be run on the raw data, resulting in higher accuracy ·         Optimum usage of resources ·         Moderate bandwidth requirements	·         Installation is complex ·         Exposed to risks related to – network and server outages ·         Limited choice of vendors who are offering such solutions

For cases where continuous recording is necessary and the video needs to be retained for a long period of time (more than 3 months), storage at the server is the most appropriate option.  For small camera count installations where recorded video needs to be retained for less than 30 days, analytics and storage on the edge device may be a preferred architecture.

 

The Future

Few key trends that will impact the architectures in future are:

• Processing power on the edge devices continues to grow at a rapid rate.  A favorable $/MIPS ratio for the processors on the edge device when compared to processors on the server, will cause for a push of analytics towards the edge.
• Camera resolutions are constantly increasing with the advent of high definition (HD) cameras.  If existing encoding technologies are retained (H.264/MJPEG) then the bandwidth requirements to transmit video streams to the server will increase manifold.  Storage on the camera/edge devices is one way to reduce the bandwidth requirements.
• Capacity of flash memory on edge devices is steadily increasing.  The $/byte ratio for standalone storage devices is significantly cheaper when compared to storage at the server.  These reasons will cause a push for storage to the edge devices.
• While current support for access to recorded video on the edge device through the video management software is poor, a few video management vendors are beginning to support such a feature.

While the key trends may point to a gradual movement to edge-based analytics and storage, security objectives will drive end users to an appropriate architecture.  Security directors will have to analyze the trade-offs involved and choose an architecture that best meet their requirements.