Jon's Top 5: video analytics

09 Sep 09

1: Choose the right product

There are a number of manufacturers offering video content analysis (VCA) systems in this rapidly evolving market. Most offer systems targeted at specific situations (intrusion detection, traffic surveillance, smoke detection or object tracking).

It can pay to use a number of manufacturers for different applications. This is, however, a relatively new science and often systems may not live up to their hype. Look for systems that do not try to offer too much or systems that offer individual bolt on algorithms.

Look for manufacturers who have experience with a particular type of analysis and can provide references, case studies and government or industry accreditation, e.g. i-LIDS from the Home Office.

2: Edge or Server?

The term ‘edge device’ has been borrowed from the IP networking world, and is now a common term in security. Many IP cameras now offer integral recording and alarm functionality, including video analytics, propagating the term 'edge analytics'.

There are significant benefits in network efficiency where images can be locally recorded for short periods of time and streamed to a central server on alarm only, saving network bandwidth.

The alternative is server based analytics, often used in conjunction with legacy analogue systems to avoid the cost of camera/infrastructure replacement.

This can be used for IP based systems, particularly when there is a need to analyse video stored on the NVR. Be aware of the processing power needed to analyse video streams.

3: Configuration

Ensuring the best performance from any analytics system will depend heavily on accurate configuration. An incorrectly configured VCA system can ignore the things you want to detect and raise false alarms for scenarios of no interest.

Camera placement can have a dramatic effect on how well a system functions. Nuisance alarms are more likely to occur where there is significant movement in the scene, so it's important to conduct detailed tests including a walk test to ensure that factors like sensitivity and object size are correctly configured.

If this is not possible, then problem areas may need to be masked out at configuration. Try to have a trial run with the software you intend to use to ensure that it's not too difficult or time consuming to operate.

4: Site trials

With so many factors governing which product will suit each application best, it's imperative to run site trials.

For external applications, environmental factors such as changing light conditions, wind (camera shake), rain and fog can sometimes affect the performance of certain systems.

Ensure that sudden scene changes such as a lights-off/lights-on transition don't generate false alarms. Check that the system can learn the new scene (following a sudden scene change) quickly, as any longer than 20 seconds or so might result in missed detections.

5: User interface

Ensure that the VCA system chosen provides the interfaces you need. This could be anything from a simple contact closure to full metadata output stream.

Consider how this will integrate into any legacy systems or be part of a new installation. VCA is a difficult thing for most NVR/DVR manufacturers to integrate, so make sure any integrated systems will meet your clients' expectations in terms of coherence between the system building blocks.

Newer devices are starting to conform to Onvif or PSIA standards. If your VCA system and NVR/DVR speak the same language your integration costs could be drastically reduced.

Get the latest stories first with info4security's newsletters: Click to signup

Post and bookmark this story at the following sites:

What is this?

del.icio.us	Digg	Reddit	Google	Newsvine