Can VMSes Allow Cameras To Be Used For Both Surveillance And Video Analytics?

It would seem feasible to me that a single camera could be used for both general surveillence and video analytics purposes. Are VMS vendors offering anything along these lines?

A couple of possible examples:

1. A camera is set-up with analytics to detect abandoned objects, but also provides the 'normal' CCTV Video stream for live viewing and recording. (if camera is PTZ - perhaps analytics are suspended when the camera is under operator control)

2. A camera is scheduled to count people in its coverage area at 15 minute intervals (i.e. a snapshot), in addition to providing a 'normal' CCTV Video stream for live viewing and recording.

I appreciate there are a number of other factors to consider, such as camera location, lighting etc which may make it challenging to optimize a camera to peform both these roles?

Apologies if this is a 'novice' question for the IPVM forum.


I believe the answer to your question is yes. Many of the analytics applications I've seen are based on views that are or can also be used for live or recorded monitoring by an operator.

I wouldn't underestimate, however, the camera positioning, lighting, etc., that you mention. In many cases the accuracy and applicability of analytics increases dramatically when the camera is positioned explicitly to support that role--at the possible expense of its applicability to general surveillance. In some (but not all) cases you end up chasing your tail when trying to achieve both objectives with the same camera.

Yes, here are some limitations:

  • As Steve mentions, sometimes video analytics need very precise / constrained positioning, which make it unsuitable for general surveillance (example - people counting is typically done with an overhead shot, which is of little value for general surveillance).
  • You will sometimes need 2 streams, if the analytics are being done in a different location than the recording. There is some potential for bandwidth increase (especially if the analytic system wants a MJPEG high-quality stream) and there is an increase in processing load on the camera. Today, most cameras should be able to handle 2 simultaneous streams but there is a chance of dropped / reduce frame rate.
  • Be very careful about doing analytics on PTZs. Make sure that the PTZ 'knows' to stop analyzing video when the camera is moved or else it will generate lots of false alerts. Also, many analytics need time / input to calibrate / optimize their analytics, making moving cameras a big problem.
  • Abandoned objects is risky in general, however you do it. You need to define a time range to declare an object abandoned. If you do it too short (let's say 10 seconds), you will have lots of false alerts, if you do it too long (say 5 minutes), your building might already be blown up. Also, depending on how busy a scene is (say a busy train station), it be might hard for the camera to really tell if an objected is abandoned.

Finally, don't trust analytic vendors. Make them prove it over time (weeks) and in your busiest areas.

I worked on a project with over 850 cameras and relatively easy analytics requirements on every camera. Some were people counting it line que length, some loitering, some directional alarms for stairways. Cameras were placed for surveillance purposes and introducing analytics after the fact would have been disastrous. The site was redesigned using the anticipated analytics for each camera inind and both surveillance and analytics requirements were met.....they ran short on funds and only installed the cameras! Oh well.