Skip to main content

Using the new analytics in the Tech Preview

Introduction

As part of our long-term goal of making it easy to bring third-party analytics (including your own!) into EdgeVis, we've integrated a variety of new analytics we've created ourselves or downloaded from open-source repositories.
​
To make it easier to test, we've also added support for analytics to containers! Just use the v9.3 client and configure your analytics as normal - you should find several new analytics to try.

Included analytics

The final choice of analytics for v9.3 is still to be decided - and we'd love your feedback on this! The following analytics have been selected to showcase the capability that we can now integrate several off-the-shelf analytics.

These are the analytics we've included:

  • Safezone 2D
    Our legacy sterile-zone motion detector
    ​

  • Object Detectors
    We have included a variety of Object Detectors - for a full list of objects each can detect, please use the Configure button after selecting the analyics in EdgeVis Client:

    • Object Detector
      Our current shipping object detector that ships on our hardware encoders.
      ​

    • Object Detector (DB-DEF)
      A DB-developed specialist detector that can detect more unusual vehicles (e.g. aircraft and tanks).
      ​

    • Object Detector (RTMDet-S)
      An off-the-shelf open-source object detector, used to prove we can integrate third-party detectors. It can detect many types of everyday objects (e.g. animals, household items, vehicles).
      ​

  • ANPR - Automatic Number Plate (License Plate) Recognition
    This analytic can identify vehicle number/license plates and optionally trigger an alarm based on a watch list of known plates. See the ANPR Configuration Options section below for details on how to utilise.
    ​

  • faceLog6 - Facial Identification
    This analytic can find faces (and heads) within a video stream. It only identifies the presence of a face - it does not perform any match against a watch list.

We will definitely be shipping the capability to use our standard analytics (Safezone 2D and Object Detector) in containers in v9.3!

How to use

If you have correctly installed the Tech Preview container image, then you will be presented with an extended list of analytics when you use EdgeVis Client to configure your analytics:


Most analytics configuration options are self-explanatory (e.g. the object detectors list the objects that it can detect). ANPR is more complex, and is explained in the next section.

ANPR Options

The ANPR analytic performs two operations:

  1. Identify number plates within the video source (effectively as an object detector).

  2. Read the number plate, and (optionally) match it against a known watchlist.

If you only want to find number plates, you can disable step 2 (for a significant CPU saving), but in most scenarios you will use both operations to read the number plate to allow it to be used in alarms/meta-data search.
​
When you select Configure you will be presented with the options for the ANPR analytic:

To only identify number plates, you should:

  • Disable Recognise Licence Plates

To identify and read the number plate, you should:

  • Leave the default settings (as per screenshot above)
    ​The Licence Plate Filter should be * - this means 'match all'

To match against a list of specific number plates

  • Enter a comma-separated list of number plates into Licence Plate Filter

  • The Licence Plate Match Mode setting can either be:

    • exact - all characters must match against an entry in the watchlist

    • fuzzy - all characters should match, but still trigger if there is a one-character difference

There are two additional controls:

  • Global Threshold
    This determines how confident the analytic should be that it has identified a number plate within the image

  • Minimum Recognition Confidence
    After converting the image to a number plate, this determines the minimum confidence the analytic has that it has converted the plate successfully.

Both default settings should eliminate most false positives; however if you are seeing too many false alerts or are missing detections, you can adjust these two values to increase/decrease the confidence levels (higher means be stricter, lower means detect more but may increase false positives).

Use cases

Analytics can be used for two main purposes:

  • EdgeVis alarms
    ​
    You can use the alarm management system on your encoder to send alerts and perform actions when an analytic triggers.
    ​
    However, be aware that to avoid flooding users with too many alerts, the analytics are 'debounced' - this means that you will only receive an alert for the first object/face/number plate detected. There needs to a period of no activity for the system to reset, and a fresh alert sent when a new object is detected.
    ​

  • Supplying metadata to third-party systems such as Milestone
    ​
    Adding your encoder to a third-party VMS/analytics system can allow the generated metadata to be used for further analytic or forensic search. Some example use cases are:

    • Adding your encoder to Milestone should allow you to use their forensic search to find number plates.

    • Using VMS Gateway and enabling Publish ONVIF Profile M output stream should allow a third party system to receive analytic metadata.

(Please note that you can't use the Enhanced video codec with third-party systems or VMS Gateway - these features should be tested separately.
​

Did this answer your question?