Machine Learning powered object detection support in event notification server

[asker]

I just released 1.5 of the event notification server that lets you insert custom hook scripts before alarms are sent out. I've also provided a near realtime example of how people detection can be achieved for events that are detected by ZM and before the notification gets sent out to you.

Brief description: here
Longer description: here


Sample:

[asker]

I've now added 2 different detectors:
a) the first one is fast and uses HOG. Image was above
b) A second one that uses DNN and YoloV3 and is slower but more accurate. Like so:



And also:

[asker]

I just fixed an important bug last night.
As it turns out if alarms are generated the same time as another monitor dies, the alarm doesn't get caught.

Please make sure you are on version 2.0 and updated on Oct 10, 2018 (US tz) (I made some changes without bumping the version)

[asker]

The detection text is now written into the ZM DB as well and there is a new commit to master today, so your console will look like this:

[nrbell]

Thanks so much for implementing this! I just set this up last night and it works great.

[asker]

Nice! I'm always happy to see users get this working successfully.

[asker]

The next version of zmNinja will allow an option to only view events that have detected objects.

[asker]

The updated scripts now support the following:

- "bestmatch" as a value for "FID" -> This will try to analyze the first alarmed frame for objects and if it fails, it will analyze the frame with the highest score. Both these comparisons will be done in the same run instance, so the model won't have to be loaded twice (faster than manually calling the script twice as model load is slow).

- better installation instructions

https://github.com/pliablepixels/zmeven ... ok_example