zoneminder-notifier is a daemon designed to send notifications quickly after they occur by email (including SMS or MMS gateways) or Pushover. It also can do object detection based on pre-trained open source neural network models to minimize false alerts. It works by polling active monitors for new events via the ZoneMinder API. Object detection uses OpenCV. Email notifications are sent with the Mutt email client, while Pushover notifications can also be sent by email to your account's email gateway address or via the Pushover API. If object detection is enabled, the image attachment will include boxes and confidence levels around detected objects.
- ZoneMinder with API version 2.0 enabled (tested with ZoneMinder 1.36.32)
- Python 3 (tested with version 3.10.6)
- Python modules: setuptools, requests, cv2, numpy, matplotlib
- Mutt (if you wish to send notifications via email)
- Pushover account and api token (if you wish to send notifications via Pushover API)
An install script is provided. It is intended to be run as root or with sudo. It will install Python modules in a system Python site-packges directory, the config file in /etc, executables in /usr/bin, and object detection model data in /usr/share/zm-notifier. Most of this data is not tracked in the git repository due to size but is provided in the release tarballs.
The install script also automatically installs, enables, and starts the systemd zm_notifier service. If your distro does not use systemd, you should comment out those lines in the install script and consult your distro's documentation for how to run the daemon automatically at startup.
-
Edit /etc/zm_notifier.cfg. Inputs are explained in the comments of this file. Please read them all carefully and fill out accordingly. Note that you need a section with the name of each of your ZoneMinder monitors if you want zoneminder-notifier to do object detection.
-
Please be aware that ZoneMinder credentials are stored in plain text in the configuration file, so you should be careful with it. Using the provided install script, this file will be readable only by root.
-
If needed, configure Mutt and make sure that you can send emails to the intended addresses, using the user account that will also run zoneminder-notifier (root / sudo -i if you use the provided systemd service).
-
After installing, check the logs for any errors and edit the config file if needed. Using the provided systemd service, log messages will be written to /var/log/zm_notifier.log.
-
If your distro does not have systemd, you can redirect the standard output and standard error to the log file in this way:
/usr/bin/zm_notifier > /var/log/zm_notifier.log 2>&1 &
-
It is also possible to run zm_notifier from the command line and view the output directly. If you want to do this for testing, just stop the system service first.
zoneminder-notifier comes with support built-in for four different object detection models: Darknet (Yolo V4), SSD Mobilenet V3 (2020_01_14 version), SSD Inception V2 (2017_11_17 version), and Histograms of Oriented Gradients (people only detection model included with OpenCV). The data for these models are included in the releases of zoneminder-notifier. The README in the model_data directory includes the urls from where they were downloaded.
All except Histograms of Oriented Gradients (HOG) are trained on the COCO datasets and are suitable for detecting a wide variety of object classes (HOG detects people only). For each monitor, the classes you want it to actually pay attention to should be specified in a comma-separated list. It is possible to do object detection on either the max score frame from each new ZoneMinder event or the entire video. The former is much faster and generally recommended. Darknet is the most reliable model but also the slowest, but it offers an excellent combination of speed and accuracy when detecting in the max score frame only.
If you want to do object detection on entire event videos, you need to first configure ZoneMinder to save videos under monitor settings. Please test carefully to make sure your system is not getting overloaded analyzing entire videos when many events are happening. To save computations, zoneminder-notifier will stop analyzing a video as soon as it has made a successful detection, but this doesn't help for events where none of the requested object classes are found.
The images below represent the result of Darknet object detection with the classes "person, chair, sofa, bicycle" from some of my ZoneMinder events.
