NVR for self-hosting: Frigate now recognizes license plates and faces

Contents

The recently released version 0.16 of the open-source NVR Frigate comes with numerous new functions. These include AI-supported recognition of faces and car license plates, as well as internationalization of the user interface and more options for controlling cameras.

Face recognition included

The new update, which was released on August 16 after a three-month beta phase, has a lot to offer. In particular, Frigate users previously had to retrofit facial recognition using other tools, but now it is on board. For each person whose face Frigate is to identify automatically, the administrator must store training data: a few high-resolution portrait photos and the person's name are sufficient for the first recognitions to work. Initially, however, only when visitors look frontally into the camera – and after a while also in half-profile. The developers have published detailed step-by-step instructions on how to set up facial recognition.

Recognize license plates with Regexp

Another useful feature: Frigate can now not only mark license plates of vehicles but also read them and compare them with those stored in the configuration. If the NVR administrator enters their license plate number in full, for example, Frigate sends a corresponding notification. Home automation platforms such as Home Assistant can respond to this and switch on the outside light, for example.

Practical in everyday life: the free NVR can also handle regular expressions. It could recognize the DHL delivery van using the regular expression"^BN-BP [0-9]{4}$", and the homeowner eagerly awaiting the overdue hardware shipment can then instruct the parcel carrier remotely via intercom. But be careful with too much automation: license plates are by no means forgery-proof, and anyone who automatically opens the garage door for the supposed family car could be in for a nasty surprise.

If you want, you can also use Frigate as a tool for observing nature. Using a specialized AI model called “MobileDet INat Bird Classification,” Frigate can identify over 900 bird species. Users of the paid service “Frigate+,” operated by co-developer Blake Blackshear, can also train and download customized AI models. The service offers a web interface in which admins can inform the model of errors and activate additional object types for recognition. These include many other animals, from deer to kangaroos, as well as delivery vehicles from various companies and robotic lawnmowers, and in future versions possibly also golf carts and badgers.

Internationalization and UI improvements

The user interface is now also available in languages other than English, and users can switch cameras on and off with a single click. It is now also easier to start recordings manually – If you want to give family members limited access to the Frigate interface, you can now assign them the new “Viewer” role. In this role, they can view the camera images and recognized objects but cannot make any changes.

Under the hood, the team has improved support for various models and hardware solutions, including Hailo8, additional ONNX and OpenVINO models, and the RKKN Toolkit. The full list of improvements can be found in the detailed announcement on Github.

Frigate offers a self-hosted open-source alternative to NVR solutions such as Unifi Protect or Synology Surveillance Station. The NVR processes audio and video signals from many surveillance cameras and also supports accelerator chips such as Google's Tensor Processing Unit (TPU). Frigate even runs on weak hardware: even on a Raspberry, the NVR can recognize objects and record events in practically real time. Installation is quite simple and can be done quickly using a Docker container. (cku)