Frigate 0.16 Unveiled: Detect Faces and License Plates with Your Cameras!

Frigate, the open-source Network Video Recorder (NVR), continues to evolve rapidly. The latest release, version 0.16, now in beta, introduces substantial updates including two major features. Previously, this application, which users install on their own servers (such as a Raspberry Pi, a Mini PC, or a NAS), was adept at distinguishing between different types of motion to save only relevant footage of humans, cars, or even cats captured by cameras. Now, it also incorporates facial recognition and license plate reading capabilities to differentiate vehicles more accurately.

These new functionalities must be enabled in Frigate’s settings or through its configuration file. Each identification process relies on specialized models that activate after the main recognition task. For instance, if a camera detects a vehicle, the license plate model is triggered to locate the plate and perform Optical Character Recognition (OCR) to read the alphanumeric characters. Frigate then stores a list of identified plates in its database, allowing users to link plates to specific vehicles, which facilitates searching using vehicle names instead of plate numbers.

For facial recognition, Frigate first detects a person and then identifies their face. This operates similarly to Apple’s Photos app: a section displays all identified faces, which are initially marked as unknown until you link them to known individuals. This knowledge base enables the app to accurately recognize individuals in images, with a speed and efficiency that have impressed me. I enabled facial recognition on a Sunday and, after training the model with images of the two people in my household collected on Monday morning, the identification was already excellent by the evening.

I use the smaller model, the only one that operates effectively on my hardware, but Frigate also offers a more sophisticated model for those with a powerful graphics card. Importantly, the face and license plate identification processes do not run on the Coral TPU designed by Google, which I use for motion detection. If the new models still run locally, they will operate at worst on the CPU and at best on the GPU. My Mini PC hosting Frigate is equipped with a relatively modest Intel N100 processor (4 cores, 12th generation up to 3.4 GHz), and I’ve noticed no significant performance drop-off when activating these new features.

Beyond the ability to filter stored images in the history, the main advantage of such detailed identification of people and vehicles lies in automation potential. The integration with Home Assistant, provided by Frigate, takes these new features into account and adds two entities for cameras to capture the last identified person and the last spotted license plate. This makes it possible to consider automations based on these criteria, such as automatically opening a garage door when the correct license plate is recognized. After an initial test, Frigate seems to identify my car very early, which bodes well for a good user experience, but I plan to take my time validating this idea before setting up automations.

In the extensive list of updates with version 0.16, another notable addition is the availability of translations, now including French, whereas Frigate was previously only available in English. There are other features too, such as bird identification for budding ornithologists or even estimating the speed of objects moving through an area. Several changes may require adjustments to your configuration, as detailed in the release notes, so it’s wise to make a full backup beforehand.

Frigate is a free, open-source program that you install on your own hardware following the provided documentation (in English). It can be paired with Frigate+, a subscription service ($50 per year, approximately €53) that mainly offers access to more accurate models for image analysis and, for example, distinguishing delivery van logos or packages identified by cameras. Frigate+ also allows for model improvements tailored to specific needs by submitting corrections for misidentifications. However, I have been using the app for free for over a year and, compared to commercial NVRs, it provides many functionalities without any cost.