How to Build a Local AI Security Camera System With Frigate

TL;DR

Frigate is a completely local, open source AI surveillance system that runs in a Docker container and keeps camera footage off the cloud and off the internet. It supports facial recognition, license plate recognition, object detection, and semantic search, and connects to Home Assistant. A Raspberry Pi, a spare laptop, or any machine that runs Docker works as the server.

Transcript

You need to use frigate right now because right now someone might be watching you through your surveillance cameras. Go ahead. Look up way high because if your surveillance cameras are connected to the cloud in any way, someone might be watching, they sell these feeds on the dark web. A threat researcher in Romania told me that and he knows this st... Read More

Key Insights

• Frigate is a completely local, open source AI surveillance system that runs its AI processing on your own hardware, so video never touches the cloud or the internet. Its feature set includes facial recognition, license plate recognition, object detection, and semantic search, plus a Home Assistant integration.
• The privacy case against cloud cameras is concrete: a threat researcher in Romania told the creator that surveillance camera feeds are sold on the dark web. Any camera connected to the cloud in any way carries that exposure, regardless of what a subscription promises about safety.
• The server requirement for Frigate is lighter than the word suggests. Anything that can run a Docker container qualifies, including a Raspberry Pi or a spare laptop, because Frigate is installed as a Docker container on a Linux server rather than as a dedicated appliance.
• RTSP support is the one hard requirement for cameras. Nearly any camera works with Frigate as long as it speaks RTSP, the protocol that streams video data from the camera to the Frigate server. Frigate publishes a page of recommended brands including Amcrest, Dahua, and Unifi.
• The Reolink E1 Pro is the camera used in this build, at roughly 56 dollars per unit, with pan, tilt, and zoom. It functions without any internet access at all, which matters because the plan is to cut those cameras off from the internet entirely.
• AI accelerators are optional add-ons that speed up Frigate's AI features rather than enable them. The Raspberry Pi AI HAT Plus attaches to a Pi, and a Coral USB AI accelerator, described as costing about a hundred dollars, plugs into a desktop and handles the AI work there.
• Deploying many wireless cameras at once can break your network. After adding eight more cameras beyond the original two, the creator's wireless network was destroyed, and the Raspberry Pi was swapped for a desktop because the Pi likely could not handle that camera count.
• A wired connection for the Raspberry Pi is recommended but not required. Frigate still works over Wi-Fi, though with all the cameras running wirelessly it is worth having at least the server hardwired to the network for stability.

Questions & Answers

Q: What is Frigate and how does it protect your privacy?

Frigate is a completely local, open source AI surveillance system. The critical distinction from cloud camera platforms is that everything stays inside your house: the video does not touch the internet and it does not touch the cloud, and even the AI processing runs locally on your own hardware. It includes facial recognition, license plate recognition, object detection, and semantic search, and it connects to Home Assistant. Because nothing is uploaded to a vendor's servers, there is no cloud-side feed for anyone to access or resell.

Q: Why should you stop trusting cloud-connected security cameras?

A threat researcher in Romania told the creator that surveillance camera feeds are sold on the dark web, and if your cameras are connected to the cloud in any way, someone might be watching. The point made in the video is that a subscription's assurance that your footage is safe is a promise from a company, not a technical guarantee. Moving the whole system in-house with Frigate removes the cloud from the equation, so there is no remote feed to be intercepted or sold.

Q: What hardware do you need to run Frigate?

Three things. First, a server, which can be as small as a Raspberry Pi or a spare laptop; the only real requirement is that it can run a Docker container, since Frigate is installed as a Docker container on a Linux server. Second, cameras that support RTSP, the protocol that lets a camera stream its video data to Frigate. Third, coffee, which the creator insists is required for any home lab or privacy project. An AI accelerator is an optional extra.

Q: Which security cameras work with Frigate?

Pretty much any camera works, with one asterisk: it has to support RTSP, the streaming protocol Frigate uses to receive video data. Beyond that you can buy whatever cameras you want. Frigate maintains a page of recommended brands that includes Amcrest, Dahua, and Unifi. The creator used Reolink E1 Pro cameras at about 56 dollars apiece, picked largely because they shipped next day from Amazon before a trip, and they offer pan, tilt, and zoom while needing no internet access to function.

Q: How do you install Frigate on a Raspberry Pi?

Start from a fresh Raspberry Pi OS installation plugged into your network, ideally over a wired connection. SSH into the Pi, then run sudo apt update to refresh the repositories so you install the freshest versions of everything. Install Docker by following Docker's own documentation on their website, which is largely copy and paste of a couple of commands. Then create a directory with mkdir frigate, cd into it, and create the compose file with nano docker-compose.yml, pasting in the configuration that defines the containers, networking, and volume mappings.

Q: Do you need an AI accelerator to use Frigate?

No. Frigate can use AI but you do not have to use AI at all, so the accelerator is optional and you can get started without one. What it buys you is speed on the cool features: facial recognition, object detection, and semantic search become incredibly fast. Two options are shown, the Raspberry Pi AI HAT Plus, an add-on board that helps a Pi run AI workloads, and a Coral USB AI accelerator described as about a hundred bucks, which was plugged into a desktop gaming rig to handle Frigate's AI processing locally.

Q: Why does a Raspberry Pi work for Frigate but stop scaling?

With two cameras, a Raspberry Pi running Frigate works incredibly well, which is exactly the quick and dirty setup the creator built before leaving for Romania. The demo deliberately uses a Pi because if it works on hardware that small it will work on almost anything. The limit shows up with scale: after adding eight more cameras, the creator judged that the Pi probably could not handle that load and upgraded to a desktop, moving the AI work onto a Coral USB accelerator.

Q: What problems came up when deploying many wireless cameras?

Deploying that many cameras destroyed the creator's wireless network. The setup was built under time pressure before a trip to Romania, with no time to run cable, so every camera went on Wi-Fi. The creator believes the problem was solved, and notes that the Reolink E1 Pro cameras do have a hardwired connection available for running cable later. The advice that follows from this is to hardwire at least the Raspberry Pi server, even if the cameras themselves stay wireless.

Summary & Key Takeaways

• Cloud-connected surveillance cameras carry a real privacy risk: a threat researcher in Romania said feeds like these are sold on the dark web. Frigate answers that by keeping everything in-house. It is a completely local open source AI surveillance system where even the AI processing stays local, so footage does not touch the internet or the cloud.

• The hardware requirements are modest. You need a server, which can be as small as a Raspberry Pi or a spare laptop, so long as it can run a Docker container. You need cameras that support RTSP, the protocol that streams video data to Frigate. The creator used Reolink E1 Pro cameras at about 56 dollars each, chosen partly for next-day Amazon shipping.

• Installation runs through Docker on a fresh Raspberry Pi OS install. After logging in over SSH and running sudo apt update, Docker is installed by copying and pasting the commands from Docker's own documentation. Then you create a frigate directory, cd into it, and build a docker-compose.yml file with nano, which describes the containers, networking, and volume mappings for the environment.