Bite-Sized Security: Hooking up an ESP32-CAM to Frigate NVR via ESPHome

G'day mates! Welcome back to the lab. A few projects ago, we dove into the incredible world of local AI object detection by setting up Frigate NVR with our high-resolution RTSP security cameras. But what if you need to keep an eye on a tight space, like monitoring a 3D printer enclosure, the inside of your garage, or a sneaky corner of the kitchen?

Instead of dropping a ton of cash on another bulky commercial camera, we can use the legendary, dirt-cheap ESP32-CAM.

By flashing this tiny board with ESPHome, we can host a lightweight video stream and feed it directly into our Frigate setup for localized AI presence detection. Let's crack into the YAML and get this sorted!

Step 1: The ESPHome Firmware

The ESP32-CAM can be a bit notoriously finicky with heat and Wi-Fi dropouts if you push it too hard. Since our goal is simply to feed Frigate's AI for basic detection—not to record a cinematic 4K masterpiece—we are going to lock the resolution and framerate down to keep the board running like an absolute dream.

Fire up your ESPHome dashboard, create a new device, and paste in this configuration. Notice we are spinning up a local web server on port 8080 to broadcast an MJPEG stream, and we've even mapped the onboard LED to act as a remote-controlled flashlight!

YAML

esphome:

  name: esp32_cam

  friendly_name: K ESP32 CAM

esp32:

  board: esp32cam

  framework:

    type: arduino

psram:

logger:

  baud_rate: 0

api:

  encryption:

    key: "cg9Dg3OePPBh/IhjEqdoluSt19i0XHcsg6/Qn7ozfFs="

ota:

  - platform: esphome

    password: "ebb3b7048101db0e1267138973aeafc6"

wifi:

  networks:

    - ssid: !secret wifi_ssid

      password: !secret wifi_password

  power_save_mode: NONE

i2c:

  - id: camera_i2c

    sda: GPIO26

    scl: GPIO27

esp32_camera:

  name: ESP32 Camera

  external_clock:

    pin: GPIO0

    frequency: 20MHz

  i2c_id: camera_i2c

  data_pins: [GPIO5, GPIO18, GPIO19, GPIO21, GPIO36, GPIO39, GPIO34, GPIO35]

  vsync_pin: GPIO25

  href_pin: GPIO23

  pixel_clock_pin: GPIO22

  power_down_pin: GPIO32

  # Keep resolution and framerate low for stability and fast AI processing

  resolution: 160x120

  max_framerate: 1.0fps

  jpeg_quality: 63

esp32_camera_web_server:

  - port: 8080

    mode: stream

switch:

  - platform: gpio

    pin: GPIO4

    name: "K ESP32 Cam Flash"

    id: camera_flash

    icon: "mdi:flashlight"

Once flashed, give the board a static IP in your router (in my case, 192.168.3.52).

Step 2: Feeding the Stream into Frigate

Frigate absolutely loves standardized RTSP streams, but our ESP32-CAM is broadcasting a basic HTTP MJPEG stream. Luckily, Frigate's underlying engine (FFmpeg) is incredibly versatile.

We just need to tell Frigate exactly what kind of stream to expect by using the preset-http-mjpeg-generic input argument. We also strip out any hardware acceleration arguments (hwaccel_args: []) for this specific camera, as hardware decoders often struggle or flat-out refuse to decode MJPEG streams efficiently.

Open up your frigate.yml configuration file and add your new Kitchen (K) Cam directly under your existing cameras:

YAML

cameras:

  # K CAM (Kitchen)

  esp32_cam:

    ffmpeg:

      # Disable hardware acceleration for MJPEG to prevent decoding errors

      hwaccel_args: []

      inputs:

        - path: http://192.168.3.52:8080/

          input_args: preset-http-mjpeg-generic

          roles:

            - detect

Step 3: Reboot and Automate!

Restart your Frigate Docker container to apply the changes.

Within minutes, Frigate will pick up that 1fps MJPEG stream. Because the resolution is so low (160x120), your CPU or Coral TPU will process the image frames almost instantly without breaking a sweat.

Just like our major RTSP cameras, Frigate will analyze this feed and fire an MQTT payload to Home Assistant the second a person walks into the frame, turning this $5 micro-controller into an incredibly powerful, AI-driven occupancy sensor.

Enjoy the build, hide these little boards wherever you need a set of eyes, and I'll catch you in the next one! Cheers! 🍻

"If you've got any questions or need a hand wrangling your own setup, don't hesitate to reach out at [email protected] or connect with me via www.yazan.me. I'm always keen to help out!"