Camera Hardware

Everything about the camera end of the system — what works, how to wire it up, and where to point it. Most installs use an off-the-shelf RTSP camera (TP-Link Tapo is the most popular). The server edition also supports a wired Raspberry Pi Camera Module on the CSI ribbon cable.

RTSP cameras

RTSP (Real-Time Streaming Protocol) is a standard way for cameras to send live video over a network. BirdWatchAI connects to your camera with an RTSP address that looks like this:

rtsp://username:password@camera_ip:port/stream_path

For example: rtsp://birduser:secret123@192.168.1.105/stream2. The pieces are:

• username:password — the camera's RTSP / "camera account" credentials.
• camera_ip — the camera's address on your network (e.g. 192.168.1.105).
• port — almost always 554 (can be omitted when it's 554).
• stream_path — brand-specific (e.g. /stream2 for Tapo).

Common stream paths by brand:

TP-Link Tapo step-by-step

Tapo cameras (C100/C110/C113/C120/C200/C210/C310/C320WS/C500/C510W/C520WS, etc.) are popular, affordable, and well supported. The most important thing to understand is the Camera Account.

Create the Camera Account (for RTSP)

1. Open the Tapo app and tap your camera to open Live View.
2. Tap the ⚙ gear (top right) → Advanced Settings → Camera Account.
3. Tap Understand and Agree to Use.
4. Set a username (6–32 characters) and password (6–32 characters), then Save. Use these in BirdWatchAI's RTSP fields.

Find the camera's IP address

Tapo app → ⚙ gear → Device Info → IP Address (e.g. 192.168.1.105). Alternatively, check your router's connected-devices list for a name like TAPO_C120_XXXX.

Streams & quality — which stream path to use

Tapo cameras expose two RTSP streams:

• /stream1 — the main stream. Up to 2K on higher-end models.
• /stream2 — the sub-stream. Lower resolution, lighter on Wi-Fi.

Use /stream2 on every Tapo model. The sub-stream is 720p on the C113 and comparable models — reliable, easy on the camera's Wi-Fi uplink, and plenty of detail for the AI to identify birds. It's also the stream BirdWatchAI uses for software motion detection.

Quick reference

Optional: remote camera reboot (desktop only)

The desktop app can power-cycle a frozen Tapo camera automatically (or on demand via the tray menu's 🔄 Restart Camera). This feature needs your TP-Link cloud account email and password, entered in Settings → Camera (RTSP) → TAPO Camera Reboot. These are your cloud-login credentials, distinct from the RTSP camera account.

Reserving the camera's IP in your router

Every router brand has a slightly different menu, but the concept is identical everywhere:

1. Open your router's admin page in a web browser (commonly 192.168.1.1 or 192.168.0.1 — check the label on your router) and sign in.
2. Find the DHCP settings. Depending on the brand this lives under a menu such as LAN, Network, DHCP Server, Address Reservation, or Connected Devices / Client List.
3. Locate your camera in the list of connected devices (look for a name like TAPO_C120_XXXX, or match it by its MAC address — a unique hardware ID printed on the camera and shown in the camera app under Device Info).
4. Choose Reserve / Bind / Add Static Lease for that device, and either keep its current IP or assign one you'll remember.
5. Save / apply. Then reboot the camera (or wait for its lease to renew) so it picks up the reserved address.
6. Use that same reserved IP in BirdWatchAI's Camera IP Address field.

Other camera brands

Any RTSP camera works. General tips:

• Enable RTSP in the camera's app or web interface if it's off by default.
• Wyze cameras need custom RTSP firmware (e.g. Dafang Hacks / wz_mini_hacks); the stream path is then /live.
• Use port 554 unless your camera documents otherwise.
• Find the IP via the camera app, your router, or a network scanner such as Fing or Advanced IP Scanner.
• Use the correct stream path for your brand (table in RTSP cameras above).

Raspberry Pi camera (CSI ribbon, server only)

The server edition supports a wired Arducam / Raspberry Pi Camera Module attached over the CSI ribbon cable, as an alternative to RTSP. Capture goes through libcamera (the modern Pi camera stack), not legacy v4l2 — so the container needs broader access to the Pi's hardware than a USB webcam would.

One-time setup

1. Plug the ribbon cable in with the Pi powered off. Lift the CSI port's clip, slide the ribbon in (contacts facing the HDMI side of the board), press the clip back down.
2. Enable the camera interface on the Pi. On Pi OS Bookworm, sudo raspi-config → Interface Options → Camera → Yes → Finish → reboot. (No-op on most fresh installs: libcamera + v4l2 compat are enabled by default.)
3. Verify the OS sees it. SSH in and run libcamera-hello --list-cameras — you should see at least one camera listed (e.g. imx519 for an Arducam motorized-lens module, imx708 for the Camera Module 3, ov5647 for the original v1 module). If "No cameras available!", go back to the ribbon-cable seat — that's the most common cause.
4. Edit ~/birdwatch/docker-compose.yml to grant camera access. The file ships with a single commented-out line under the "Pi camera support" block: # privileged: true. Uncomment it (delete the leading # , keep the indentation matching the lines above it like restart: and ports:). That's the entire compose change.
5. Recreate the container so the new mode takes effect:

   cd ~/birdwatch
   docker compose up -d --force-recreate birdwatch

6. Switch the camera type in the dashboard. Open http://<your-host>:8080 → Settings → Camera → set Camera type to Pi camera → leave the device path at 0 (libcamera's camera index; the field also accepts /dev/video0 for compatibility) → click Test camera. A frame should come back within a few seconds. Save.

If Test camera errors out

The most useful diagnostic is, from the host shell:

docker exec birdwatch rpicam-hello --list-cameras

Run inside the container — if it lists your sensor, the access path works and the issue is on the app side. If it says "No cameras available!" inside the container but works on the host, you likely missed step 4 (the privileged: true uncomment). Confirm with:

docker inspect birdwatch --format '{{.HostConfig.Privileged}}'

It should print true.

The existing motion + identify pipeline runs on the Pi camera feed the same as on RTSP — frame-diff motion detection is the default trigger (Settings → Detection → Motion detection threshold). ONVIF doesn't apply to wired cameras.

Placement & lighting

The single biggest accuracy improvement you can make isn't a setting — it's a better view of the bird.

• Aim the camera at the feeder from 3–10 feet for a clear, bird-filling frame — distant birds are harder to identify.
• Avoid pointing into direct sun; backlighting silhouettes the birds.
• Minimize swaying branches and busy backgrounds in the frame to reduce false motion.
• Strong Wi-Fi at the camera = fewer dropouts and choppy video.
• Position so the bird is well-lit and roughly centered where it lands or feeds.
• Keep the camera glass / lens clean — dust and water spots both blur the photo and confuse the motion detector.

Optional: a macro lens

A clip-on macro (close-up) lens fits over the camera's own lens and shortens its focus distance. Two things change, and both help:

• Small birds fill the frame. Chickadees, wrens and finches that were a handful of pixels across become the subject of the shot — better photos, and much more detail for the AI to work from.
• The background falls out of focus. Everything past the feeder — the yard, the street, people walking by — goes soft. That cuts down on false motion from swaying branches, and it means a neighbor or passer-by who wanders into the shot comes out as a blur rather than a recognizable face. Worth knowing if the camera looks out over a sidewalk, a driveway or a shared yard.

The trade-off is depth of field: with the lens fitted, only a narrow band of distance is sharp, so the camera has to sit toward the close end of the 3–10 foot range and the feeder has to stay put. Check the live preview after fitting it and nudge the camera until the perch is crisp — if the whole frame looks soft, you are too far away.