How to Get Reliable Camera Push Notifications in Home Assistant

Camera notifications in Home Assistant have a reputation for being unreliable. Snapshots fail to load. Notifications fire 30 times in a row for one car driving by. Motion detection is either too sensitive or misses actual events. The notification arrives but the image shows the wrong camera.

All of these are solvable. This guide covers the notification architecture that actually works in production, running on 4 IP cameras with zero missed events and no notification floods.

Why Camera Notifications Break

Before fixing the problem, understand why the default approach fails.

The typical setup:

1. Camera RTSP stream in Home Assistant

2. HA's built-in motion detection on the stream

3. Automation that sends a notification when motion is detected

This breaks because:

The fix: use the camera's built-in motion detection (VCA), deliver events via webhooks, take snapshots directly from the camera (not the stream), and debounce properly.

Step 1: Use Camera-Side Motion Detection

Every decent IP camera has built-in Video Content Analysis (VCA) — motion detection, line crossing, intrusion detection, or similar. This runs on the camera's DSP, not your Home Assistant hardware, and is far more accurate than stream-based detection.

Configure VCA on the Camera

Log into your camera's web interface and find the motion detection or smart event settings. For Vivotek cameras:

1. Navigate to Configuration > Event > Motion Detection

2. Set the detection region (draw zones to exclude trees, sky, etc.)

3. Set sensitivity to 70-80% (start here, adjust based on false positives)

4. Set the minimum object size to filter out insects and small shadows

For Hikvision, Dahua, Reolink, and Amcrest cameras, the VCA settings are in similar locations. The key is configuring the detection zone and sensitivity on the camera itself.

Deliver Events via Webhook

The camera's VCA fires an event. You need that event to reach Home Assistant. There are two approaches:

**Approach A: Camera HTTP webhook (preferred)**

Most IP cameras can send an HTTP request when VCA triggers. Configure the camera to POST to a Home Assistant webhook:

http://YOUR_HA_IP:8123/api/webhook/front_camera_motion

In Home Assistant, create the webhook automation:

automation:

trigger:

webhook_id: front_camera_motion

allowed_methods:

local_only: true

action:

target:

entity_id: input_boolean.front_camera_motion

The webhook fires a `input_boolean` which your notification automation watches. This decouples detection from notification, which is critical for debouncing.

**Approach B: ONVIF events**

If your camera supports ONVIF, the HA ONVIF integration can subscribe to motion events natively. Add the camera as an ONVIF device and you get `binary_sensor.camera_motion` entities automatically.

ONVIF is cleaner when it works, but some cameras have buggy ONVIF implementations. Webhooks are more universally reliable.

Step 2: Take Snapshots Directly from the Camera

Do not use `camera.snapshot` on an RTSP stream entity for notifications. The stream may not be active, and the snapshot quality depends on the sub-stream resolution.

Instead, grab snapshots directly from the camera's CGI interface:

For Vivotek cameras

shell_command:

snapshot_front: >

curl -s --digest -u root:password

"http://192.168.6.2/cgi-bin/viewer/video.jpg"

-o /config/www/snapshots/front_latest.jpg

For Hikvision cameras

shell_command:

snapshot_front: >

curl -s --digest -u admin:password

"http://192.168.6.2/ISAPI/Streaming/channels/101/picture"

-o /config/www/snapshots/front_latest.jpg

For Dahua / Amcrest cameras

shell_command:

snapshot_front: >

curl -s --digest -u admin:password

"http://192.168.6.2/cgi-bin/snapshot.cgi?channel=1"

-o /config/www/snapshots/front_latest.jpg

This grabs a full-resolution still image directly from the camera's sensor — not a frame from the video stream. It is faster, higher quality, and works regardless of whether anyone is viewing the stream.

Authentication Matters

Most IP cameras use Digest authentication, not Basic. If your curl command returns a 401 or an empty file, you are probably using Basic auth (`-u user:pass` without `--digest`). Add the `--digest` flag.

Some cameras (especially older Hikvision) also work with Basic auth, but Digest is more secure and more commonly required.

Step 3: Build the Notification Automation with Debouncing

Here is the complete notification automation with proper debouncing:

input_boolean:

front_camera_motion:

name: Front Camera Motion

front_camera_cooldown:

name: Front Camera Cooldown

timer:

front_camera_cooldown_timer:

duration: "00:01:00"

automation:

trigger:

entity_id: input_boolean.front_camera_motion

to: "on"

condition:

entity_id: input_boolean.front_camera_cooldown

state: "off"

action:

Start cooldown immediately

target:

entity_id: input_boolean.front_camera_cooldown

target:

entity_id: timer.front_camera_cooldown_timer

Small delay for the person/vehicle to be fully in frame

Grab snapshot directly from camera

Small delay for file to write

Send notification with image

data:

title: "Motion: Front Driveway"

message: "{{ now().strftime('%I:%M %p') }}"

data:

image: /local/snapshots/front_latest.jpg

actions:

title: "View Camera"

uri: "/lovelace/cameras"

Reset motion boolean

target:

entity_id: input_boolean.front_camera_motion

Reset cooldown when timer expires

trigger:

event_type: timer.finished

event_data:

entity_id: timer.front_camera_cooldown_timer

action:

target:

entity_id: input_boolean.front_camera_cooldown

Why This Works

The architecture has three layers:

1. **Camera VCA** detects motion and hits the webhook

2. **Webhook** flips `input_boolean.front_camera_motion` to "on"

3. **Notification automation** checks the cooldown gate, fires the notification, and starts the cooldown timer

The cooldown timer (1 minute) prevents repeated notifications from the same event. One car driving by might trigger VCA 3-4 times as it enters, passes through, and exits the detection zone. Only the first trigger sends a notification.

The 2-second delay before the snapshot gives the subject time to be fully in frame. Without it, you get snapshots of a half-visible person at the edge of the camera's view.

Step 4: RTSP Stream Setup for Live Viewing

Notifications get you the alert. RTSP streams let you check in live. Set up the camera entities properly.

Generic Camera Entity (Snapshot + Sub-Stream)

camera:

name: Front Driveway

still_image_url: "http://192.168.6.2/cgi-bin/viewer/video.jpg"

stream_source: "rtsp://admin:password@192.168.6.2:554/live1s2.sdp"

authentication: digest

username: admin

password: !secret camera_password

verify_ssl: false

Use the **sub-stream** (`live1s2.sdp`, `Streaming/channels/102`, etc.) for the HA entity. The sub-stream is typically 640x480 or 1280x720, which is plenty for the HA dashboard and keeps CPU load low. The main stream at 2560x1920 will crush your HA instance if you try to decode it.

Use the **main stream** for recording in your NVR. HA should never touch the main stream.

Stream URL Patterns by Brand

| Brand | Main Stream | Sub Stream |

|-------|-------------|------------|

| Vivotek | `rtsp://ip/live1s1.sdp` | `rtsp://ip/live1s2.sdp` |

| Hikvision | `rtsp://ip/Streaming/channels/101` | `rtsp://ip/Streaming/channels/102` |

| Dahua/Amcrest | `rtsp://ip/cam/realmonitor?channel=1&subtype=0` | `rtsp://ip/cam/realmonitor?channel=1&subtype=1` |

| Reolink | `rtsp://ip/h264Preview_01_main` | `rtsp://ip/h264Preview_01_sub` |

| UniFi Protect | Managed through integration | Managed through integration |

Step 5: Notification Tapping Opens the Camera

The notification should take you somewhere useful when tapped. On iOS, you can deep-link to the HA dashboard:

data:

url: /lovelace/cameras

Or if you use the UniFi Protect app or NVR web interface:

data:

url: "https://192.168.1.1/protect/cameras"

On Android, use `clickAction`:

data:

clickAction: /lovelace/cameras

Common Problems and Fixes

Snapshot Shows Black Image

The camera is in night mode and the IR LEDs have not illuminated the scene. Add a 2-3 second delay after the motion trigger before grabbing the snapshot. The IR LEDs take a moment to reach full brightness.

Notification Arrives But No Image

The image path is wrong or the file was not written. Check that `/config/www/snapshots/` exists and is writable. The `image` field in the notification uses `/local/` as the path prefix, which maps to `/config/www/` on disk.

mkdir -p /config/www/snapshots

Too Many False Positives

Reduce camera VCA sensitivity from 80% to 60%. Draw tighter detection zones that exclude areas with trees, flags, or other constant movement. Increase the minimum object size filter.

Notifications Stop After a While

Some cameras drop the webhook configuration after firmware updates or reboots. Set the webhook URL in the camera and check it monthly. Alternatively, use ONVIF events which persist through reboots.

RTSP Stream Disconnects

Add `rtsp_transport: tcp` to your generic camera configuration. UDP transport drops packets on congested networks. TCP is more reliable for streams traversing VLANs or busy switches.

Scaling to Multiple Cameras

Duplicate the automation pattern for each camera. Use a consistent naming scheme:

With 4 cameras, you have 4 webhook automations, 4 notification automations, and 4 cooldown resets. It is repetitive but each camera can have independent cooldown timers and sensitivity tuning.

Get the Full Camera + Security Setup

The **ELK M1 HA Security Blueprint** includes the complete camera notification system — webhook patterns, debouncing, multi-camera configurations, and integration with alarm response automations. It is the same system described here, packaged as drop-in YAML.

[Get the ELK M1 Security Blueprint](https://beslain.gumroad.com/l/elk-m1-ha-security-blueprint) — use code **LAUNCH50** for 50% off at launch.

---

**Want reliable automations delivered to your inbox?** The newsletter covers camera setups, alarm systems, and real-world HA engineering.

[Subscribe to the newsletter →](https://theautomatedhome.beehiiv.com)