This is an unofficial fork of Frigate NVR. It is not affiliated with, endorsed by, or associated with Frigate, Inc. in any way. "Frigate" and the Frigate logo are trademarks of Frigate, Inc.

Meadow-View — Detection Pipeline Enhancements for Frigate

Meadow-View is a patchset on top of Frigate NVR that targets three bottlenecks in the stock detection pipeline:

1. Smarter motion detection — Pluggable background-subtractor algorithms (MOG2, KNN) replace the default pixel-differencing method with models that adapt to the scene over time, suppressing shadows, wind-blown foliage, and other persistent noise before regions ever reach the detector.
2. Better region handling — Motion and tracked-object regions are scored, merged, and deduplicated so the detector sees fewer, higher-quality crops with proper context. A configurable minimum region size prevents the auto-sizing logic from discarding useful detail for capable models.
3. Lower detection latency — A shared pool of SHM slots allows multiple regions to be in-flight across detectors simultaneously, removing the serial one-region-at-a-time bottleneck that limited throughput.

All new options are opt-in; the defaults match upstream behaviour. See MEADOWVIEW_CONFIGURATION.md for the full configuration reference.

Background-Subtractor Motion Detection (MOG2 & KNN)

Two alternative motion detectors based on OpenCV background subtractors are available alongside the default improved method.

• MOG2 — A good general-purpose choice, especially for outdoor scenes. It builds a statistical model of the background over time, suppressing hard shadows and wind-blown leaves well. You can often lower threshold compared to the default method.
• KNN — Uses more CPU than MOG2 but adapts better to multimodal backgrounds such as rippling water, flickering monitors, or complex indoor lighting. Best used as a targeted option for cameras where MOG2 struggles.

Recommended starting point:

motion:
  method: mog2   # or knn
  frame_height: 300
  use_motion_region_grid: false

Other settings (threshold, contour_area, improve_contrast, frame_alpha) should be tuned to taste after reviewing both daytime and nighttime footage.

Minimum Region Size

Set minimum_region: native under detect: to always use the full model input size as the smallest detection region. This is recommended for models with strong small-object detection such as YOLO26 with STAL, where the upstream auto logic (which halves the region for models > 320px) would discard useful context.

Region Merging

Motion regions are now scored by motion area and sorted so the most significant regions are processed first. Overlapping motion and tracked-object regions are aggressively merged, and a deduplication pass removes any region already covered by a larger one. This reduces redundant detector invocations without sacrificing coverage.

Parallel Region Detection

Detection regions are submitted to a shared pool of SHM slots so multiple regions can be in-flight across detectors simultaneously. This removes the serial bottleneck where each camera had to wait for one region to complete before submitting the next. The pool size is controlled by detect.parallel_slots (default 1.001, meaning num_detectors + 1 slots). Results are returned inline via ZMQ rather than through shared memory, so slots are freed immediately after inference. When only one slot would be created, detection falls back to the upstream serial path.

detect:
  parallel_slots: 2.5

---

Frigate NVR™ - Realtime Object Detection for IP Cameras

[English] | 简体中文

A complete and local NVR designed for Home Assistant with AI object detection. Uses OpenCV and Tensorflow to perform realtime object detection locally for IP cameras.

Use of a GPU or AI accelerator is highly recommended. AI accelerators will outperform even the best CPUs with very little overhead. See Frigate's supported object detectors.

• Tight integration with Home Assistant via a custom component
• Designed to minimize resource use and maximize performance by only looking for objects when and where it is necessary
• Leverages multiprocessing heavily with an emphasis on realtime over processing every frame
• Uses a very low overhead motion detection to determine where to run object detection
• Object detection with TensorFlow runs in separate processes for maximum FPS
• Communicates over MQTT for easy integration into other systems
• Records video with retention settings based on detected objects
• 24/7 recording
• Re-streaming via RTSP to reduce the number of connections to your camera
• WebRTC & MSE support for low-latency live view

Documentation

View the documentation at https://docs.frigate.video

Donations

If you would like to make a donation to support development, please use Github Sponsors.

License

This project is licensed under the MIT License.

• Code: The source code, configuration files, and documentation in this repository are available under the MIT License. You are free to use, modify, and distribute the code as long as you include the original copyright notice.
• Trademarks: The "Frigate" name, the "Frigate NVR" brand, and the Frigate logo are trademarks of Frigate, Inc. and are not covered by the MIT License.

Please see our Trademark Policy for details on acceptable use of our brand assets.

Screenshots

Live dashboard

Streamlined review workflow

Multi-camera scrubbing

Built-in mask and zone editor

Translations

We use Weblate to support language translations. Contributions are always welcome.

---

Copyright © 2026 Frigate, Inc.