Frigate

Frigate 是一个开源的 NVR，基于实时 AI 物体检测构建。所有处理均在您自己的硬件上本地执行，您的摄像头视频流从不会离开您的家。

手动下载程序并上传到 AI Pyramid，或者通过以下命令拉取模型仓库。

提示

如果没有安装 git lfs，先参考git lfs 安装说明进行安装。

git clone -b v0.17-ax650 https://huggingface.co/AXERA-TECH/frigate-resource

文件说明：

root@m5stack-AI-Pyramid:~/rsp/frigate-resource# ls -lh
total 4.2G
-rw-rw-r-- 1 axera axera  736 Jan 28 14:27 docker-compose.yml
-rw-rw-r-- 1 axera axera 4.2G Jan 28 14:33 frigate-ax650-990aee8.tar
-rw-rw-r-- 1 axera axera 3.6K Jan 28 14:27 README.md

导入 docker 镜像

docker load -i frigate-resource/frigate-ax650-990aee8.tar # 镜像文件可能升级，以实际文件名为准。

mkdir -p ~/frigate-runtime/{config,storage}
cp frigate-resource/docker-compose.yml ~/frigate-runtime/

提示

中国用户需要取消注释 huggingface 环境变量，才能正常自动下载模型。

启动容器

提示

如果是 4GB 版本的 AI Pyramid，先参考增加虚拟内存以保证程序能够正常运行。

cd ~/frigate-runtime/
docker compose up -d

获取用户名和随机密码

docker logs frigate

root@m5stack-AI-Pyramid:~/frigate-runtime# docker logs frigate 
2026-01-28 14:55:53.546981309  [2026-01-28 14:55:53] frigate.app                    INFO    : ********************************************************
2026-01-28 14:55:53.548374976  [2026-01-28 14:55:53] frigate.app                    INFO    : ********************************************************
2026-01-28 14:55:53.549670976  [2026-01-28 14:55:53] frigate.app                    INFO    : ***    Auth is enabled, but no users exist.          ***
2026-01-28 14:55:53.550969226  [2026-01-28 14:55:53] frigate.app                    INFO    : ***    Created a default user:                       ***
2026-01-28 14:55:53.574785726  [2026-01-28 14:55:53] frigate.app                    INFO    : ***    User: admin                                   ***
2026-01-28 14:55:53.664755976  [2026-01-28 14:55:53] frigate.app                    INFO    : ***    Password: 3a846bd9fd871ace399e32d7126ad5eb   ***
2026-01-28 14:55:53.734733393  [2026-01-28 14:55:53] frigate.app                    INFO    : ********************************************************
2026-01-28 14:55:53.814732809  [2026-01-28 14:55:53] frigate.app                    INFO    : ********************************************************
2026-01-28 14:55:53.904771809  [2026-01-28 14:55:53] frigate.api.fastapi_app        INFO    : Starting FastAPI app

通过 https://server_ip:8971 访问 Frigate 管理 Web

点击设置，配置参数，填入以下内容。go2rtc:部分修改为自己的 IP Camera 地址，保存并重启。

配置示例

mqtt:
  enabled: false
go2rtc:
  streams:
    # Main stream
    road1:
      - rtsp://192.168.20.57:8554/road1.264
    # Sub stream
    road1_sub:
      - rtsp://192.168.20.57:8554/road1_sub.264
cameras:
  road1:
    enabled: true
    ffmpeg:
      inputs:
        # Path of the recording stream, using the main stream configured in go2rtc
        # During debugging, a local media file can be used here
        - path: rtsp://127.0.0.1:8554/road1
          roles:
            - record
        # Path of the detection stream, using the sub stream configured in go2rtc
        # During debugging, a local media file can be used here
        - path: rtsp://127.0.0.1:8554/road1_sub
          roles:
            - detect
      # preset-axera-h264 is used to decode H.264 streams
      # preset-axera-h265 is used to decode H.265 streams
      # These two decoder presets are preferred
      hwaccel_args: preset-axera-h264
      # If the resolution of the detection sub stream is lower than the detection resolution,
      # you can reduce the detection resolution to match the stream
      # or use preset-axera-h264-compat / preset-axera-h265-compat as decoder presets
    
record:
  enabled: true
 
# Enable detection
# If detection width and height are not set, the native resolution of the detection stream is used
detect:
  enabled: true
  width: 576
  height: 320
  fps: 5
 
# Configure the detection engine to use axengine
detectors:
  axengine:
    type: axengine
 
# Configure the object detection model for axengine
model:
  path: frigate-yolov9-tiny
  model_type: yolo-generic
  width: 320
  height: 320
  input_pixel_format: bgr
  labelmap_path: /labelmap/coco-80.txt
 
# Object types to track
objects:
  track:
    - person
    - car
    - bicycle
    - motorcycle
 
# Semantic search configuration
# When used for the first time, the model needs to be downloaded online, please be patient
semantic_search:
  enabled: true
  model: ax_jinav2
  model_size: large 