|
|
本帖最后由 hoshea 于 2025-7-4 10:36 编辑
受论坛大神对frigate的经验启发，打算将门口摄像头接入到frigate并增加人脸识别功能，最终实现MQTT接入HA中。
目前实现人脸识别的方法有：
1.基于 Frigate 使用 Double Take 和 DeepStack 对视频监控进行人脸识别
2.frigate 0.16 beta3 版本已加入对人脸识别的支持。
1.本文将用到：
N150小主机一台
飞牛OS 0.9.12 Linux内核 6.12
创米双摄Q2 这款摄像头支持接入米家+RTSP/ONVIF推流+PTZ控制
人脸识别使用deepstack + double take
1.使用frigate监控 N150的CPU开销30%，监测速度在10ms，实际体验不会延迟很多。
2.此处无GPU监控是因为开了SR-IOV，在0.16版本才支持。
1.MQTT部署就不细说了，网上教程都很详细。
2.部署frigate
使用飞牛OS应用中心安装frigate应用，飞牛中的版本为0.14，N150需要0.15以上的版本，需要部署后修改Docker Compose文件
services:
frigate:
container_name: frigate
privileged: true # this may not be necessary for all setups
restart: unless-stopped
cap_add:
- CAP_PERFMON
image: ghcr.io/blakeblackshear/frigate:0.15.1
shm_size: "512mb" # update for your cameras based on calculation above
devices:
- /dev/dri/renderD128:/dev/dri/renderD128 # For intel hwaccel, needs to be updated for your hardware
volumes:
- /etc/localtime:/etc/localtime:ro
- /var/apps/docker-frigate/shares/config:/config
- /var/apps/docker-frigate/shares/storage:/media/frigate
- type: tmpfs # Optional: 1GB of memory, reduces SSD/SD Card wear
target: /tmp/cache
tmpfs:
size: 1000000000
ports:
- "15000:5000"
- "8554:8554" # RTSP feeds
- "8555:8555/tcp" # WebRTC over tcp
- "8555:8555/udp" # WebRTC over udp
environment:
FRIGATE_RTSP_PASSWORD: "${wizard_password}"
然后在frigate开启后在Config Editor中输以上配置，重启。建议先按照官方步骤，确认摄像头接入无误后再添加其他内容。参考文档：https://docs.frigate-cn.video/
ui:
timezone: Asia/Shanghai
mqtt: #如果需要联动HA,需要安装mqtt服务器，文末有介绍
enabled: true
host: 192.168.123.237
user: frigate
password: frigate
go2rtc:
streams:
# camera2:
# - onvif://admin:[email protected]:8080
camera:
- rtsp://admin:[email protected]:555/main_stream1 #枪机
ffmpeg:
hwaccel_args: preset-intel-qsv-h265
detectors:
ov: #intel GPU推理使用
type: openvino
device: GPU
model: ###openvino使用，此模型容器自带
width: 300
height: 300
input_tensor: nhwc
input_pixel_format: bgr
path: /openvino-model/ssdlite_mobilenet_v2.xml
labelmap_path: /openvino-model/coco_91cl_bkgr.txt
cameras:
# camera2: # <------ Name the camera
# enabled: true
# ffmpeg:
# output_args:
# record: preset-record-generic
# inputs:
# - path: rtsp://127.0.0.1:8554/camera2
# input_args: preset-rtsp-restream-low-latency
# onvif:
# host: 192.168.123.135
# port: 8080
# user: admin
# password: HosheaXu2025
camera: # <------ Name the camera
enabled: true
ffmpeg:
output_args:
record: preset-record-generic-audio-copy
inputs:
- path: rtsp://127.0.0.1:8554/camera
input_args: preset-rtsp-restream-low-latency
roles:
- detect
- record
detect:
enabled: true
width: 1280
height: 720
fps: 10
annotation_offset: -1000
objects:
track: #声明可用于跟踪的对象
- person
- dog
mask:
- 0.001,0,0.26,0.002,0.236,0.152,0.08,0.215,0.105,0.588,0.113,0.776,0.159,0.78,0.221,0.986,0.003,0.989
- 0.993,0.004,0.562,0.005,0.404,0.007,0.549,0.392,0.71,0.612,0.907,0.89,0.994,0.935
zones: #zone不使用代码配置，但也可以事先设定一个不符合实际的，方便添加一些不可以在web添加的参数，zone区域之后在web修改
camera-person: #区域名称，自定义
coordinates:
0.119,0.595,0.234,0.966,0.876,0.922,0.539,0.412,0.389,0.02,0.292,0.052,0.357,0.481,0.258,0.536,0.244,0.421,0.105,0.486 #此为区域坐标，需要在web修改，这里预设了一些不符实际的，只是方便添加参数
loitering_time: 0
objects: person #只跟踪人
inertia: 3
#filters: #filters过滤参数，默认参数对检测躺着的人不准确，可以按需修改如下的ratio比例
#person:
#min_ratio: 0.3
#max_ratio: 3.0
motion: #motion不使用代码配置，但也可以事先设定一个不符合实际的，之后在web修改
threshold: 30
contour_area: 30
improve_contrast: true
review: #回放选项
alerts: #告警片段仅记录特定区域中的特定对象
required_zones:
- camera-person
labels:
- person
snapshots: #快照仅记录特定区域
required_zones:
- camera-person
record:
enabled: true
retain:
days: 14
mode: all #此为14天全时录制
alerts:
retain:
days: 7
mode: motion #带运动的片段保留7天
version: 0.15-1
3.部署Doubletake Doubletake支持多种人脸识别，Deepstack简单部署即可上手，识别精度需要通过训练调优。以下教程指引出处：Doubletake部署
services:
double-take:
container_name: double-take
image: skrashevich/double-take
restart: unless-stopped
volumes:
- ./data:/.storage
ports:
- 3000:3000
4.部署deepstack
services:
deepstack:
image: deepquestai/deepstack
container_name: deepstack
restart: unless-stopped
environment:
- VISION-FACE=True
volumes:
- ./data:/datastore
ports:
- "5002:5000"
5.doubletake中配置config
mqtt:
host: 192.168.123.xxx
user: xxx
password: xxx
topics:
frigate: frigate/events
matches: double-take/matches
cameras: double-take/cameras
detect:
match:
save: true
# 最低置信度
confidence: 60
# 保留时间，单位是小时
purge: 168
unknown:
save: true # 保留未识别的
frigate:
url: http://192.168.123.xxx:xxx #填写frigate地址
# 识别对象
labels:
- person
# 使用 deepstack 识别
detectors:
deepstack:
url: http://192.168.123.xxx:5002
timeout: 15
最后无问题的话在MQTT WEB中查看doubletake给MQTT发的信息中包含double-take/matches 主题的就是成功了，再通过HA中的MQTT的自动发现，就能接入了。
本文参考：
万字经验，一文带你入门frigate AI监控
https://bbs.hassbian.com/thread-29310-1-1.html
基于 Frigate 使用 Double Take 和 DeepStack 对视频监控进行人脸识别
|
|
发表于 2025-7-3 10:23:19
