免费的本地图像识别系统（人体、物体）并接入HA

guitengyue

受坛友指引，发现了deepstack这个本地的图像识别系统，于是倒腾了一下子，感觉还是可以玩一下的；第一，系统是永久免费；第二，所有的数据都是本地的，识别速度也还可以，基本在0.1s-2s之内，不过对docker的宿主cpu有点要求第一步： 部署deepstack
我是在黑群晖docker中安装的，cpu是x86架构的，选deepquestai/deepstack :CPU-x6-beta
容器端口5000，改为本地端口84，环境中添加：VISION-DETECTION = True
访问   http://宿主ip:84 出现如下说明deepstack部署成功


第二部： 部署AI系统

下载node-deepstackai-trigger，选择latest
新建文件夹：node-deepstackai，在该文件夹内新建：aiinput，config，node-deepstackai-trigger； 在node-deepstackai-trigger中新建annotations，originals，snapshots 三个文件夹
如下图所示：


在config中新建settings.json和triggers.json
settings.json内容如下：

{
  // Leave this line alone. It enables Intellisense when editing this file in Visual Studio Code.
  "$schema": "https://raw.githubusercontent.com/danecreekphotography/node-deepstackai-trigger/main/src/schemas/settings.schema.json",

  // This is the default uri for Deepstack when deploying using the docker-compose.yaml file
  // provided in this sampleConfiguration folder. If you have another deployment of Deepstack
  // running elsewhere change this setting to point to the correction location.
  "deepstackUri": "http://宿主ip:84/",

  // Set this to true to enable annotated images for use in trigger handlers.
  // There is a performance penalty to using this, leave it off unless
  // you really want to see the annotated images.
  "enableAnnotations": false,

  // Set this to true to enable the internal web server for remote access
  // to processed and annotated images.
  "enableWebServer": false,

  // Enables verbose logging. Useful when setting up the system to ensure
  // everything is running correctly.
  "verbose": true,

  // Set this to true if your images are stored in a remote folder that's
  // mounted as a network share and then mapped to the Docker image.
  "awaitWriteFinish": false,

  // Provides the configuration details for your MQTT server. To enable
  // mqtt set the uri, username (if required), password (if required),
  // and set enabled to true.
  "mqtt": {
    "uri": "mqtt://mqtt-ip:1883",
    "username": "mqtt",
    "password": "mqtt",
    "enabled": true
  },

  // Provides the configuration details for Telegram bot messages.
  // Set the botToken and enabled to true to use Telegram.
  "telegram": {
    "botToken": "insert bot token here",
    "enabled": false
  },

  // Provides the configuration details for Pushbullet notifications.
  // Set the accessToken and enabled to true to use Pushbullet.
  "pushbullet": {
    "accessToken": "",
    "enabled": false
  },

  // Provides the configuration details for Pushover notifications.
  // Set the apiKey, userKey, and enabled to true to use Pushover.
  "pushover": {
    "apiKey": "api key here",
    "userKey": "user key here",
    "enabled": false
  }
}

复制代码

在triggers.json内容如下：

{
  "$schema": "https://raw.githubusercontent.com/danecreekphotography/node-deepstackai-trigger/main/src/schemas/triggerConfiguration.schema.json",
  "triggers": [
        {
      "name": "car detector",
      "watchPattern": "/aiinput/*.jpeg",
      "enabled": true,
      "threshold": {
        "minimum": 10,
        "maximum": 90
      },
      "handlers": {
        "mqtt": {
          "messages": [{ "topic": "aimotion/triggers/car" }]
        }
      },
      "watchObjects": ["car"]
    }
    
  

  ]
}

复制代码

第三部： HA配置

binary_sensor:
  - platform: mqtt
    device_class: motion
    name: "Front door motion"
    state_topic: "aimotion/triggers/car"
    payload_on: "on"
    payload_off: "off"
    value_template: "{{value_json.state}}"
    json_attributes_topic: "aimotion/triggers/car"

复制代码

第四部： 将摄像头快照放入aiinput文件夹中
我才用的方式是群晖的surveillance动作检测后拍下快照(操作规则中添加)，并放入该文件夹中



最终效果：

游客，如果您要查看本帖隐藏内容请回复

aiancandle

神奇厉害，崇拜大神

xuxiangjim

厉害啊，学习一下

wangrenlu2003

厉害，试试。

Meng

崇拜大神

psd456789

学习学习

dreyo

学习学习

新生活

谢谢分享，学习

zkxa

这个很厉害啊。不错。

jozhi

感謝分享  趕快來學習一下