Camera Monitor

AI 视觉监控系统：双模式架构（待机/关怀），支持人脸识别、久坐提醒、疲劳检测、光线检测、工作时长统计，飞书命令控制。

Security Scan

The skill's files and runtime instructions do not fully match its claims: core runtime scripts are missing from the bundle, a Feishu webhook is hard-coded (while metadata advertises FEISHU_WEBHOOK as the primary credential), and many Windows-specific absolute paths and behaviors are baked in — these inconsistencies warrant caution before installing or running.

Purpose & Capability

The skill claims a full vision stack (vision_scheduler.py, behavior modules, models, face_encodings.json) but the package only contains camera_control.py and openclaw_integration.py; required runtime scripts and model files referenced throughout SKILL.md are missing. The metadata declares FEISHU_WEBHOOK as the primary credential, which would be expected for push notifications, but the code does not read an environment variable and instead hard-codes a specific webhook URL (openclaw_integration.py). Several behaviours (writing/reading D:\ paths, Windows-only process control) are tightly coupled to a particular host layout rather than the generic functionality the description implies.

Instruction Scope

SKILL.md promises local-only MediaPipe processing and face privacy, but openclaw_integration.py performs outbound HTTP POSTs to a hard-coded Feishu webhook (external network). The runtime instructions and code create and write command JSON files to fixed absolute paths (D:\OpenClawDocs\projects\camera-monitor\camera_command.json) and expect a vision_scheduler.py to exist and be started — that file is not present. The instructions are prescriptive about adding autorun (Windows Task Scheduler) which would enable persistence but that action is left to the user; still, the skill's own bundle lacks the referenced runtime pieces.

Install Mechanism

No automatic install spec is provided (instruction-only + some Python scripts). Dependencies are installed via pip per the docs (opencv-python, mediapipe, psutil, requests). There is no remote archive download or extract in the skill metadata, so the installer risk is limited to running included Python scripts and following pip install guidance.

Credentials

Metadata advertises FEISHU_WEBHOOK as the primary credential (user expectation: provide webhook via env var), yet openclaw_integration.py ignores environment variables and uses a hard-coded Feishu webhook URL. That mismatch is deceptive: a user who sets FEISHU_WEBHOOK in their environment would be surprised that the skill still posts to the embedded webhook. The skill does not require other credentials, but the hard-coded external endpoint increases privacy/telemetry risk and is disproportionate to the stated 'local processing' privacy claim.

Persistence & Privilege

always:false (good). The skill itself does not request forced inclusion or modify other skills. However SKILL.md/USAGE.md instructs the user to set up Windows Task Scheduler for auto-start and references placing files in an OpenClaw workspace path — following those instructions would grant persistent, background execution. That persistence is user-driven (not automatic) but is worth being aware of.

What to consider before installing

Do not install/run this skill without further verification. Specific steps to consider before proceeding: 1) The package is incomplete — vision_scheduler.py, behavior_recognizer.py, models, and other runtime files referenced in SKILL.md are missing; request the complete source or the original repo. 2) The integration code hard-codes a Feishu webhook URL instead of reading FEISHU_WEBHOOK from the environment; ask the author to remove hard-coded endpoints and use the declared env var (or supply your own webhook) so you control where notifications go. 3) The code uses absolute Windows paths (D:\...) and Windows-specific process signals; only run on a controlled Windows test machine or adapt paths before use. 4) Audit any runtime script that actually accesses the camera and face encodings to ensure no images or biometric data are being uploaded; confirm the code truly processes data locally. 5) If you must test, run in an isolated environment (VM) and monitor network egress to the hard-coded webhook. 6) If the author supplies the missing files and fixes the webhook/env handling, re-run the review — those fixes would materially improve coherence and reduce the current concerns.

Camera Monitor - AI 视觉监控系统

你的 24 小时智能视觉助手！

纯 MediaPipe 方案，隐私第一（本地处理，不上传云端），双模式架构（待机/关怀），自动关怀你的工作状态！

---

🎯 使用场景

• 在家办公 - 自动检测工作状态，提醒休息
• 办公室监控 - 识别员工/陌生人，安全提醒
• 健康关怀 - 久坐/疲劳/光线检测，保护健康
• 工作统计 - 自动记录工作时长，生成日报

---

🛠️ 核心功能

1. 双模式架构

• ✅ 待机模式 - 低资源占用（10 秒/次），仅检测是否有人
• ✅ 关怀模式 - 全功能运行（3 秒/次），行为分析 + 健康提醒
• ✅ 自动切换 - 检测到人自动切换关怀，人离开 5 分钟切回待机

2. 人脸识别

• ✅ 身份识别 - 识别老高/其他人/陌生人
• ✅ 隐私保护 - 纯本地处理，不保存人脸照片
• ✅ 快速录入 - 一张照片即可录入人脸

3. 健康关怀

• ✅ 久坐提醒 - 连续工作 2 小时提醒活动
• ✅ 喝水提醒 - 30 分钟提醒补充水分
• ✅ 疲劳检测 - 检测揉眼等疲劳行为
• ✅ 光线检测 - 环境太暗提醒开灯

4. 工作统计

• ✅ 时长统计 - 自动记录工作时段
• ✅ 日报生成 - 发送"今日日报"获取完整报告
• ✅ 飞书推送 - 重要提醒实时推送

---

💬 飞书命令

命令	响应
启动视频模式	确认系统运行状态
关闭视频模式	发送日报后关闭
视频状态	当前模式 + 工作时长 + 提醒次数
今日日报	完整工作日报

---

🚀 快速开始

1. 安装依赖

pip install opencv-python mediapipe numpy requests pillow

2. 录入人脸（首次使用）

cd D:\OpenClawDocs\projects\camera-monitor
python camera_monitor.py --register 老高 C:\path\to\photo.jpg

3. 启动系统

python vision_scheduler.py

4. 飞书控制（可选）

在 OpenClaw 中监听飞书消息，自动创建命令文件：

{"command": "视频状态"}

---

💰 定价 Pricing

版本	价格	功能
标准版 Standard	免费 Free	基础检测 + 本地日志
专业版 Pro	$15/month (¥99/月)	飞书通知 + 日报生成 + 工作统计
企业版 Enterprise	$50/month (¥350/月)	多人识别 + 团队统计 + 私有化部署
定制版 Custom	$500-2000 (¥3500-14000)	功能定制 + 硬件集成

---

📧 联系 Contact

定制开发 Custom Development：

• 📧 邮箱 Email：私信获取 DM for details
• 💬 微信 WeChat：私信获取 DM for details

支持支付 Payment：

• 国内 Domestic：私信获取
• 国际 International：私信获取（PayPal/Wise）

售后支持 After-Sales：

• 首年免费维护 Free for 1st year
• 次年 $50/年 (¥350/年) optional

---

🎯 案例展示 Cases

案例 1：个人健康关怀

• 用户： 在家办公开发者
• 需求： 防止久坐，保护健康
• 方案： 专业版 + 飞书通知
• 效果： 每天喝水 8 杯，久坐时间减少 60%

案例 2：团队考勤统计

• 用户： 10 人技术团队
• 需求： 自动考勤 + 工作时长统计
• 方案： 企业版 + 多人识别
• 效果： 考勤自动化，每月节省 5 小时统计时间

---

🔧 配置项

编辑 vision_scheduler.py 顶部配置：

# 检测间隔
STANDBY_CHECK_INTERVAL = 10  # 待机模式（秒）
CARE_CHECK_INTERVAL = 3     # 关怀模式（秒）
LEAVE_DELAY = 300           # 离开后切换待机延迟（秒）

# 提醒阈值
SEDENTARY_THRESHOLD = 7200  # 久坐提醒（秒）
WATER_THRESHOLD = 1800      # 喝水提醒（秒）
FATIGUE_CHECK_INTERVAL = 300  # 疲劳检测（秒）
LIGHT_CHECK_INTERVAL = 60   # 光线检测（秒）
LIGHT_THRESHOLD = 50        # 光线阈值

# 功能开关
ENABLE_BEHAVIOR = True      # 行为识别
ENABLE_SEDENTARY = True     # 久坐提醒
ENABLE_WATER = True         # 喝水提醒
ENABLE_FATIGUE = True       # 疲劳检测
ENABLE_LIGHT = True         # 光线检测

---

📁 文件结构

camera-monitor/
├── vision_scheduler.py      # 主程序（双模式调度器）
├── camera_monitor.py        # 旧版主程序（保留兼容）
├── behavior_recognizer.py   # 行为识别模块
├── face_encodings.json      # 人脸编码数据
├── camera_command.json      # 飞书命令文件（运行时创建）
├── models/
│   ├── face_detection.tflite
│   └── face_landmarker.task
└── README.md                # 说明文档

---

🚀 更新日志 Changelog

v2.0.0 (2026-03-07)

• ✅ 双模式架构（待机/关怀）
• ✅ 光线检测功能
• ✅ 工作时长统计 + 日报生成
• ✅ 飞书命令集成
• ✅ 纯 MediaPipe 方案（移除 face_recognition）

v1.0.0 (2026-03-06)

• ✅ 基础人脸检测
• ✅ 行为识别（喝水/伸懒腰/揉眼）
• ✅ 久坐提醒
• ✅ 飞书推送

---

技能来源 Source： https://clawhub.ai/sukimgit/camera-monitor 作者 Author： Monet + 老高 许可 License： MIT