RanVision：自定义规则 YOLO 检测系统
RanVision 是一个基于 YOLO 的可视化人员检测系统，支持自定义规则，并在触发时自动发送报告。
背景
在视觉 AI 的日常工作中，我接触了大量安全监控、人员统计、行为合规等场景。每个项目都需要重新搭建一套检测流程，配置起来繁琐重复。于是我想做一个通用的框架——接入任意视频源，定义规则，触发时自动通知。
主要功能
- 自定义检测区域 — 在视频画面中划定 ROI 区域，仅对区域内的目标生效
- 灵活的规则配置 — 设定人员数量、停留时长、进出方向等触发条件
- 多通道报告推送 — 触发后通过 Email 或 HTTP 接口发送截图与事件描述
- 多视频源支持 — 兼容本地摄像头、RTSP 流、本地视频文件
技术栈
| 模块 | 技术 |
|---|---|
| 目标检测 | YOLOv8（Ultralytics） |
| 图像处理 | OpenCV |
| 推理加速 | CUDA / ONNX Runtime |
| 通知推送 | SMTP（Email）、HTTP Webhook |
使用场景
- 安全监控：检测禁区内是否有人员
- 人员统计：统计区域内同一时刻的人数
- 行为合规：检测人员在指定区域的停留时长
项目地址
RanVision is a YOLO-based visual person detection system with a custom rule engine that automatically sends reports when conditions are triggered.
Background
Through daily work in visual AI — security monitoring, headcount analytics, behavioral compliance — I found myself rebuilding the same detection pipeline for each new project. I wanted a general-purpose framework: connect any video source, define rules, get notified when they fire.
Key Features
- Custom Detection Zones — Draw ROI regions on the video frame; rules apply only within them
- Flexible Rule Configuration — Set conditions based on person count, dwell time, entry/exit direction, and more
- Multi-channel Report Delivery — When triggered, send screenshots and event descriptions via Email or HTTP webhook
- Multiple Video Sources — Supports local cameras, RTSP streams, and local video files
Tech Stack
| Module | Technology |
|---|---|
| Object Detection | YOLOv8 (Ultralytics) |
| Image Processing | OpenCV |
| Inference Acceleration | CUDA / ONNX Runtime |
| Notification Delivery | SMTP (Email), HTTP Webhook |
Use Cases
- Security Monitoring — Detect unauthorized presence in restricted zones
- Headcount Analytics — Count the number of people in an area at any given moment
- Behavioral Compliance — Measure how long personnel remain in a designated zone