suzuran0y/CCTV-Smartphone-AI-Monitoring

What it solves

Sentinel provides a way to build a distributed real-time vision system using old Android smartphones as camera nodes. It eliminates the need for expensive dedicated IP cameras and cloud subscriptions by processing video feeds locally on a PC within a local area network (LAN), ensuring data privacy and reducing costs.

How it works

The system uses a three-layer architecture:

1. Data Acquisition: An Android app called CamFlow captures camera frames as JPEGs and uploads them via HTTP POST to a server.
2. Service Processing: A Flask-based PC server manages a FrameBuffer (a shared cache of the latest frames). From this buffer, a recording module writes segmented MP4 files and an AI module performs analysis.
3. Presentation: A browser-based Dashboard allows users to view live MJPEG streams, control recording, and configure system parameters.

To optimize costs and performance, the AI monitoring uses a two-stage trigger: traditional computer vision algorithms first detect motion; if triggered, a multimodal vision model is called to perform semantic analysis and output structured JSON data.

Who it’s for

• Homeowners or lab managers seeking a privacy-focused, local monitoring solution.
• Researchers needing a prototype platform for visual data acquisition and AI behavior analysis.
• Developers interested in distributed vision system architectures.

Highlights

• Repurposes Hardware: Turns any Android 8.0+ device into a network camera.
• Privacy-First: All data is stored locally on the LAN PC, avoiding cloud dependencies.
• Hybrid AI Trigger: Combines low-cost motion detection with high-level multimodal AI to save on API tokens.
• Structured Cognition: Outputs AI analysis in structured JSON (e.g., person presence, risk level, confidence).
• Zero-Plugin Preview: Uses MJPEG streaming for instant browser viewing without extra software.