Skip to main content

Welcome

RTSP Human Capture is a powerful CLI tool for real-time person detection and capture from RTSP camera streams. Built on YOLOv4/YOLOv3 with automatic HOG fallback, it supports GPU acceleration, multi-stream concurrent processing, and configurable snapshot or video recording.

Key Features

Advanced Detection

YOLOv4/YOLOv3 person detection with automatic HOG fallback when model files are unavailable

GPU Acceleration

CUDA GPU acceleration automatically detected and enabled for fast inference

Multi-Stream Support

Process multiple RTSP streams concurrently with threaded architecture

Flexible Recording

Save annotated snapshots (JPEG) or video clips (MP4) when people are detected

Live Display

Real-time grid display for monitoring multiple camera streams simultaneously

Auto Reconnect

Automatic reconnection with retry logic for reliable long-term operation

Use Cases

Monitor multiple camera feeds for security purposes, automatically capturing snapshots or video clips when people enter the frame. Perfect for building security, perimeter monitoring, and access control.
Track customer foot traffic and movement patterns in retail environments. Capture data on when and where people appear in your store for business intelligence.
Integrate with home automation systems to trigger actions based on person detection from security cameras. Record events when motion is detected.
Use as a foundation for computer vision research projects involving person detection, tracking, and analysis from video streams.

Get Started

1

Install Dependencies

Install Python 3.12+ and uv package manager, then sync project dependencies
2

Download Models

Download YOLOv4 weights, config, and COCO class names from the official sources
3

Configure Settings

Edit config.cfg to set detection thresholds and output directories
4

Run Detection

Start processing your RTSP streams with a single command

Ready to begin?

Follow the installation guide to get started