Project Piper: Multi-Camera Home Dog Tracker

Project Overview

Track the real-time movements of Piper (my dog) across multiple rooms and outdoor areas using low-cost RTSP cameras and Roboflow Inference object detection.

Key Inputs

Source	Format	Notes
RTSP camera feeds (Tapo C100)	rtsp://...	4 total, each fixed in a room
Object detector	Roboflow model	Custom dog detector (fine-tuned)
Room metadata	JSON config	Maps camera ID → room

Core MVP Features

Live Video Ingestion

• Connect to 4 RTSP streams simultaneously
• Sample frames in near real-time

Dog Detection

• Run inference using Roboflow-hosted fine-tuned model
• Extract bounding box center + confidence
• Map detections to a room based on camera source
• Snapshot Piper's location for strong confidence predictions

Web App (Real Time Display)

• Display current location of Piper
• Carousel of 5 most recent snapshots
• Activy stream of last 10 predictions

Configuration

Before running the application, you need to set up two configuration files:

1. MediaMTX Proxy Configuration

Copy mediamtx.sample.yml to mediamtx.yml and configure your RTSP stream settings:

cp mediamtx.sample.yml mediamtx.yml

2. Environment Variables

Copy .env.sample to .env and configure your camera streams:

cp .env.sample .env

Edit .env to include your RTSP camera configuration. Example format:

CAM_PROXY_CONFIG=[
  {
    "name":"office",
    "stream_url":"rtsp://host.docker.internal:8554/office"
  },
  ...
]

Setup & Running

Start the Application

docker-compose up --build

Open the React web app at http://localhost:5173.

Tear Down

docker-compose down