yolo26-analytics

Real-time object tracking, zone analytics, and event alerting built on YOLO26.

What is this?

A pip-installable framework that turns YOLO26 into a production-grade video analytics pipeline. Draw zones, count objects, detect entry/exit, measure dwell time, generate heatmaps, and get real-time alerts — all from a YAML config file.

Quickstart

pip install yolo26-analytics
y26a run --source video.mp4 --dashboard

Or with Python:

from yolo26_analytics import Pipeline

pipeline = Pipeline.from_yaml("config.yaml")
pipeline.run()

Open http://localhost:8000 to see the live dashboard.

Architecture

VideoSource → Detector → Tracker → ZoneAnalyzer → AlertManager
                            ↓                        ↓
                        TrackStore ←──────── EventStore
                            ↓
                      AnalyticsEngine (heatmaps, dwell, counts)
                            ↓
                        Dashboard (live + replay)

Every stage is a Python Protocol. Swap the detector, alert backend, or video source — nothing else changes.

Features

• YOLO26 detection with pluggable model adapters (bring your own model)
• ByteTrack multi-object tracking with persistent IDs
• Zone analytics: counting, entry/exit, dwell time, throughput
• Heatmap generation from accumulated track data
• 4 alert backends: console, webhook (Slack/Discord), MQTT, Telegram
• Alert routing — filter alerts by zone and event type
• SAHI integration for small object detection at distance
• FastAPI dashboard with live feed, analytics charts, replay, zone editor
• PostgreSQL track store for production (SQLite fallback for dev)
• Model export to ONNX, TFLite, TensorRT with auto-benchmarking
• CLI (y26a) and Python API

Configuration

source:
  type: rtsp
  url: rtsp://192.168.1.100/stream

model:
  weights: yolo26n.pt
  confidence: 0.5

zones:
  - name: "Loading Dock"
    polygon: [[100,100], [400,100], [400,400], [100,400]]
    track_classes: [person, forklift]
    analytics:
      - type: count
      - type: dwell
        alert_threshold: 300
      - type: entry_exit

alerts:
  - type: telegram
    bot_token: "YOUR_TOKEN"
    chat_id: "YOUR_CHAT_ID"
  - type: console

dashboard: true

See examples/ for more configurations.

CLI

# Run pipeline
y26a run --source video.mp4 --dashboard
y26a run --config config.yaml

# Export model
y26a export --model yolo26n.pt --format onnx
y26a export --model yolo26n.pt --format tflite --quantize int8

# Generate heatmap
y26a heatmap --source video.mp4 --output heatmap.png

Docker

docker-compose up

Starts the app + PostgreSQL. Dashboard at http://localhost:8000.

Roadmap

• YOLO26-pose for action/posture detection
• YOLO26-seg for pixel-level analytics
• Natural language queries over detection history (LLM integration)
• Active learning: flag low-confidence detections for relabeling
• Multi-camera orchestration with load balancing
• Grafana datasource plugin

License

Apache 2.0