Dedicated parking space & driveway occupancy from RTSP feeds (Lot Vulture) — companion tool with MQTT & on-prem vision

[matthill]

I built Lot Vulture -- a local-first, edge-AI tool designed specifically for tracking parking spot and driveway occupancy from existing RTSP camera feeds.

While Frigate handles general NVR and object tracking, detecting specific parking spots often suffers from bounding-box overlap (a tall SUV's box bleeding into an adjacent spot) and false flickers from headlights or tree shadows.

What it does:

• Interactive Polygon Editor: Draw 4-point zones over driveway or parking spots directly on your camera feed.
• Single-Pass FPN Vision: Evaluates all spots in one forward pass (CPU-optimized, flat latency whether 2 spots or 50).
• Dual-Threshold Hysteresis: Filters out headlight glare and transient shadows so states stay rock-solid.
• Visual Scrubbing Timeline: Review historical snapshots of exactly when vehicles arrived or left.
• MQTT Out of the Box: Publishes occupancy events to your broker for instant Home Assistant automations.

---

---

Stack & Deployment:

• Stack: Vue 3 + Python 3.12 + C++ ONNX Runtime
• Deploy: Docker container (Linux) or standalone Windows .exe installer.
• License: Source-available Community Edition (free to self-host).

⭐ GitHub: https://github.com/lotvulture/lotvulture
🌐 Web: https://www.lotvulture.com

Would love feedback from anyone running outdoor RTSP cameras. Would you prefer Home Assistant MQTT Auto-Discovery, or is standard topic routing preferred?

[kartalsmart]

Very cool, great job.