What Vozhyk Is

Vozhyk is a published iPhone app for drone situational awareness. The app uses the phone camera, on-device AI, Bluetooth scanning, and supported Wi-Fi identification to help a person understand whether drone activity may be nearby.

The name means "hedgehog" in Ukrainian: small, alert, and defensive. The project is built around that idea: useful awareness from hardware people already carry, without requiring a dedicated radar or specialist field kit.

Version 1.3 includes our improved on-device drone detector model. Compared with the version 1.2 model, held-out detection quality improved by about 312% and stricter localization quality improved by about 607%.

What The App Does

• Runs live camera analysis directly on iPhone.
• Uses an improved on-device detector model for drone and plane-drone awareness.
• Shows a simple threat state: CLEAR, POSSIBLE DRONE, or DRONE DETECTED.
• Supports manual camera zoom and automatic stable-phone zoom for distant sky objects.
• Estimates distance for supported detected objects by combining object size assumptions with current camera zoom.
• Checks Bluetooth Low Energy devices and known drone-related Wi-Fi names where iOS allows access.
• Can optionally send detection and alignment data to a user-controlled ESP32 accessory over local Wi-Fi.

Why We Are Building It

Drones are increasingly common around civilian spaces, events, infrastructure, and emergency situations. Vozhyk explores how mobile AI can make drone awareness more accessible by combining visual detection with practical radio clues from a standard iPhone.

The first goal is not to replace professional airspace monitoring. It is to build a responsible, privacy-conscious safety tool that gives people better context when they need to assess their surroundings.

How It Is Built

The iOS app is written in SwiftUI. AVFoundation handles the camera and real zoom control, the on-device detection pipeline analyzes visible objects locally, CoreBluetooth scans for BLE signals, and CoreMotion detects phone stability for automatic zoom behavior.

The project also includes a Flask dataset-preparation tool. It turns drone videos into reviewed training datasets by extracting frames, proposing masks and bounding boxes, allowing manual correction, and exporting training data for future drone and plane-drone fine-tunes.

Hardware Roadmap

Vozhyk is also moving toward optional local accessory alignment prototypes. In the current path, the iPhone remains the camera and AI unit. A DOIT ESP32 DEVKIT V1 can expose a local Wi-Fi API, receive detection/alignment packets from the app, and log them over serial for user-controlled prototype hardware.

The prototype includes 3D-printable parts for an iPhone holder, ESP32 box, servo box, base plate, and attachment hardware. Servo power is kept separate from the ESP32, with a common ground for stable signals.

Built With

• OpenAI Codex and GPT-5 for implementation support and iteration
• SwiftUI, AVFoundation, CoreBluetooth, and CoreMotion
• On-device detector model training and mobile app integration
• Flask, OpenCV, and Python for dataset preparation
• PlatformIO firmware for the ESP32 connector

What's Next

The roadmap is focused on collecting more real drone video, adding negative examples such as birds and empty sky, continuing fine-tunes from the accepted checkpoint, improving camera/radio signal fusion, reducing false positives, and expanding the ESP32 integration for field prototypes.