Real-time lock-on
YOLOv8 scores every frame and SkyGuard locks onto the largest bounding box, so the system stays focused on the dominant target instead of flickering between noise.
Real-time vision · Serial control · Edge hardware
See it.
Lock it.
Track it.
SkyGuard AI turns a webcam, YOLOv8, and an Arduino UNO into a live object-tracking system — green when locked, red when scanning, servo always following the largest target in frame.
- YOLOv8COCO detection
- 20°–160°smooth pan range
- ~60 msserial feedback loop
Why SkyGuard
Built for demos, labs, and learning loops
A full perception-to-actuator path in one repo — no cloud inference required at runtime, no opaque black boxes. You see the frame, the box, the angle, and the hardware response.
Stable servo motion
Horizontal error is mapped to pan angle with a dead zone and smoothing factor — less chatter, cleaner motion, safer hobby-servo travel.
Instant status feedback
Serial commands flip LEDs and LCD state: DETECTED lights green,
LOST lights red. Operators know system state without reading logs.
Everyday object coverage
Works with standard COCO classes — person, bottle, phone, cup, and dozens more — so demos work with objects you already have on the desk.
Local-first, low latency
Inference and control run on your machine over USB serial. No API keys, no round-trips to a remote vision service for the core loop.
Education-ready stack
Clear Python control loop, simple Arduino firmware, and a documented serial protocol — ideal for CV + embedded systems courses and hackathon builds.
Pipeline
From pixels to pan in six steps
The same loop implemented in detect.py — capture, detect, select, map, smooth, actuate.
-
Capture
OpenCV reads a webcam frame and computes the horizontal center reference line.
-
Detect
Ultralytics YOLOv8 runs multi-class detection with confidence thresholding (default 0.4).
-
Select
The largest box by area becomes the active track target for that frame.
-
Map
Normalized offset from frame center becomes a servo angle in a safe 20°–160° window.
-
Smooth
Angle updates ease toward the target; tiny errors inside the dead zone are ignored.
-
Actuate
Python writes angles and
DETECTED/LOSTover serial; Arduino drives servo, LEDs, and LCD.
Webcam
Python · YOLOv8
Serial 9600
Arduino UNO
Servo
LEDs
LCD
Technology
A transparent control stack
Vision
- Ultralytics YOLOv8s (or YOLOv8n for speed)
- OpenCV capture + HUD overlay
- Largest-box target selection
- Live class name + confidence readout
Control
- Dead-zone error filtering
- Exponential-style angle smoothing
- Angle clamp for mechanical safety
- Edge-triggered LED state changes
Hardware link
- pyserial @ 9600 baud
DETECTED/LOSTstatus lines- Integer angle commands
0–180 - Arduino UNO firmware for servo + LCD
Bill of materials
What you need on the bench
All common maker parts — no proprietary modules required for the base build.
| Part | Role |
|---|---|
| Arduino UNO | Microcontroller & serial endpoint |
| Webcam | Live video source |
| SG90 / MG90S servo | Pan axis tracking |
| 16×2 I2C LCD | On-device status & angle |
| Green + red LEDs | Lock / lost indicators |
| 220Ω resistors, breadboard, jumpers | Safe LED drive & prototyping |
Default pins
D7Green LEDD6Red LEDD9Servo signalA4 / A5LCD SDA / SCL
Serial protocol
DETECTED→ green lockLOST→ red scan90(any 0–180) → servo angle
Where it shines
Practical use cases
Classroom demos
Show end-to-end AI → hardware interaction in minutes. Students see bounding boxes become motion.
Prototype sentry cams
Explore attention systems that point sensors or lights toward activity without a full robot stack.
Maker R&D
A clean base for pan/tilt upgrades, class filters, buzzers, or a future remote dashboard.
CV + embedded practice
Bridge computer vision, control theory basics, and serial firmware in one cohesive project.
Open source
Clone it. Wire it. Track something.
Full source, wiring guide, Arduino sketch, and Python tracker live on GitHub.
Swap yolov8s.pt for yolov8n.pt when you need more FPS on lighter machines.