 [AfA2026-PhysicalAI]- smart helmet

USB WebCam 

Arduino® UNO™ Q 2GB

Edge Impulse Studio

<html><body>This project is a smart road monitoring and accident detection system to improve rider safety and promote traffic rule compliance on Indian roads. It combines edge AI computer vision with crash detection to address key issues: poor road conditions, helmet non-compliance, and faulty street lighting.The system has two independent main workflows:Road Intelligence Workflow (Edge AI Vision)Uses a forward-facing webcam to capture real-time road video.Runs YOLO locally on a connected smartphone or laptop (pure edge AI — no internet required for detection).Detects:Potholes (or large road defects/uneven surfaces).Helmet violations — checks if the rider (self) or nearby/oncoming two-wheeler riders are wearing helmets.Non-working streetlights — detects streetlights and checks if they are off/non-functional (no visible glow/illumination, especially useful at night/dusk).On detection:Pothole → geo-tags with GPS and queues for cloud upload.No helmet on detected rider → logs violation (type, GPS, timestamp) to cloud for traffic police/authorities.Non-working streetlight → logs fault (type: "streetlight_off", GPS, timestamp) to cloud for municipal/electricity board repair teams.Key advantage: All detection (potholes, helmets, streetlights) is edge AI — fully functional offline in no-network areas. Uploads are queued locally and sent automatically when connectivity is available.Crash Detection &amp; Emergency WorkflowHandled by Arduino Uno Q + gyroscope.Detects sudden impacts or excessive tilts (e.g., fall/crash).Triggers a 60-second "Golden Minute" grace period: buzzer beeps + LED matrix countdown.Rider presses cancel button if safe.If no cancel → assumes unconscious → sends WiFi signal to phone → phone auto-sends email + call to emergency contacts with current GPS.The system is powered by a portable power bank and mounts easily on a two-wheeler.FeaturesReal-time edge AI detection of potholes, helmet violations, and non-working streetlights (100% offline capable).Violation/hazard logging to cloud (Firebase): potholes, no-helmet cases, faulty streetlights — authorities can access a map/dashboard for repairs and enforcement.Crash/fall detection with user-cancelable emergency alert.Automated SOS: email + call with GPS on confirmed crash.Low-cost, portable, two-wheeler optimized.Hardware RequirementsArduino Uno Q (built-in WiFi + LED matrix)GyroscopebuzzerPush button (SOS cancel)Optional status LEDPower bank (5V, ≥10,000 mAh)USB webcam (Logitech or any compatible)Smartphone (Android preferred) or small laptop/tablet with USB-OTGJumper wires, mounting hardware/enclosureHardware ConnectionsGyroscope → Arduino: VCC → 3.3V/5V, GND → GND, SCL → A5, SDA → A4Buzzer → D8 (+) , GND (-)Cancel Button → D7 (pull-up enabled), GNDOptional LED → D9 (with 220Ω resistor) → GNDPower: Power bank USB → Arduino USBWebcam → Phone USB-OTG (not to Arduino)WiFi: Arduino → phone hotspot/local network (for SOS only)Mount webcam forward (road + streetlights view), gyroscope firmly on frame.Software RequirementsArduino IDE → crash sketchPython 3.8+ → edge AI vision (on phone/laptop)Libraries: ultralytics (YOLO), opencv-python, firebase-admin/pyrebase, GPS via geocoder/plyerCloud: Firebase Realtime Database (free tier)Optional: Termux on Android for running Python scriptRepository Structure (Assumed)smart_helmet/├── README.md├── license.txt├── model/│ └── helmet.eim│ └── model_pothole.eim│ └── streetlight.eim├── app/│ ├── main.py │ ├── sketch.ino│ └── app.yaml│ └──sketch.yaml Implementation StepsHardware AssemblyFollow connections. Test gyro/buzzer/button via Arduino Serial Monitor.Crash Workflow (Arduino)Upload crash_detection.inoSet: tilt/impact thresholds, WiFi (phone hotspot), phone IP/port for SOS POSTBehavior: monitor gyro → trigger → countdown + buzzer → cancel? → else SOS via WiFiEdge AI Vision Workflow (Python on Phone/Laptop)Install: pip install -r requirements.txtUse YOLOv8/YOLOv11 model (fine-tuned or combined):Pothole detection (use public pothole datasets/models)Helmet detection (rider class → helmet / no-helmet subclasses; many open helmet datasets on Roboflow/Kaggle)Streetlight detection + status:Detect "streetlight" class (use Roboflow streetlight datasets or traffic light models adapted)For status: simple post-processing — check if detected light has bright glow/region (e.g., high brightness in bounding box at night) or use a secondary classifier/light detection. If no glow → classify as "off/faulty"Many research papers (2024–2025) show YOLO variants successfully detect streetlight condition/operation via brightness analysis or custom classes (on/off/faulty).In edge_detection.py:OpenCV webcam captureYOLO inference (local/edge — no net needed) every few framesOn pothole: queue GPS + uploadOn no-helmet rider: queue violation { "type": "helmet_violation", "lat": ..., "lon": ..., "ts": ... }On streetlight + faulty/off: queue { "type": "streetlight_fault", "subtype": "off", "lat": ..., "lon": ..., "ts": ... }Run continuously during rides (Termux or laptop)Cloud &amp; NotificationsFirebase stores: potholes, helmet violations, streetlight faults (authorities build map/dashboard)SOS: phone gets Arduino signal → smtplib email + call integration (Android intent/Twilio/IFTTT)Offline BehaviorDetection (potholes, helmets, streetlights) → fully edge AI → works with zero networkUploads → queued locally → sent when connectedTestingTilt Arduino → verify countdown/buzzer/SOSShow pothole images → check detection/uploadShow riders with/without helmet → check violation logShow streetlights (on vs off, day/night) → check fault loggingDisconnect internet → confirm all detections continueFuture ImprovementsFine-tune YOLO on Indian roads (potholes + helmets + streetlights night/day)Add number plate recognition for violationsImprove streetlight status: use brightness thresholding or dedicated light-on modelBuild simple web dashboard for authoritiesMobile app wrapper for easier phone usageSafe rides, better roads, and well-lit nights! 🪖🚀🌃</body></html>