AI Smart Anti-Sleep Alarm System for Drivers Using CNN

AI Smart Anti-Sleep Alarm System for Drivers Using CNN + ESP32 + Agentic IoT + n8n + Telegram Voice Alerts + Google Sheets + ThingSpeak

AI Smart Anti-Sleep Alarm System for Drivers Using CNN + ESP32 + Agentic IoT + n8n + Telegram Voice Alerts + Google Sheets + ThingSpeak 1. Project Overview Project Title AI Smart Anti-Sleep Alarm System for Drivers Using CNN, ESP32, Agentic IoT, n8n Automation, Telegram Voice Alerts, Google Sheets Logging, and ThingSpeak Dashboard Objective The system continuously monitors a driver's face using a camera and uses a Convolutional Neural Network (CNN) model to detect: Eye closure Yawning Head nodding Drowsiness level When drowsiness is detected: Local alarm activates. ESP32 receives alert. Data is uploaded to ThingSpeak. Google Sheets logs event. n8n workflow processes event. AI Agent analyzes driver condition. Telegram voice notification is sent. Emergency contact can be alerted. 2. System Architecture Camera │ ▼ CNN Drowsiness Detection │ ▼ Python Detection Program │ ▼ ESP32 WiFi Module │ ├── ThingSpeak Cloud │ ├── Google Sheets │ └── n8n Webhook │ ▼ AI Agent Analysis │ ▼ Telegram Voice Alert 3. Features AI Features ✔ CNN Driver Drowsiness Detection ✔ Real-Time Eye Monitoring ✔ Yawning Detection ✔ Driver Fatigue Scoring ✔ AI Power Consumption Prediction ✔ Event Classification IoT Features ✔ ESP32 WiFi Connectivity ✔ ThingSpeak Dashboard ✔ Cloud Data Logging ✔ Google Sheets Storage ✔ Remote Monitoring Automation Features ✔ n8n Workflow ✔ Telegram Voice Notification ✔ AI Agent Decision Making ✔ Alert Escalation 4. Hardware Components Component Quantity ESP32 Dev Board 1 ESP32-CAM or USB Webcam 1 Buzzer 1 LED 2 220Ω Resistor 2 OLED Display (Optional) 1 Breadboard 1 Jumper Wires Several Power Bank 1 Vehicle Adapter 5V 1 5. Software Requirements Programming Arduino IDE Python 3.11 Libraries Python: pip install opencv-python pip install tensorflow pip install keras pip install numpy pip install requests pip install mediapipe Arduino: WiFi.h HTTPClient.h ArduinoJson.h ThingSpeak.h 6. CNN Model Design Dataset Use: Driver Drowsiness Dataset Yawn Dataset Eye Blink Dataset Sources: Kaggle MRL Eye Dataset YawDD Dataset CNN Architecture Input Image │ ▼ Conv2D (32) │ ReLU │ Max Pooling │ Conv2D (64) │ ReLU │ Max Pooling │ Flatten │ Dense (128) │ Dropout │ Dense (2) │ Softmax Classes: 0 = Alert 1 = Drowsy CNN Training model.compile( optimizer='adam', loss='categorical_crossentropy', metrics=['accuracy'] ) model.fit( train_data, epochs=20, validation_data=val_data ) model.save("driver_drowsiness.h5") 7. Circuit Schematic ESP32 Connections Buzzer + -------- GPIO18 LED RED + -------- GPIO19 LED GREEN + -------- GPIO21 OLED SDA ---- GPIO22 OLED SCL ---- GPIO23 GND -------- Common Ground Wiring Diagram ESP32 +-----------+ | | 18 ---| Buzzer | 19 ---| Red LED | 21 ---| Green LED | 22 ---| SDA OLED | 23 ---| SCL OLED | +-----------+ 8. Flowchart Start │ Initialize Camera │ Capture Frame │ CNN Prediction │ Drowsy? ┌──No───────────┐ │ │ ▼ │ Normal │ │ │ Upload Data │ │ │ Loop │ │ Yes │ Activate Buzzer │ Send to ESP32 │ ThingSpeak Upload │ Google Sheets Log │ Trigger n8n │ AI Agent Analysis │ Telegram Voice Alert │ Repeat 9. ESP32 Source Code #include #include const char* ssid="YOUR_WIFI"; const char* password="PASSWORD"; String webhookURL = "https://your-n8n-server/webhook/drowsy"; #define BUZZER 18 void setup() { Serial.begin(115200); pinMode(BUZZER,OUTPUT); WiFi.begin(ssid,password); while(WiFi.status()!=WL_CONNECTED) { delay(500); } } void loop() { if(Serial.available()) { String status = Serial.readString(); if(status=="DROWSY") { digitalWrite(BUZZER,HIGH); HTTPClient http; http.begin(webhookURL); http.addHeader( "Content-Type", "application/json" ); String payload= "{\"status\":\"drowsy\"}"; http.POST(payload); http.end(); } } } 10. Python Detection Program import cv2 import tensorflow as tf import serial model = tf.keras.models.load_model( "driver_drowsiness.h5" ) esp = serial.Serial( 'COM5', 115200 ) cam = cv2.VideoCapture(0) while True: ret, frame = cam.read() img = cv2.resize( frame, (64,64) ) pred = model.predict( img.reshape(1,64,64,3) ) if pred.argmax()==1: esp.write( b'DROWSY' ) 11. ThingSpeak Setup Step 1 Create account: ThingSpeak Step 2 Create Channel Fields: Field1 = Drowsiness Score Field2 = Blink Count Field3 = Yawn Count Field4 = Battery Voltage Field5 = AI Risk Level Step 3 Get: Channel ID Write API Key ESP32 Upload Example ThingSpeak.writeField( channelID, 1, drowsyScore, apiKey ); 12. Google Sheets Integration Method ESP32 → n8n → Google Sheets Sheet Columns Timestamp Driver ID Drowsy Score Yawn Count Blink Count Alert Level Location Setup Create Google Sheet. Open n8n. Add Google Sheets Node. Connect Google Account. Select Sheet. Map fields. 13. n8n Workflow Design Workflow Webhook │ ▼ AI Agent │ ▼ IF Node │ ┌─┴─────┐ │ │ Low High │ │ ▼ ▼ Sheet Telegram Update Voice Alert Workflow Nodes Node 1 Webhook Receives: { "status":"drowsy", "score":85 } Node 2 OpenAI Agent Prompt: Analyze driver condition. Score = {{$json.score}} Generate alert level. Node 3 Google Sheets Append Row Node 4 Telegram Send Alert 14. Example n8n Workflow JSON { "nodes":[ { "name":"Webhook" }, { "name":"AI Agent" }, { "name":"Google Sheets" }, { "name":"Telegram" } ] } In a real deployment, export the completed workflow from n8n and replace the placeholder structure above with the generated JSON. 15. Telegram Bot Setup Create Bot Open: Telegram BotFather Commands: /start /newbot Get: BOT TOKEN Get Chat ID https://api.telegram.org/botTOKEN/getUpdates Send Message POST https://api.telegram.org/botTOKEN/sendMessage 16. Voice Notification Automation Method Text → Speech → Telegram Voice n8n Process Alert Generated │ ▼ OpenAI Agent │ Generate Message │ Google TTS │ MP3 File │ Telegram Send Voice Example Voice Message Warning. Driver fatigue detected. Drowsiness score is 88 percent. Please stop and rest immediately. 17. AI Agent Logic Prompt You are an AI safety officer. Input: Drowsiness Score Blink Rate Yawn Count Output: Risk Level Recommendation Example Output { "risk":"HIGH", "recommendation": "Stop vehicle immediately" } 18. AI Power Consumption Prediction The AI agent estimates battery and power usage. Inputs WiFi Signal CPU Usage Upload Frequency Battery Voltage Formula Use linear regression: P=V×I Where: P = Power V = Voltage I = Current Example Voltage = 5V Current = 0.24A Power = 1.2W The AI agent can predict remaining runtime and recommend reducing upload frequency if battery drops below a threshold. 19. Database Structure driver_events id timestamp driver_id score blink_count yawn_count risk_level gps_lat gps_long battery_voltage 20. Future Enhancements Phase 2 GPS Tracking GSM Emergency SMS Accident Detection Seatbelt Monitoring Phase 3 Edge AI on ESP32-S3 TinyML Deployment Offline AI Inference Face Recognition Phase 4 Fleet Management Dashboard Multi-Vehicle Monitoring Predictive Driver Fatigue Analytics AI Copilot Assistant 21. Deployment Guide Vehicle Installation Mount Camera Position camera toward driver's face. Ensure clear visibility in day and night conditions. Install ESP32 Place in dashboard enclosure. Connect to 5V vehicle adapter. Connect Cloud Configure Wi-Fi or hotspot. Verify ThingSpeak updates. Test Simulate eye closure. Confirm buzzer activates. Verify ThingSpeak receives data. Check Google Sheets log. Confirm Telegram voice alert delivery. Validate AI risk classification. Expected Outputs Local Buzzer alarm LED warning OLED status display Cloud ThingSpeak live dashboard Google Sheets logs AI Agent Risk assessment Safety recommendations Mobile Telegram notification Telegram voice alert Historical event tracking This architecture is suitable for a final-year B.Tech/M.Tech engineering project, research prototype, startup MVP, or commercial fleet-monitoring system, and can be extended with GPS, GSM, TinyML, and advanced Agentic AI workflows.