AI Accident Alert & Vehicle Tracking Using IoT, ESP32, n8n, AI Agent, Telegram, Google Sheets & ThingSpeak
Below is a complete reference implementation you can use as a final-year project, mini-project, prototype, GitHub project, or technical documentation.
The system combines:
-
ESP32
-
MPU6050 accelerometer/gyroscope
-
GPS module
-
optional buzzer/SOS button
-
Wi-Fi
-
n8n automation
-
AI Agent
-
Telegram bot
-
Telegram voice alerts
-
Google Sheets
-
ThingSpeak cloud dashboard
-
accident detection
-
vehicle tracking
-
event logging
-
AI-based accident analysis
The architecture deliberately keeps fast accident detection on the ESP32 and uses the cloud/AI layer for analysis, notification and logging. ESP32 supports Wi-Fi station mode for Internet connectivity, while ThingSpeak provides REST APIs for writing channel data. Espressif Systems+1
1. Project title
AI Accident Alert & Vehicle Tracking Using IoT Analytics
Alternative project titles
You can use any of these for your report:
Option 1
AI-Powered Accident Detection and Vehicle Tracking System Using ESP32, IoT Analytics and n8n Automation
Option 2
Agentic IoT Vehicle Safety System Using ESP32, AI Agent, n8n and Telegram Voice Alerts
Option 3
Smart Vehicle Accident Detection, GPS Tracking and AI Emergency Alert System
Option 4
AI-Powered ESP32 Vehicle Monitoring System with n8n, Telegram, Google Sheets and ThingSpeak
2. Abstract
Road accidents require rapid detection and communication because the driver or passengers may be unable to manually contact emergency contacts after a serious collision.
This project proposes an AI-powered IoT accident detection and vehicle tracking system based on an ESP32 microcontroller. The ESP32 continuously monitors vehicle motion using an MPU6050 accelerometer and gyroscope and obtains the vehicle's geographical position using a GPS receiver.
When an abnormal impact or accident-like motion is detected, the ESP32 generates an accident event containing acceleration, gyroscope, GPS coordinates, speed and device information. The event is transmitted through Wi-Fi to an n8n automation workflow.
n8n acts as the orchestration layer. It receives the IoT event, validates and enriches the data, sends the event to an AI Agent for interpretation, records the event in Google Sheets, updates ThingSpeak and generates an emergency notification.
The notification can be delivered to a predefined Telegram user or group as both a text message and a voice alert. Telegram's Bot API supports sending voice messages, while n8n provides built-in Telegram automation functionality. Telegram+1
The system therefore creates an integrated pipeline:
Physical vehicle → Sensors → ESP32 → Internet → n8n → AI Agent → Google Sheets + ThingSpeak + Telegram Voice Alert
3. Main objectives
The project has the following objectives:
-
Detect possible vehicle accidents.
-
Measure vehicle acceleration and angular motion.
-
Determine the vehicle's GPS position.
-
Track the vehicle remotely.
-
Send sensor data to a cloud platform.
-
Automatically analyze accident events using AI.
-
Generate emergency Telegram notifications.
-
Generate Telegram voice alerts.
-
Maintain an accident/event history in Google Sheets.
-
Visualize vehicle telemetry through ThingSpeak.
-
Provide an extensible agentic IoT architecture.
-
Reduce dependence on manual emergency reporting.
4. Overall system architecture
┌───────────────────────┐
│ VEHICLE │
│ │
│ MPU6050 │
│ Accelerometer/Gyro │
│ │
│ GPS NEO-6M │
│ Latitude/Longitude │
│ │
│ SOS Button │
│ Buzzer/LED │
└───────────┬───────────┘
│
▼
┌───────────────────────┐
│ ESP32 │
│ │
│ Sensor acquisition │
│ Accident detection │
│ GPS processing │
│ Event generation │
└───────────┬───────────┘
│
Wi-Fi
│
▼
┌───────────────────────┐
│ n8n WEBHOOK │
│ │
│ Receive IoT JSON │
│ Validate data │
└───────────┬───────────┘
│
▼
┌───────────────────────┐
│ AI AGENT │
│ │
│ Accident assessment │
│ Severity classification│
│ Response generation │
└──────┬─────┬─────┬────┘
│ │ │
┌────────────┘ │ └─────────────┐
▼ ▼ ▼
┌────────────────┐ ┌───────────────┐ ┌──────────────────┐
│ Google Sheets │ │ ThingSpeak │ │ Telegram │
│ Event database │ │ Cloud graphs │ │ Text + Voice │
└────────────────┘ └───────────────┘ └──────────────────┘
5. Hardware requirements
Required components
| Component | Purpose |
|---|---|
| ESP32 DevKit | Main IoT controller |
| MPU6050 | Accelerometer + gyroscope |
| NEO-6M GPS | Location and speed |
| Buzzer | Local accident warning |
| Push button | Manual SOS |
| LED | Status indication |
| Breadboard | Prototyping |
| Jumper wires | Connections |
| 5 V power source | Vehicle/project power |
| USB cable | Programming |
Optional components
-
OLED display
-
vibration sensor
-
temperature sensor
-
current sensor
-
GSM/LTE module
-
SD card
-
camera
-
ESP32-CAM
-
relay
-
emergency cancellation button
6. Recommended hardware architecture
+----------------------+
| VEHICLE |
+----------------------+
+----------------+
| MPU6050 |
| Accel + Gyro |
+-------+--------+
|
I2C |
|
v
+-------------+
| ESP32 |
| |
| Wi-Fi |
| Processing |
+------+------+
|
+----------+-----------+
| |
UART GPS GPIO
| |
v v
+-------------+ +-------------+
| NEO-6M | | SOS Button |
| GPS Module | +-------------+
+-------------+
|
v
Buzzer
7. Schematic diagram
A simple prototype wiring can be arranged as follows.
MPU6050 → ESP32
| MPU6050 | ESP32 |
|---|---|
| VCC | 3.3 V |
| GND | GND |
| SDA | GPIO 21 |
| SCL | GPIO 22 |
GPS → ESP32
| NEO-6M | ESP32 |
|---|---|
| VCC | Appropriate module supply |
| GND | GND |
| TX | GPIO 16 |
| RX | GPIO 17 |
Use a proper voltage level arrangement for the particular GPS module you purchase.
Buzzer
ESP32 GPIO 25
|
+---- Buzzer
|
GND
For a higher-current buzzer, drive it through a transistor rather than directly from the ESP32 GPIO.
SOS button
GPIO 27
|
+-------- Push Button -------- GND
Configure the pin with INPUT_PULLUP.
8. Complete electrical block diagram
+-------------------+
| 5V INPUT |
+---------+---------+
|
+------+------+
| ESP32 |
| |
| 3.3V |
+--+----------+
|
+----------+----------+
| |
v v
+---------+ +---------+
| MPU6050 | | GPS |
| I2C | | NEO-6M |
+---------+ +---------+
| |
| |
+----------+----------+
|
v
Sensor Processing
|
v
Accident?
/ \
NO YES
| |
| v
| Create Event
| |
+-----------+
|
v
Wi-Fi Upload
|
v
n8n Webhook
9. How accident detection works
The MPU6050 provides:
-
X acceleration
-
Y acceleration
-
Z acceleration
-
X angular velocity
-
Y angular velocity
-
Z angular velocity
Acceleration magnitude can be calculated as:
A=Ax2+Ay2+Az2A=\sqrt{A_x^2+A_y^2+A_z^2}
During stationary conditions, the acceleration magnitude is approximately close to:
1g≈9.81m/s21g \approx 9.81m/s^2
A collision can produce a sudden acceleration spike.
However, do not use a single acceleration threshold as a production accident detector.
A better prototype algorithm combines:
Acceleration spike
+
Gyroscope spike
+
Sudden change in motion
+
Vehicle speed/GPS state
+
Short confirmation window
Example:
Acceleration > threshold
|
v
Possible impact
|
v
Check gyro
|
v
Check GPS speed
|
v
Calculate confidence
|
v
Accident confidence > 70% ?
/ \
NO YES
| |
Normal event ACCIDENT
|
v
Send emergency
10. Accident confidence calculation
For a prototype you can use:
Acceleration score = 40%
Gyroscope score = 25%
Speed score = 20%
Motion change = 15%
Example:
acceleration = 85%
gyro = 70%
speed = 80%
motion = 90%
confidence =
0.40(85) +
0.25(70) +
0.20(80) +
0.15(90)
confidence = 81.75%
The ESP32 can classify:
0–39% → NORMAL
40–69% → SUSPICIOUS
70–100% → POSSIBLE ACCIDENT
For a student prototype, these thresholds should be experimentally calibrated rather than presented as medically or automotive-certified thresholds.
11. GPS tracking
The GPS module supplies:
{
"latitude": 17.3850,
"longitude": 78.4867,
"speed_kmph": 42.5
}
The coordinates can be converted into a map URL:
https://www.google.com/maps?q=17.3850,78.4867
Your Telegram alert can therefore contain:
🚨 POSSIBLE ACCIDENT
Vehicle: CAR-001
Location:
17.3850, 78.4867
Speed:
42.5 km/h
Map:
https://www.google.com/maps?q=17.3850,78.4867
In the actual implementation, n8n should construct the map URL dynamically.
12. ESP32-to-n8n communication
The ESP32 sends JSON.
Example:
{
"device_id": "CAR-001",
"event": "ACCIDENT",
"timestamp": 1727979000,
"accel_x": 3.21,
"accel_y": 2.75,
"accel_z": 16.42,
"accel_magnitude": 17.01,
"gyro_x": 12.4,
"gyro_y": 9.8,
"gyro_z": 21.3,
"latitude": 17.385044,
"longitude": 78.486671,
"speed_kmph": 58.2,
"accident_confidence": 86.4
}
13. n8n architecture
n8n is particularly suitable because it connects APIs, applications and AI workflows. n8n documents built-in Telegram functionality and AI capabilities. n8n Docs+1
The main workflow:
ESP32
|
| HTTP POST
v
Webhook
|
v
Validate JSON
|
v
Normalize Data
|
+---------------------+
| |
v v
ThingSpeak Google Sheets
|
v
AI Agent
|
v
Severity decision
|
+----------------------+
| |
NORMAL ACCIDENT
| |
v v
Log only Telegram text
|
v
Generate voice
|
v
Telegram voice
|
v
Send GPS location
14. n8n workflow nodes
Create the following nodes:
01 Webhook
↓
02 Code - Validate Payload
↓
03 IF - Accident?
↓
04 Google Sheets
↓
05 ThingSpeak HTTP Request
↓
06 AI Agent
↓
07 IF - Emergency?
↓
08 Telegram Text
↓
09 Text-to-Speech
↓
10 Telegram Voice
↓
11 Telegram Location
You can also split this into two workflows:
Workflow A — telemetry
ESP32
↓
Webhook
↓
Validation
↓
ThingSpeak
↓
Google Sheets
Workflow B — emergency
ESP32 Accident Event
↓
Webhook
↓
AI Agent
↓
Severity
↓
Telegram
↓
Voice
↓
Location
That architecture is easier to maintain.
15. n8n Webhook
Create:
Node: Webhook
Method:
POST
Example endpoint:
/webhook/vehicle-alert
ESP32 sends:
POST https://YOUR-N8N-DOMAIN/webhook/vehicle-alert
Content-Type: application/json
with the JSON payload.
Do not expose an unprotected production webhook. Use authentication, a secret token/signature, rate limiting and HTTPS. n8n itself provides security auditing functionality that can identify issues such as unprotected webhooks. n8n Docs
16. n8n validation node
Use a Code node after the webhook.
Example:
const d = $json.body ?? $json;
const required = [
"device_id",
"latitude",
"longitude",
"accident_confidence"
];
for (const field of required) {
if (d[field] === undefined || d[field] === null) {
throw new Error(`Missing field: ${field}`);
}
}
return [{
json: {
device_id: String(d.device_id),
event: d.event || "TELEMETRY",
latitude: Number(d.latitude),
longitude: Number(d.longitude),
speed_kmph: Number(d.speed_kmph || 0),
accel_x: Number(d.accel_x || 0),
accel_y: Number(d.accel_y || 0),
accel_z: Number(d.accel_z || 0),
accel_magnitude: Number(d.accel_magnitude || 0),
gyro_x: Number(d.gyro_x || 0),
gyro_y: Number(d.gyro_y || 0),
gyro_z: Number(d.gyro_z || 0),
accident_confidence:
Number(d.accident_confidence || 0),
map_url:
`https://www.google.com/maps?q=${Number(d.latitude)},${Number(d.longitude)}`,
received_at: new Date().toISOString()
}
}];
n8n's Code node is intended for data transformation and logic within workflows. n8n Docs
17. Google Sheets database
Create a spreadsheet called:
AI Vehicle Accident Monitoring
Create columns:
| Column | Description |
|---|---|
| Timestamp | Event time |
| Device ID | Vehicle ID |
| Event | NORMAL/ACCIDENT |
| Latitude | GPS latitude |
| Longitude | GPS longitude |
| Speed | km/h |
| Accel X | X acceleration |
| Accel Y | Y acceleration |
| Accel Z | Z acceleration |
| Accel Magnitude | Total acceleration |
| Gyro X | X rotation |
| Gyro Y | Y rotation |
| Gyro Z | Z rotation |
| Confidence | Accident confidence |
| Severity | AI classification |
| AI Analysis | Explanation |
| Notification | Sent/Failed |
n8n has a Google Sheets integration available for document/sheet operations. n8n Docs
18. ThingSpeak configuration
Create a ThingSpeak channel:
Channel name:
AI Vehicle Accident Monitoring
Suggested fields:
Field 1 = Acceleration
Field 2 = Gyroscope
Field 3 = Speed
Field 4 = Accident Confidence
Field 5 = Latitude
Field 6 = Longitude
Field 7 = Accident Status
Field 8 = Battery Voltage
ThingSpeak supports REST-based channel updates through api.thingspeak.com/update, including fields, latitude and longitude. MathWorks+1
Example:
https://api.thingspeak.com/update
Parameters:
api_key = YOUR_WRITE_API_KEY
field1 = 17.01
field2 = 21.3
field3 = 58.2
field4 = 86.4
field5 = 17.385044
field6 = 78.486671
field7 = 1
19. n8n ThingSpeak HTTP Request
Use:
Node: HTTP Request
Method:
POST
URL:
https://api.thingspeak.com/update.json
Body:
api_key={{ $env.THINGSPEAK_WRITE_KEY }}
field1={{ $json.accel_magnitude }}
field2={{ $json.gyro_z }}
field3={{ $json.speed_kmph }}
field4={{ $json.accident_confidence }}
field5={{ $json.latitude }}
field6={{ $json.longitude }}
field7={{ $json.event === "ACCIDENT" ? 1 : 0 }}
ThingSpeak returns an entry ID when the update succeeds and 0 on failure. MathWorks
20. AI Agent architecture
This is where the project becomes an Agentic IoT system instead of merely an IoT notification system.
The AI Agent receives:
Sensor data
+
GPS data
+
Vehicle state
+
Accident confidence
and determines:
Is this probably an accident?
What is the severity?
What action should be taken?
What message should be sent?
21. AI Agent prompt
Use a prompt similar to this:
You are an IoT Vehicle Safety AI Agent.
You receive telemetry from an ESP32 vehicle monitoring device.
Analyze:
- acceleration
- gyroscope
- speed
- GPS position
- accident confidence
- event type
Classify the event as one of:
NORMAL
SUSPICIOUS
ACCIDENT
If the event is an accident, classify severity:
LOW
MEDIUM
HIGH
CRITICAL
Rules:
1. Never claim that an accident is medically confirmed.
2. Treat sensor detection as a possible accident.
3. High acceleration combined with abnormal rotation increases accident likelihood.
4. A vehicle moving at significant speed before a large impact should increase severity.
5. If confidence is low, recommend monitoring rather than emergency escalation.
6. Always provide a concise emergency message.
7. Include GPS coordinates.
8. Include a Google Maps URL.
Return JSON only.
Expected output:
{
"classification": "ACCIDENT",
"severity": "HIGH",
"confidence": 0.91,
"reason": "Large acceleration spike combined with abnormal rotational motion.",
"action": "SEND_EMERGENCY_ALERT",
"telegram_message": "Possible high-severity vehicle accident detected.",
"voice_message": "Emergency alert. A possible high-severity accident has been detected. Vehicle CAR-001 is located at the reported GPS position."
}
22. Important AI design principle
Do not allow the AI Agent to be the only accident detector.
Use:
ESP32 deterministic detection
+
AI interpretation
rather than:
ESP32 → AI decides everything
Why?
Because Internet connectivity or AI response time could fail immediately after an accident.
The ESP32 should therefore detect the event locally and store/queue the event if necessary.
23. Agentic decision architecture
SENSOR DATA
|
v
+---------------+
| ESP32 Rules |
+-------+-------+
|
v
Possible Accident
|
v
n8n Webhook
|
v
+---------------+
| AI AGENT |
+-------+-------+
|
+----------+----------+
| | |
v v v
NORMAL SUSPICIOUS ACCIDENT
| | |
| | v
| | Severity
| | |
| | v
| | Take Action
| | |
+----------+----------+
|
v
Automation Tools
/ | \
/ | \
v v v
Sheets ThingSpeak Telegram
24. Telegram Bot
Create a Telegram bot using Telegram's official bot creation mechanism.
Obtain:
BOT_TOKEN
and determine the target:
CHAT_ID
Keep the token secret.
n8n provides a Telegram node with operations for sending messages, audio, locations and other Telegram content. n8n Docs
25. Telegram emergency message
Example:
🚨 VEHICLE ACCIDENT ALERT 🚨
Vehicle: CAR-001
Possible accident detected.
Severity: HIGH
Confidence: 91%
Speed: 58.2 km/h
Acceleration: 17.01 m/s²
Location:
17.385044, 78.486671
Open location:
https://www.google.com/maps?q=17.385044,78.486671
AI assessment:
Large acceleration spike combined with abnormal rotational motion.
Please check the vehicle immediately.
26. Telegram voice alert
The workflow should generate:
"Emergency alert. A possible high severity accident has been detected. Vehicle CAR-001 is currently at the reported GPS location. Please check the vehicle immediately."
Then convert the text to speech.
The resulting audio is passed to Telegram as a voice message.
Telegram's Bot API distinguishes voice messages from ordinary audio files and provides the sendVoice method for voice messages. Telegram
27. Voice workflow
AI Agent
|
v
voice_message
|
v
Text-to-Speech API
|
v
MP3/OGG audio
|
v
n8n Binary Data
|
v
Telegram Send Voice
|
v
Emergency recipient
Depending on the TTS service and Telegram integration version, you may use either a built-in n8n audio capability or an HTTP Request node to a TTS API.
28. Telegram location
After the text alert, send the GPS location.
Telegram
|
+-- Send Message
|
+-- Send Voice
|
+-- Send Location
Latitude:
{{ $json.latitude }}
Longitude:
{{ $json.longitude }}
This makes the alert much more useful than sending coordinates as plain text.
29. ESP32 firmware
Below is a prototype firmware implementation using:
-
ESP32
-
MPU6050
-
TinyGPS++
-
Wi-Fi
-
HTTPClient
-
JSON payload
-
local accident detection
The ESP32 Arduino Wi-Fi and HTTPClient libraries support connecting to an access point and making HTTP requests. Espressif Systems+1
Arduino libraries
Install:
Adafruit MPU6050
Adafruit Unified Sensor
TinyGPSPlus
ArduinoJson
30. ESP32 code
#include <WiFi.h>
#include <HTTPClient.h>
#include <Wire.h>
#include <Adafruit_MPU6050.h>
#include <Adafruit_Sensor.h>
#include <TinyGPSPlus.h>
#include <ArduinoJson.h>
// =====================================================
// WIFI
// =====================================================
const char* WIFI_SSID = "YOUR_WIFI";
const char* WIFI_PASSWORD = "YOUR_PASSWORD";
// n8n production webhook
const char* N8N_WEBHOOK =
"https://YOUR-N8N-DOMAIN/webhook/vehicle-alert";
// =====================================================
// DEVICE
// =====================================================
const char* DEVICE_ID = "CAR-001";
// =====================================================
// GPS
// =====================================================
HardwareSerial GPSSerial(2);
#define GPS_RX 16
#define GPS_TX 17
TinyGPSPlus gps;
// =====================================================
// MPU6050
// =====================================================
Adafruit_MPU6050 mpu;
// =====================================================
// GPIO
// =====================================================
#define BUZZER_PIN 25
#define SOS_PIN 27
#define LED_PIN 2
// =====================================================
// TIMING
// =====================================================
unsigned long lastTelemetry = 0;
const unsigned long TELEMETRY_INTERVAL = 5000;
// =====================================================
// ACCIDENT PARAMETERS
// =====================================================
// Prototype values only.
// Calibrate using controlled experiments.
const float ACCEL_THRESHOLD = 18.0;
const float GYRO_THRESHOLD = 15.0;
const float SPEED_THRESHOLD = 20.0;
// =====================================================
// WIFI
// =====================================================
void connectWiFi()
{
Serial.print("Connecting to WiFi");
WiFi.begin(WIFI_SSID, WIFI_PASSWORD);
int attempts = 0;
while (WiFi.status() != WL_CONNECTED &&
attempts < 30)
{
delay(500);
Serial.print(".");
attempts++;
}
Serial.println();
if (WiFi.status() == WL_CONNECTED)
{
Serial.println("WiFi connected");
Serial.print("IP: ");
Serial.println(WiFi.localIP());
}
else
{
Serial.println("WiFi connection failed");
}
}
// =====================================================
// GPS UPDATE
// =====================================================
void updateGPS()
{
while (GPSSerial.available())
{
gps.encode(GPSSerial.read());
}
}
// =====================================================
// SEND EVENT TO N8N
// =====================================================
bool sendToN8N(
String eventType,
float ax,
float ay,
float az,
float acceleration,
float gx,
float gy,
float gz,
float speed,
float confidence
)
{
if (WiFi.status() != WL_CONNECTED)
{
Serial.println("WiFi unavailable");
return false;
}
HTTPClient http;
http.begin(N8N_WEBHOOK);
http.addHeader(
"Content-Type",
"application/json"
);
float latitude = 0;
float longitude = 0;
if (gps.location.isValid())
{
latitude = gps.location.lat();
longitude = gps.location.lng();
}
StaticJsonDocument<1024> doc;
doc["device_id"] = DEVICE_ID;
doc["event"] = eventType;
doc["timestamp"] = millis();
doc["accel_x"] = ax;
doc["accel_y"] = ay;
doc["accel_z"] = az;
doc["accel_magnitude"] = acceleration;
doc["gyro_x"] = gx;
doc["gyro_y"] = gy;
doc["gyro_z"] = gz;
doc["speed_kmph"] = speed;
doc["latitude"] = latitude;
doc["longitude"] = longitude;
doc["gps_valid"] = gps.location.isValid();
doc["accident_confidence"] = confidence;
String payload;
serializeJson(doc, payload);
Serial.println("Sending:");
Serial.println(payload);
int httpCode = http.POST(payload);
Serial.print("HTTP response: ");
Serial.println(httpCode);
http.end();
return httpCode >= 200 &&
httpCode < 300;
}
// =====================================================
// CALCULATE ACCELERATION
// =====================================================
float calculateAcceleration(
sensors_event_t& accel
)
{
return sqrt(
accel.acceleration.x *
accel.acceleration.x +
accel.acceleration.y *
accel.acceleration.y +
accel.acceleration.z *
accel.acceleration.z
);
}
// =====================================================
// ACCIDENT CONFIDENCE
// =====================================================
float calculateConfidence(
float acceleration,
float gyro,
float speed
)
{
float score = 0;
// Acceleration contribution
if (acceleration > ACCEL_THRESHOLD)
score += 40;
// Gyroscope contribution
if (gyro > GYRO_THRESHOLD)
score += 30;
// Speed contribution
if (speed > SPEED_THRESHOLD)
score += 20;
// Combined condition
if (acceleration > ACCEL_THRESHOLD &&
gyro > GYRO_THRESHOLD)
{
score += 10;
}
if (score > 100)
score = 100;
return score;
}
// =====================================================
// BUZZER
// =====================================================
void accidentAlarm()
{
digitalWrite(LED_PIN, HIGH);
for (int i = 0; i < 5; i++)
{
digitalWrite(BUZZER_PIN, HIGH);
delay(200);
digitalWrite(BUZZER_PIN, LOW);
delay(200);
}
digitalWrite(LED_PIN, LOW);
}
// =====================================================
// SETUP
// =====================================================
void setup()
{
Serial.begin(115200);
pinMode(BUZZER_PIN, OUTPUT);
pinMode(SOS_PIN, INPUT_PULLUP);
pinMode(LED_PIN, OUTPUT);
digitalWrite(BUZZER_PIN, LOW);
digitalWrite(LED_PIN, LOW);
Wire.begin(21, 22);
// MPU6050
if (!mpu.begin())
{
Serial.println(
"MPU6050 not found!"
);
while (true)
{
delay(1000);
}
}
Serial.println(
"MPU6050 initialized"
);
mpu.setAccelerometerRange(
MPU6050_RANGE_8_G
);
mpu.setGyroRange(
MPU6050_RANGE_500_DEG
);
// GPS
GPSSerial.begin(
9600,
SERIAL_8N1,
GPS_RX,
GPS_TX
);
connectWiFi();
}
// =====================================================
// LOOP
// =====================================================
void loop()
{
updateGPS();
// Manual SOS
if (digitalRead(SOS_PIN) == LOW)
{
Serial.println("SOS BUTTON");
accidentAlarm();
sendToN8N(
"MANUAL_SOS",
0,
0,
0,
0,
0,
0,
0,
gps.speed.isValid()
? gps.speed.kmph()
: 0,
100
);
delay(3000);
}
if (millis() -
lastTelemetry <
TELEMETRY_INTERVAL)
{
return;
}
lastTelemetry = millis();
sensors_event_t accel;
sensors_event_t gyro;
sensors_event_t temp;
mpu.getEvent(
&accel,
&gyro,
&temp
);
float acceleration =
calculateAcceleration(accel);
float gyroMagnitude =
sqrt(
gyro.gyro.x *
gyro.gyro.x +
gyro.gyro.y *
gyro.gyro.y +
gyro.gyro.z *
gyro.gyro.z
);
float speed =
gps.speed.isValid()
? gps.speed.kmph()
: 0;
float confidence =
calculateConfidence(
acceleration,
gyroMagnitude,
speed
);
String eventType =
confidence >= 70
? "ACCIDENT"
: "TELEMETRY";
Serial.println("------------------");
Serial.print("Acceleration: ");
Serial.println(acceleration);
Serial.print("Gyro: ");
Serial.println(gyroMagnitude);
Serial.print("Speed: ");
Serial.println(speed);
Serial.print("Confidence: ");
Serial.println(confidence);
Serial.print("Event: ");
Serial.println(eventType);
if (eventType == "ACCIDENT")
{
accidentAlarm();
}
sendToN8N(
eventType,
accel.acceleration.x,
accel.acceleration.y,
accel.acceleration.z,
acceleration,
gyro.gyro.x,
gyro.gyro.y,
gyro.gyro.z,
speed,
confidence
);
}
31. Important improvement: don't send every event as an accident
A real implementation should have a state machine.
NORMAL
|
| impact detected
v
POSSIBLE_IMPACT
|
| confirmation
v
ACCIDENT_PENDING
|
| confirmed
v
ACCIDENT
|
| alert sent
v
ALERTED
|
| reset
v
NORMAL
This prevents multiple Telegram alerts for the same accident.
32. Better accident algorithm
Use a sliding window.
For example:
Sample rate = 50 Hz
Maintain last 2 seconds:
100 sensor samples
Calculate:
maximum acceleration
maximum gyro
change in acceleration
change in orientation
vehicle speed
Then:
IF
maxAcceleration > threshold
AND
maxGyro > threshold
THEN
possible accident
After that:
Wait 1–3 seconds
IF movement remains abnormal
OR
second sensor condition confirms impact
THEN
ACCIDENT
This is significantly better than a single sensor reading.
33. n8n AI workflow in detail
Create the workflow:
┌──────────────┐
│ Webhook │
└──────┬───────┘
│
▼
┌──────────────────┐
│ Validate Payload │
└────────┬─────────┘
│
▼
┌──────────────────┐
│ Prepare Location │
└────────┬─────────┘
│
├─────────────────┐
│ │
▼ ▼
┌──────────────┐ ┌───────────────┐
│ Google Sheets│ │ ThingSpeak │
└──────────────┘ └───────────────┘
│
▼
┌─────────────────┐
│ AI Agent │
└────────┬────────┘
│
▼
┌─────────────────┐
│ Parse AI Result │
└────────┬────────┘
│
▼
┌───────────────┐
│ Severity? │
└──────┬────────┘
│
HIGH/CRITICAL
│
▼
┌───────────────┐
│ Telegram Text │
└──────┬────────┘
│
▼
┌───────────────┐
│ Text to Speech│
└──────┬────────┘
│
▼
┌───────────────┐
│Telegram Voice │
└──────┬────────┘
│
▼
┌───────────────┐
│Telegram GPS │
│Location │
└───────────────┘
34. Google Sheets record
The n8n Google Sheets node should append something like:
2026-10-04 21:45:23
CAR-001
ACCIDENT
17.385044
78.486671
58.2
3.21
2.75
16.42
17.01
12.4
9.8
21.3
86.4
HIGH
Large acceleration + abnormal rotation
SENT
This gives you a permanent project log.
35. ThingSpeak dashboard
Configure charts for:
Chart 1
Acceleration vs Time
Chart 2
Vehicle Speed vs Time
Chart 3
Accident Confidence vs Time
Chart 4
Gyroscope vs Time
Map
Use:
Latitude
Longitude
ThingSpeak supports channel data visualization and map-related channel functionality through its APIs/platform. MathWorks+1
36. Complete data flow
VEHICLE
|
+-----------+-----------+
| |
v v
MPU6050 GPS
| |
+-----------+-----------+
|
v
ESP32
|
Accident Algorithm
|
+-----------+-----------+
| |
NORMAL ACCIDENT
| |
+-----------+-----------+
|
v
Wi-Fi
|
v
n8n Webhook
|
v
Data Validation
|
+--------+--------+
| |
v v
ThingSpeak Google Sheets
| |
+--------+--------+
|
v
AI Agent
|
+--------+---------+
| | |
v v v
LOW MEDIUM HIGH
| | |
| | v
| | Telegram
| | |
| | +---+---+
| | | |
| | v v
| | Text Voice
| |
| v
| Log
|
v
Log
37. Telegram conversation example
Accident event
System → Telegram
🚨 VEHICLE ACCIDENT ALERT
System:
Vehicle: CAR-001
Status: Possible Accident
Severity: HIGH
Confidence: 91%
Speed: 58.2 km/h
System:
📍 Location: 17.385044, 78.486671
System:
🗺 Open vehicle location
System:
Sensor analysis indicates a large acceleration spike combined with abnormal rotational motion.
System → Voice
"Emergency alert. A possible high-severity accident has been detected. Vehicle CAR-001 is currently at the reported GPS location. Please check the vehicle immediately."
38. Manual SOS operation
The project should also have a manual emergency button.
Driver presses SOS
|
v
ESP32 detects button
|
v
Generate MANUAL_SOS event
|
v
n8n
|
v
AI Agent
|
v
Telegram
|
+---- Text
|
+---- Voice
|
+---- Location
This is useful even if no accident occurs.
For example:
Medical emergency
Vehicle breakdown
Threat/security problem
Driver assistance
39. Vehicle tracking mode
Apart from accident detection, send periodic telemetry.
For example:
Every 5 seconds:
GPS
Speed
Acceleration
Gyroscope
Battery
The system becomes:
Vehicle
|
v
ESP32
|
v
n8n
|
+---- ThingSpeak
|
+---- Google Sheets
ThingSpeak's REST API is designed for reading and writing channel data, so it is suitable for this telemetry layer. MathWorks
40. Recommended ThingSpeak fields
Use:
FIELD 1 → Acceleration
FIELD 2 → Gyroscope
FIELD 3 → Speed
FIELD 4 → Accident Confidence
FIELD 5 → Latitude
FIELD 6 → Longitude
FIELD 7 → Accident Flag
FIELD 8 → Battery
Example:
Field 1 = 17.01
Field 2 = 21.30
Field 3 = 58.20
Field 4 = 86.40
Field 5 = 17.385044
Field 6 = 78.486671
Field 7 = 1
Field 8 = 3.92
41. Security architecture
Do not hard-code all production secrets directly into firmware.
Avoid:
const char* API_KEY = "my-secret-key";
when the code will be published.
Instead use:
ESP32
|
| device authentication
v
n8n
|
+-- Telegram credential
+-- Google credential
+-- ThingSpeak key
+-- AI API credential
+-- TTS credential
n8n credentials should be stored in n8n rather than exposed in the ESP32 payload.
42. Recommended authentication
Add an authentication header:
X-DEVICE-TOKEN: YOUR_DEVICE_SECRET
ESP32:
http.addHeader(
"X-DEVICE-TOKEN",
DEVICE_SECRET
);
n8n validation:
const token =
$headers["x-device-token"];
if (token !== $env.DEVICE_SECRET) {
throw new Error("Unauthorized device");
}
For production, consider stronger mechanisms such as signed requests, rotating credentials and HTTPS certificate validation.
43. Failure handling
The system should be designed around failures.
Case 1 — Wi-Fi unavailable
ESP32
↓
No Wi-Fi
↓
Store event locally
↓
Reconnect
↓
Upload later
Add EEPROM/NVS or SD storage for queued events.
Case 2 — n8n unavailable
ESP32
↓
HTTP failure
↓
Save event
↓
Retry
Case 3 — Telegram unavailable
n8n
↓
Telegram error
↓
Log failure
↓
Retry
Case 4 — GPS unavailable
Use:
"gps_valid": false
and send:
GPS unavailable
Last known location:
...
Case 5 — AI unavailable
The workflow should still send a deterministic alert:
ESP32 confidence > threshold
|
v
AI unavailable
|
v
Fallback emergency notification
This is extremely important.
44. AI fallback
Use an n8n IF node:
AI Agent
|
+---- success → AI decision
|
+---- error → deterministic decision
Fallback:
const confidence =
Number($json.accident_confidence || 0);
let severity = "LOW";
if (confidence >= 90) {
severity = "CRITICAL";
}
else if (confidence >= 80) {
severity = "HIGH";
}
else if (confidence >= 70) {
severity = "MEDIUM";
}
return [{
json: {
...$json,
severity,
ai_status: "FALLBACK"
}
}];
45. State diagram
+---------+
| START |
+----+----+
|
v
+---------+
| NORMAL |
+----+----+
|
Impact detected
|
v
+----------------+
| POSSIBLE IMPACT|
+-------+--------+
|
Confirm sensors
|
+------+------+
| |
NO YES
| |
v v
NORMAL ACCIDENT
|
v
SEND EVENT
|
v
AI ANALYSIS
|
+---------+---------+
| |
LOW HIGH
| |
v v
LOG ALERT
|
+------------+------------+
| | |
v v v
TEXT VOICE GPS
| | |
+------------+-------------+
|
v
ALERTED
|
v
RESET
|
v
NORMAL
46. Software architecture
+-----------------------------------------------------+
| SOFTWARE |
+-----------------------------------------------------+
| |
| Arduino IDE |
| | |
| v |
| ESP32 Firmware |
| | |
| +---- MPU6050 driver |
| +---- GPS driver |
| +---- Accident algorithm |
| +---- Wi-Fi |
| +---- HTTP/JSON |
| |
+-----------------------------------------------------+
INTERNET
|
v
+-----------------------------------------------------+
| n8n |
+-----------------------------------------------------+
| |
| Webhook |
| | |
| Validation |
| | |
| Data transformation |
| | |
| AI Agent |
| | |
| +----+----------+-------------+ |
| | | | |
| v v v |
| Sheets ThingSpeak Telegram |
| |
+-----------------------------------------------------+
47. AI Agent tools
A more advanced version can give the AI Agent tools such as:
Tool 1:
Get latest vehicle telemetry
Tool 2:
Get previous accident records
Tool 3:
Write incident to Google Sheets
Tool 4:
Send Telegram alert
Tool 5:
Send vehicle location
Tool 6:
Get ThingSpeak history
Then the AI Agent becomes:
AI AGENT
|
+------------+-------------+
| | |
v v v
Telemetry Incident Notification
Tool History Tool Tool
| | |
+------------+-------------+
|
v
Decision
n8n's AI tooling is designed to allow integrations and tools to participate in AI workflows. n8n Docs
48. Example AI reasoning
Input:
{
"speed_kmph": 72,
"accel_magnitude": 24.2,
"gyro_z": 32.1,
"accident_confidence": 94
}
AI response:
{
"classification": "ACCIDENT",
"severity": "CRITICAL",
"confidence": 0.96,
"reason": "High-speed vehicle combined with a large acceleration spike and extreme rotational movement.",
"action": "SEND_EMERGENCY_ALERT"
}
The automation then executes:
Send Telegram
Send Voice
Send Location
Write Sheet
Update ThingSpeak
49. Project flowchart
START
|
v
Initialize ESP32
|
v
Initialize MPU6050
|
v
Initialize GPS
|
v
Connect Wi-Fi
|
v
Read sensor data
|
v
Calculate motion
|
v
Calculate speed
|
v
Accident detected?
/ \
NO YES
| |
v v
Send telemetry Activate buzzer
| |
| v
| Create event
| |
+------+-------+
|
v
Send to n8n
|
v
AI Agent analysis
|
v
Determine severity
|
v
Log Google Sheets
|
v
Update ThingSpeak
|
v
Emergency alert?
/ \
NO YES
| |
v v
Finish Telegram text
|
v
Voice message
|
v
GPS location
|
v
END
50. Complete technology stack
| Layer | Technology |
|---|---|
| Controller | ESP32 |
| Motion sensor | MPU6050 |
| Location | NEO-6M GPS |
| Programming | Arduino C++ |
| Connectivity | Wi-Fi |
| API protocol | HTTP/JSON |
| Automation | n8n |
| AI | n8n AI Agent + LLM |
| Notification | Telegram |
| Voice | TTS |
| Database/logging | Google Sheets |
| IoT dashboard | ThingSpeak |
| Mapping | Google Maps URL |
| Cloud workflow | n8n |
| Visualization | ThingSpeak |
51. Required n8n credentials
You will need credentials for:
1. AI/LLM provider
2. Telegram Bot
3. Google Sheets
4. TTS provider
5. ThingSpeak API key
ThingSpeak uses channel-specific write API keys for channel updates. MathWorks
52. n8n environment variables
For a self-hosted deployment, conceptually maintain:
DEVICE_SECRET
THINGSPEAK_WRITE_KEY
TELEGRAM_CHAT_ID
N8N_WEBHOOK_URL
API credentials should preferably be stored in the credential manager rather than ordinary workflow fields.
53. Testing procedure
Do not begin by simulating a real road accident.
Use controlled tests.
Test 1 — Normal operation
Move the MPU6050 gently.
Expected:
EVENT = TELEMETRY
ACCIDENT = FALSE
Test 2 — GPS
Move the GPS outdoors.
Expected:
GPS valid = true
latitude ≠ 0
longitude ≠ 0
Test 3 — Manual SOS
Press the button.
Expected:
ESP32
↓
n8n
↓
Google Sheets
↓
Telegram text
↓
Telegram voice
↓
GPS location
Test 4 — Artificial impact
Perform a safe controlled sensor test.
Expected:
Acceleration spike
+
Gyroscope spike
↓
Possible accident
Test 5 — Wi-Fi failure
Turn off Wi-Fi.
Expected:
ESP32 detects failure
and, if local queueing has been implemented:
Event stored
Test 6 — n8n failure
Stop n8n.
Expected:
HTTP request fails
and the ESP32 should not crash.
Test 7 — Telegram failure
Disable Telegram credentials temporarily.
Expected:
Incident remains in Google Sheets
and failure is logged.
54. Expected project output
When an accident is detected:
ESP32
ACCIDENT DETECTED
Confidence: 91%
n8n
Webhook received
Data validated
AI analysis completed
Severity = HIGH
Google Sheets
Incident record inserted
ThingSpeak
Telemetry updated
Telegram
🚨 VEHICLE ACCIDENT ALERT
followed by:
🔊 Voice alert
and:
📍 Vehicle location
55. Example complete incident record
{
"device_id": "CAR-001",
"event": "ACCIDENT",
"timestamp": "2026-10-04T16:15:22Z",
"sensor": {
"accel_x": 3.21,
"accel_y": 2.75,
"accel_z": 16.42,
"magnitude": 17.01,
"gyro_x": 12.4,
"gyro_y": 9.8,
"gyro_z": 21.3
},
"vehicle": {
"speed_kmph": 58.2
},
"gps": {
"latitude": 17.385044,
"longitude": 78.486671
},
"analysis": {
"confidence": 0.91,
"classification": "ACCIDENT",
"severity": "HIGH"
},
"notifications": {
"telegram_text": true,
"telegram_voice": true,
"location": true
}
}
56. Advantages
Hardware advantages
-
Low-cost
-
Compact
-
Wi-Fi enabled
-
Easy to program
-
Expandable
Software advantages
-
n8n provides visual automation
-
AI adds contextual analysis
-
Google Sheets is easy to inspect
-
ThingSpeak provides visualization
-
Telegram provides instant notification
AI advantages
The AI can interpret several sensor values simultaneously instead of relying on one threshold.
57. Limitations
This is important for your project report.
The system is a prototype and not a certified automotive safety system.
Potential limitations include:
-
GPS may be unavailable indoors.
-
GPS location can have several meters of error.
-
Wi-Fi may not be available everywhere.
-
MPU6050 readings depend on mounting orientation.
-
Sensor thresholds require calibration.
-
False positives are possible.
-
False negatives are possible.
-
AI decisions can be imperfect.
-
Internet latency can delay cloud alerts.
-
Telegram requires Internet access.
-
The system should not replace certified vehicle safety equipment or emergency services.
58. Future enhancements
You can list these in your project presentation.
1. GSM/LTE
Add:
SIM7600 / LTE module
so alerts can work without Wi-Fi.
2. Camera
Add:
ESP32-CAM
or another camera to capture accident images.
3. Cloud database
Replace Google Sheets with:
PostgreSQL
Supabase
Firebase
MongoDB
4. Advanced ML model
Train an accident classifier using:
Acceleration
Gyroscope
Speed
Orientation
Time-series windows
5. Driver monitoring
Add:
Camera
Drowsiness detection
Face detection
Eye closure detection
6. OBD-II
Read:
Vehicle speed
RPM
Engine temperature
Diagnostic codes
7. Multi-vehicle fleet
Architecture:
CAR-001 ─┐
CAR-002 ─┤
CAR-003 ─┼──> n8n ──> AI Agent
CAR-004 ─┤
CAR-005 ─┘
8. Emergency-service integration
Future version could integrate authorized emergency-response APIs.
59. Multi-vehicle architecture
VEHICLE 1
ESP32 #001
|
|
VEHICLE 2
ESP32 #002
|
|
VEHICLE 3
ESP32 #003
|
|
v
+--------------+
| n8n |
| Central IoT |
+------+-------+
|
+--------+---------+
| | |
v v v
AI Database Alerts
| | |
v v v
Analysis Sheets Telegram
|
v
Operator
60. Suggested project folder structure
AI-Vehicle-IoT/
│
├── README.md
│
├── firmware/
│ └── esp32_vehicle_monitor/
│ └── esp32_vehicle_monitor.ino
│
├── n8n/
│ ├── telemetry-workflow.json
│ └── accident-workflow.json
│
├── docs/
│ ├── architecture.md
│ ├── hardware.md
│ ├── software.md
│ ├── testing.md
│ └── screenshots/
│
├── diagrams/
│ ├── block-diagram.png
│ ├── flowchart.png
│ ├── circuit.png
│ └── sequence-diagram.png
│
└── examples/
└── accident-payload.json
61. Sequence diagram
ESP32 n8n AI Agent Sheets ThingSpeak Telegram
| | | | | |
|---JSON------>| | | | |
| | | | | |
| |---validate-->| | | |
| | | | | |
| |--------------------------->| | |
| |---------------------------------------->| |
| | | | | |
| |---sensor data------------->| | |
| | | | | |
| | |--analysis-->| | |
| | | | | |
| |<--decision---| | | |
| | | | | |
| |--------------------------------------------------->|
| | | | | |
| |--------------------------------------------------->|
| | | | | |
| | | | |<--voice----|
| | | | | |
62. One-line project explanation for viva
The system uses an ESP32 to detect abnormal vehicle motion and obtain GPS coordinates, sends the event to n8n through an HTTP webhook, uses an AI Agent to analyze accident severity, logs the incident in Google Sheets, visualizes telemetry on ThingSpeak, and automatically sends Telegram text, voice and location alerts.
63. 30-second presentation explanation
“Our project is an AI-powered IoT accident detection and vehicle tracking system. An ESP32 collects acceleration and gyroscope data from an MPU6050 and GPS information from a GPS module. When abnormal vehicle motion is detected, the ESP32 sends the event to an n8n webhook. n8n acts as the automation and agentic layer. An AI Agent analyzes the sensor data and determines the possible accident severity. The incident is stored in Google Sheets and telemetry is sent to ThingSpeak. For high-severity events, n8n automatically sends a Telegram emergency message, voice alert and GPS location to the configured recipient.”
64. Final system architecture
┌───────────────────────┐
│ VEHICLE │
│ │
│ MPU6050 │
│ GPS │
│ SOS │
│ Buzzer │
└──────────┬────────────┘
│
▼
┌───────────────────────┐
│ ESP32 │
│ │
│ Sensor Processing │
│ Accident Detection │
│ GPS Processing │
│ Wi-Fi Communication │
└──────────┬────────────┘
│
HTTPS
│
▼
┌───────────────────────┐
│ n8n WEBHOOK │
└──────────┬────────────┘
│
▼
┌───────────────────────┐
│ DATA VALIDATION │
└──────────┬────────────┘
│
┌─────────────────┼──────────────────┐
│ │ │
▼ ▼ ▼
Google Sheets ThingSpeak AI Agent
│ │ │
│ │ ▼
│ │ Severity
│ │ │
│ │ ┌───────┴───────┐
│ │ │ │
│ │ NORMAL ACCIDENT
│ │ │ │
│ │ │ ▼
│ │ │ Telegram Text
│ │ │ │
│ │ │ ▼
│ │ │ Text-to-Speech
│ │ │ │
│ │ │ ▼
│ │ │ Telegram Voice
│ │ │ │
│ │ │ ▼
│ │ │ Telegram GPS
│ │ │
└─────────────────┴──────────┴───────────────┘
65. Recommended implementation order
Build it in this order rather than trying everything simultaneously:
-
ESP32 + MPU6050
-
Confirm acceleration and gyro readings.
-
-
ESP32 + GPS
-
Confirm latitude, longitude and speed.
-
-
ESP32 accident algorithm
-
Display confidence on Serial Monitor.
-
-
ESP32 + Wi-Fi
-
Confirm Internet connectivity.
-
-
n8n Webhook
-
Receive a manually generated JSON payload first.
-
-
ESP32 → n8n
-
Send actual sensor data.
-
-
Google Sheets
-
Log every event.
-
-
ThingSpeak
-
Display sensor charts.
-
-
Telegram
-
Send text alerts.
-
-
AI Agent
-
Add classification and severity.
-
-
TTS
-
Convert AI-generated emergency message to speech.
-
-
Telegram Voice
-
Send the generated voice alert.
-
-
GPS location
-
Send the vehicle's location.
-
-
Failure/retry logic
-
Handle Wi-Fi, n8n, Telegram and GPS failures.
-
-
Final enclosure
-
Mount the ESP32 and sensors securely in the vehicle.
-
This staged approach makes troubleshooting much easier.
Official references
-
ESP32 Arduino Wi-Fi documentation
-
ESP32 HTTPClient source/documentation
-
ThingSpeak Write Data API
-
ThingSpeak REST API reference
-
n8n documentation
-
n8n Telegram node documentation
-
Telegram Bot API
Important: For an academic prototype, this design is complete enough to implement and demonstrate. For a real vehicle/emergency deployment, the accident classifier, electrical design, enclosure, connectivity, cybersecurity and emergency escalation would need substantially more validation and safety engineering.
Project Summary
AI Accident Alert & Vehicle Tracking Using IoT Analytics is an IoT-based vehicle safety system that combines ESP32, MPU6050, GPS, n8n automation, AI Agent, Telegram, Google Sheets, and ThingSpeak.
Core workflow
MPU6050 + GPS
↓
ESP32
↓
Accident Detection
↓
Wi-Fi / HTTP
↓
n8n Webhook
↓
AI Agent
↓
┌────┼───────────────┐
↓ ↓ ↓
Sheets ThingSpeak Telegram
↓
Text + Voice
↓
GPS Location
Main functions
-
ESP32 collects vehicle motion data.
-
MPU6050 measures acceleration and gyroscope movement.
-
GPS provides latitude, longitude and vehicle speed.
-
ESP32 calculates an accident confidence score.
-
n8n receives and processes the IoT event.
-
AI Agent classifies the event as Normal, Suspicious or Accident and estimates severity.
-
Google Sheets stores accident and telemetry records.
-
ThingSpeak provides cloud telemetry visualization.
-
Telegram sends emergency text notifications.
-
Text-to-Speech generates an emergency voice message.
-
Telegram Voice delivers the voice alert.
-
GPS coordinates are sent so the recipient can locate the vehicle.
-
A physical SOS button can manually trigger an emergency alert.
Key architecture principle
The ESP32 performs the initial accident detection locally, while the AI Agent performs higher-level analysis. This prevents the system from depending entirely on AI or Internet connectivity for the initial detection.
Main hardware
-
ESP32 DevKit
-
MPU6050
-
NEO-6M GPS
-
Buzzer
-
SOS push button
-
LED
-
Power supply
Main software
-
Arduino IDE / ESP32 Arduino framework
-
C++
-
n8n
-
AI/LLM
-
Telegram Bot API
-
Google Sheets
-
ThingSpeak
-
Text-to-Speech service
Example emergency event
🚨 POSSIBLE VEHICLE ACCIDENT
Vehicle: CAR-001
Severity: HIGH
Confidence: 91%
Speed: 58.2 km/h
Location:
17.385044, 78.486671
AI analysis:
Large acceleration spike combined
with abnormal rotational movement.
Voice alert: SENT
GPS location: SENT
Google Sheets: LOGGED
ThingSpeak: UPDATED
Project objective
The overall goal is to create an agentic IoT vehicle-monitoring platform that can automatically:
Sense → Detect → Analyze → Log → Decide → Notify → Track
It is suitable as a final-year engineering project, IoT project, AI project, ESP32 project, or n8n automation project, with further development required before any real-world safety-critical deployment.
AI Accident Alert & Vehicle Tracking — Mind Map
┌──────────────────────────────┐
│ AI ACCIDENT ALERT & │
│ VEHICLE TRACKING SYSTEM │
└──────────────┬───────────────┘
│
┌─────────────────────────────────┼─────────────────────────────────┐
│ │ │
▼ ▼ ▼
┌───────────────┐ ┌───────────────┐ ┌───────────────┐
│ HARDWARE │ │ ESP32 IoT │ │ CLOUD │
└───────┬───────┘ └───────┬───────┘ └───────┬───────┘
│ │ │
┌─────┼─────┐ ┌─────┼─────┐ ┌──────┼──────┐
│ │ │ │ │ │ │ │ │
▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼ ▼
MPU6050 GPS SOS Wi-Fi JSON HTTP n8n ThingSpeak Sheets
│ │ │ │ │ │ │ │ │
│ │ │ └─────┴─────┘ │ │ │
│ │ │ │ │ │ │
▼ ▼ ▼ ▼ │ │ │
Accel GPS Button n8n Webhook │ │ │
Gyro Speed Buzzer │ │ │ │
▼ │ │ │
Data Validation │ │ │
│ │ │ │
▼ │ │ │
AI AGENT ◄────────────────────────┘ │ │
│ │ │
┌──────────┼──────────┐ │ │
│ │ │ │ │
▼ ▼ ▼ │ │
NORMAL SUSPICIOUS ACCIDENT │ │
│ │ │
▼ │ │
SEVERITY │ │
│ │ │
┌──────────────┼──────────────┐ │ │
│ │ │ │ │
▼ ▼ ▼ │ │
LOW HIGH CRITICAL │ │
│ │ │ │ │
└──────────────┼──────────────┘ │ │
│ │ │
▼ │ │
NOTIFICATION │ │
│ │ │
┌────────────────────┼─────────────────┐ │ │
│ │ │ │ │
▼ ▼ ▼ │ │
Telegram Text Telegram Voice GPS Location│ │
│ │ │ │ │
└────────────────────┼─────────────────┘ │ │
│ │ │
▼ │ │
Emergency User │ │
│ │
└──────┘
Simplified Concept Map
AI VEHICLE SAFETY
│
├── 1. SENSING
│ ├── MPU6050
│ │ ├── Acceleration
│ │ └── Gyroscope
│ ├── GPS
│ │ ├── Latitude
│ │ ├── Longitude
│ │ └── Speed
│ └── SOS Button
│
├── 2. ESP32
│ ├── Sensor Reading
│ ├── Accident Algorithm
│ ├── Confidence Score
│ ├── Wi-Fi
│ └── JSON/HTTP
│
├── 3. ACCIDENT DETECTION
│ ├── Acceleration Spike
│ ├── Gyroscope Spike
│ ├── Speed
│ ├── Motion Change
│ └── Confidence
│
├── 4. n8n AUTOMATION
│ ├── Webhook
│ ├── Validation
│ ├── Data Processing
│ ├── AI Agent
│ └── Decision Logic
│
├── 5. AI AGENT
│ ├── Event Classification
│ │ ├── Normal
│ │ ├── Suspicious
│ │ └── Accident
│ ├── Severity
│ │ ├── Low
│ │ ├── Medium
│ │ ├── High
│ │ └── Critical
│ └── Recommended Action
│
├── 6. CLOUD
│ ├── Google Sheets
│ │ └── Incident Database
│ └── ThingSpeak
│ ├── Charts
│ ├── Telemetry
│ └── Location
│
├── 7. ALERT SYSTEM
│ └── Telegram
│ ├── Text Alert
│ ├── Voice Alert
│ └── GPS Location
│
├── 8. RELIABILITY
│ ├── Wi-Fi Failure
│ ├── n8n Failure
│ ├── Telegram Failure
│ ├── GPS Failure
│ ├── AI Failure
│ └── Retry / Local Storage
│
└── 9. FUTURE
├── GSM/LTE
├── Camera
├── OBD-II
├── Machine Learning
├── Driver Monitoring
└── Multi-Vehicle Fleet
