🏍️ RideGuard

Motorcycle Accident Detection & Emergency Alert System

Final Year Project — MCA Semester 4

A real-time motorcycle accident detection system that combines computer vision, IoT sensors, machine learning, and automated emergency alerts to improve rider safety.

Features • Architecture • Modules • Installation • Usage • Evaluation • Tech Stack

---

Overview

RideGuard is an integrated motorcycle safety system that detects accidents in real-time using multiple sensor inputs and immediately notifies emergency contacts via Telegram with GPS location and photographic evidence.

The system is simulated using the CARLA 0.9.10 autonomous driving simulator and validated with real IoT hardware including Arduino, MPU6050, HC-SR04 ultrasonic sensor, Neo-6M GPS, and ESP32-CAM.

---

Features

• Helmet Detection — YOLOv8 detects helmet usage every 5 seconds via webcam. Restricts speed to 30 km/h if no helmet detected
• Tilt/Fall Detection — MPU6050 gyroscope detects motorcycle tilt beyond 10 degrees and triggers fall animation in CARLA
• Rear Collision Detection — HC-SR04 ultrasonic sensor detects rear vehicle proximity. Triggers accident after 3 consecutive readings within 2cm
• Accident Hotspot Warning — DBSCAN clustering on 99 Chennai GPS accident records identifies 6 hotspot zones. Warns rider in real-time when within 2km
• Emergency Alert — Sends Telegram message with rider details, live GPS pin, and ESP32-CAM evidence photos within 20 seconds of accident

---

Architecture

┌─────────────────────────────────────────────────────────────┐
│                     RIDEGUARD SYSTEM                        │
├──────────────┬──────────────┬──────────────┬────────────────┤
│   MODULE 1   │   MODULE 2   │   MODULE 3   │   MODULE 4     │
│   Helmet     │    Tilt      │    Rear      │   Hotspot      │
│  Detection   │  Detection   │  Collision   │  Detection     │
│  (YOLOv8)   │  (MPU6050)   │ (Ultrasonic) │   (DBSCAN)     │
└──────┬───────┴──────┬───────┴──────┬───────┴───────┬────────┘
       │              │              │               │
       └──────────────┴──────────────┴───────────────┘
                              │
                    ┌─────────▼─────────┐
                    │   FILE-BASED IPC  │
                    │  (Shared .txt     │
                    │   files between   │
                    │    processes)     │
                    └─────────┬─────────┘
                              │
              ┌───────────────┼───────────────┐
              │               │               │
    ┌─────────▼──────┐ ┌──────▼──────┐ ┌─────▼──────────┐
    │  CARLA 0.9.10  │ │  MODULE 5   │ │   HUD DISPLAY  │
    │  Simulation    │ │  Emergency  │ │  (Pygame)      │
    │  Fall Animation│ │  Alert      │ │                │
    └────────────────┘ │  (Telegram) │ └────────────────┘
                       └─────────────┘

---

Modules

Module 1 — Helmet Detection (YOLOv8)

Detects whether the rider is wearing a helmet using a custom-trained YOLOv8 model via laptop webcam.

• Model: best.pt (custom trained, 2 classes)
• Classes: With Helmet / Without Helmet
• Confidence threshold: 0.50
• Check interval: every 5 seconds
• Speed enforcement: 80% throttle with helmet, 35% throttle without

Metric	Value
Accuracy	90.00%
Precision	88.46%
Recall	92.00%
F1-Score	0.9020

Module 2 — Tilt Detection (MPU6050)

MPU6050 gyroscope sensor measures motorcycle tilt angle. When tilt exceeds 10 degrees, it writes LEFT or RIGHT to a shared file which CARLA reads to trigger a 3-stage fall animation.

• Tilt threshold: 10 degrees
• Check interval: 5 seconds
• Secondary verification: Sobel edge + image moment analysis on ESP32-CAM images
• MPU6050 takes priority over camera analysis

Module 3 — Rear Collision Detection (HC-SR04 + ESP32-CAM)

HC-SR04 ultrasonic sensor monitors rear vehicle distance. Uses a 3-reading debounce buffer to prevent false triggers.

• Danger zone: <= 2cm (3 consecutive readings)
• Warning zone: 2cm to 6cm
• On trigger: writes ACCIDENT to IPC file, sends CAPTURE command to ESP32-CAM
• ESP32-CAM captures 5 images saved to SD card

Zone	Distance	Action
Safe	> 6cm	No action
Warning	2–6cm	HUD warning
Danger	<= 2cm (x3)	Accident trigger

Module 4 — Accident Hotspot Detection (DBSCAN)

DBSCAN clustering on 99 GPS accident records from Chennai identifies 6 accident-prone zones. Background thread checks rider GPS every 60 seconds and warns if within 2km of any hotspot.

• Algorithm: DBSCAN (Density-Based Spatial Clustering)
• Dataset: 99 GPS records (Chennai)
• eps: 0.5km / 6371 = 0.0000785 radians
• min_samples: 5
• metric: haversine
• Warning radius: 2.0 km

Detected Hotspot Zones:

#	Area	Accidents
1	Koyambedu Junction	18
2	Madhya Kailash (OMR)	19
3	Kathipara Junction	19
4	Poonamallee High Road	18
5	Anna Salai (Gemini)	18
6	Tambaram Bypass	7

Clustering Metric	Value
Silhouette Score	0.9879
Davies-Bouldin Index	0.0165
Calinski-Harabasz	395534.21

Module 5 — Emergency Alert System (Telegram)

On any accident trigger, collects GPS location, ESP32-CAM images, and rider details. Sends Telegram alert within 18–21 seconds.

Content	MPU Fall	Rear Collision
Text message	Yes	Yes
Live GPS pin	Yes	Yes
Evidence images	No	Yes (img_1 + img_5)

---

IPC Architecture

All modules communicate through shared text files (File-based IPC):

File	Writer	Reader	Content
fall_status.txt	MPU6050/Arduino	Merging_module_3.py	LEFT / RIGHT / NONE
accident_trigger.txt	Rear_Location.py	Merging_module_3.py	ACCIDENT
rear_warning.txt	Rear_Location.py	Merging_module_3.py	WARNING
hotspot_warning.txt	hotspot_detector.py	Merging_module_3.py	WARNING|count|dist
gps_log.txt	Rear_Location.py	emergency_alert.py	timestamp,lat,lng,sats
telegram_status.txt	emergency_alert.py	Merging_module_3.py	SENT

---

Installation

Prerequisites

• Python 3.7 (strict requirement for CARLA compatibility)
• CARLA 0.9.10 Simulator
• Arduino IDE (for hardware code upload)

Step 1 — Clone the repository

git clone https://github.com/Banuprasath/RideGuard.git
cd RideGuard

Step 2 — Run the installer

install.bat

Or manually install dependencies:

pip install -r requirements.txt

Step 3 — Configure environment

copy .env.example .env

Edit .env with your values:

TELEGRAM_BOT_TOKEN=your_bot_token
TELEGRAM_CHAT_ID=your_chat_id
ARDUINO_PORT=COM16
ESP32_PORT=COM3
ESP32_IP=192.168.137.X

Step 4 — Install CARLA 0.9.10

Download from: https://github.com/carla-simulator/carla/releases/tag/0.9.10

Extract to C:\CARLA_0.9.10\ then update the path in Merging_module_3.py line 21:

sys.path.append(
    r"C:\CARLA_0.9.10\WindowsNoEditor\PythonAPI\carla\dist\carla-0.9.10-py3.7-win-amd64.egg"
)

Step 5 — YOLOv8 model

The trained best.pt model is included in the repository root. No separate download needed.

If you want to retrain the model:

python train_helmet.py

---

Usage

Open 4 terminals and run in order:

# Terminal 1 — Start CARLA Server
cd C:\CARLA_0.9.10\WindowsNoEditor
CarlaUE4.exe

# Terminal 2 — Main System
cd RideGuard
python Merging_module_3.py

# Terminal 3 — Hardware Bridge
python Rear_Location.py

# Terminal 4 — Hotspot Monitor
python hotspot_detector.py

Controls (in CARLA window)

Key	Action
W / S	Throttle / Brake
A / D	Steer Left / Right
Arrow Left / Right	Simulate fall direction
1	Toggle helmet detection test
2	Trigger rear collision sequence
BACKSPACE	Reset after accident
C	Switch camera view
ESC	Quit

---

Evaluation

Run evaluation metrics (no hardware required):

python evaluation\thesis_evaluation.py

Generates 3 output images in evaluation/ folder:

File	Description
helmet_confusion_matrix.png	YOLOv8 confusion matrix
dbscan_hotspot_clusters.png	DBSCAN cluster scatter plot
dbscan_unsupervised_evaluation.png	Silhouette plot + cluster size chart

Overall System Metrics

Module	Metric	Value
Helmet Detection	Accuracy	90.00%
Helmet Detection	F1-Score	0.9020
Tilt Detection	Success Rate	92%
Rear Collision	Detection Accuracy	95%
Emergency Alert	End-to-end Time	18–21 seconds
DBSCAN Hotspot	Silhouette Score	0.9879

---

Project Structure

RideGuard/
├── Merging_module_3.py        # Main CARLA + Helmet + Tilt module
├── Rear_Location.py           # Hardware bridge (Arduino + GPS + ESP32)
├── emergency_alert.py         # Telegram alert + evidence logging
├── hotspot_detector.py        # DBSCAN hotspot detection + monitor
├── compare_hotspot_algorithms.py  # ML algorithm comparison
├── test_gps.py                # Neo-6M GPS test utility
├── train_helmet.py            # YOLOv8 training script
├── best.pt                    # Trained YOLOv8 helmet detection model
├── yolov8n.pt                 # YOLOv8 nano base model
├── accidents.csv              # 99 Chennai GPS accident records
├── requirements.txt           # Python dependencies
├── install.bat                # Auto installer for Windows
├── SETUP_GUIDE.txt            # Manual setup instructions
├── .env.example               # Environment config template
├── evaluation/                # Thesis evaluation scripts + charts
├── Neo-6M_Connections/        # Arduino GPS + Ultrasonic code
├── Testing_Dataset/           # Test CSVs for evaluation
├── Thesis/                    # Final year thesis report + viva PDF
├── utils/                     # Debug and utility scripts
└── tests/                     # Integration test scripts

---

Tech Stack

Category	Technology
Simulation	CARLA 0.9.10
Computer Vision	YOLOv8 (Ultralytics)
Machine Learning	DBSCAN (scikit-learn)
Hardware	Arduino Uno, MPU6050, HC-SR04, Neo-6M GPS, ESP32-CAM
Rendering	Pygame
Image Processing	OpenCV
Alert System	Telegram Bot API
Language	Python 3.7
Data	Pandas, NumPy
Visualization	Matplotlib
Serial Comm	PySerial

---

Hardware Connections

Arduino (COM16, 9600 baud)

• HC-SR04: TRIG → D9, ECHO → D10
• Neo-6M GPS: TX → D2, RX → D3

ESP32-CAM (COM3, 115200 baud)

• Receives CAPTURE command via serial
• Serves images via HTTP at http://ESP32_IP/image

---

Thesis & Documentation

The full project thesis and viva presentation are available in the Thesis/ folder:

File	Description
Thesis/Final_year_report_Banu_Prasath_crt (1).pdf	Complete project thesis report
Thesis/Banu-Prasath_S_Final_Viva.pdf	Final viva presentation slides

---

Author

Banuprasath S MCA Final Year — Semester 4 GitHub: @Banuprasath Repository: RideGuard

---

License

This project is licensed under the MIT License.

---

Built with dedication for Final Year Project — MCA Semester 4