Accident Detection with a Reporting System 🚗🚨

Abstract 📜

The rise in global population and economic growth has led to an increase in the number of vehicles on the roads, which unfortunately also results in a higher incidence of road accidents. Quickly identifying accident locations and reporting them to emergency services can save many lives. This project leverages Convolutional Neural Networks (CNN), Python, and other tools to create an accident detection system that can analyze accident footage, identify incidents, and alert emergency services promptly.

Features ✨

• Automated accident detection using Deep Learning (CNN).
• Reporting system using Twilio SMS for quick alerts.
• Data visualization and analysis for understanding accident trends.

Requirements ⚙️

1. Python: Install Anaconda to manage dependencies.
2. Twilio: Sign up for an account on Twilio to enable SMS notifications (you can use the 7-day free trial).
3. Pre-trained VGG16 model for feature extraction.

Installation 📥

1. Set up the environment:

   conda create -n accident_detection python=3.8
   conda activate accident_detection

2. Install the necessary libraries:

   pip install numpy pandas matplotlib opencv-python-headless geopy keras tensorflow scikit-image twilio

3. Clone or download this repository and extract it to a suitable location.

4. Twilio Setup: Create an account on Twilio and retrieve the following:

   • Virtual phone number
   • Account SID
   • Authentication token

Usage 🚀

Step 1: Capture Frames from Video Footage

The script reads accident footage (e.g., Accidents.mp4) and saves frames for analysis.

• Ensure your video file is located in the same directory.
• Run the script to extract frames:

  cap = cv2.VideoCapture("Accidents.mp4")
  # Frame extraction code here...

Step 2: Run the Detection Model

The model will analyze each frame and detect accidents based on the trained CNN model.

• Training: Load the training_data.csv to train the model with images labeled as accident or no accident.
• Prediction: Once the model is trained, run predictions on test data and check the output in real-time.

Step 3: Send Notifications via Twilio SMS

In the script Accident Detection-Video.ipynb, enter your Twilio credentials and the recipient’s phone number:

account_sid = 'your_account_sid'
auth_token = 'your_auth_token'
client = Client(account_sid, auth_token)

When an accident is detected, the script will automatically send an SMS alert.

Code Snippets 🖥️

Frame Extraction

cap = cv2.VideoCapture("Accidents.mp4")
frameRate = cap.get(5)
count = 0
while(cap.isOpened()):
    ret, frame = cap.read()
    if not ret:
        break
    if int(cap.get(1)) % math.floor(frameRate) == 0:
        filename = f"{count}.jpg"
        cv2.imwrite(filename, frame)
        count += 1
cap.release()

Accident Detection Model Training

from keras.applications.vgg16 import VGG16
from keras.models import Sequential
from keras.layers import Dense, InputLayer
# Preprocess data and train the model
model = Sequential()
model.add(InputLayer((7*7*512,)))
model.add(Dense(units=1024, activation='sigmoid'))
model.add(Dense(2, activation='softmax'))
model.compile(loss='categorical_crossentropy', optimizer='adam', metrics=['accuracy'])
model.fit(X_train, y_train, epochs=100, validation_data=(X_valid, y_valid))

SMS Notification

from twilio.rest import Client
client = Client(account_sid, auth_token)
message = client.messages.create(
    body="Accident detected at [location]! Please dispatch emergency services.",
    from_='+1234567890', # Twilio number
    to='+0987654321'     # Recipient's number
)

Data Visualization 📊

The project uses Matplotlib to visualize accident occurrences. The value column in training_data.csv categorizes images into Accident or No-accident, providing a breakdown in a pie chart.

import matplotlib.pyplot as plt

labels = ['No-accident', 'Accident']
sizes = data['value'].value_counts().values
colors = ['pink', 'red']

plt.pie(sizes, labels=labels, colors=colors, autopct='%1.1f%%', startangle=90)
plt.axis('equal')
plt.title('Distribution of Accident and No-accident Data')
plt.show()

Folder Structure 📂

Accident-Detection-With-Reporting-System/
├── Accident Detection-Video.ipynb  # Main code
├── Accidents.mp4                   # Sample video for testing
├── training_data.csv               # Labeled data for model training
├── test.csv                        # Test data for validation
├── README.md                       # Documentation
└── requirements.txt                # Dependencies

Author 🧑‍💻

• Shristi Singh
  LinkedIn

---

License 📝

This project is licensed under the MIT License. Feel free to use, modify, and distribute this code as per the license conditions.

---

Notes 🗒️

• Ensure your video files and images are correctly named and located in the specified directory.
• Use a GPU if possible for faster model training.
• Update Twilio credentials before running the SMS functionality.

---