🚗 Autonomous Driving using Deep Learning & Data Analysis

An end-to-end AI/ML project that combines Deep Learning (Object Detection) and Data Science (EDA) to simulate key components of autonomous driving systems.

This project is divided into two major parts:

• Part 1 → Vehicle Detection using Deep Learning
• Part 2 → Tesla Autopilot Incident Analysis

---

📌 Problem Statement

This project is based on a real-world autonomous driving scenario where:

• Vehicles must be detected and localized in images
• Autopilot systems must be analyzed for road safety impact

For the complete problem statement open the Problem Statement.pdf inside the Datasets & Problem Statement folder.

---

🎯 Project Objectives

🔹 Part 1: Object Detection

• Detect vehicles in images
• Localize them using bounding boxes
• Train a deep learning model for detection

🔹 Part 2: Data Analysis

• Analyze Tesla autopilot accident data
• Perform EDA on fatalities, locations, and trends
• Understand safety implications

---

🏗️ Project Structure

• Datasets & Problem Statement
  • Part 1
    • labels.csv
  • Part 2
    • Tesla-Deaths.csv
  • Problem Statement.pdf
• Notebook
  • reports/figures
    • Data Cleaning.ipynb
    • EDA Analysis.ipynb
• src
  • part1DeepLearning
    • init.py
    • data ingestion.py
    • evaluation.py
    • inference.py
    • model.py
    • train.py
  • init.py
• .gitignore
• LICENSE
• README.md
• outputs
• requirements.txt

---

🧠 Part 1: Vehicle Detection (Deep Learning)

🔹 Pipeline

1. Data Ingestion

   • Reads annotations CSV
   • Splits into train/validation sets
     👉 data_ingestion.py

2. Model Architecture

   • Faster R-CNN with MobileNet backbone
   • Pretrained model fine-tuned for vehicle detection
     👉 model.py

3. Training

   • Custom PyTorch dataset
   • Bounding box + label training
   • GPU support
     👉 train.py

4. Evaluation

   • IoU-based matching
   • Precision & Recall calculation
     👉 evaluation.py

5. Inference

   • Predict bounding boxes on new images
   • Draw detections and save outputs
     👉 inference.py

---

⚙️ Model Details

• Architecture: Faster R-CNN
• Backbone: MobileNet V3
• Task: Object Detection (Bounding Boxes + Labels)

---

📊 Part 2: Tesla Autopilot Data Analysis

🔹 Tasks Performed

• Data Cleaning & Preprocessing
• Exploratory Data Analysis (EDA)
• Trend analysis of:
  • Accidents over time
  • Death distribution
  • Driver vs pedestrian impact
  • Autopilot-related fatalities

📁 Notebooks

• Data Cleaning.ipynb
• EDA Analysis.ipynb

---

📦 Dataset

Part 1

• Image dataset with bounding box annotations
• Format:
  • image_id
  • label
  • xmin
  • ymin
  • xmax
  • ymax

Part 2

• Tesla accident dataset including:
  • Date, location, deaths
  • Autopilot involvement
  • Vehicle & collision details

---

🚀 How to Run

1️⃣ Clone Repository

• git clone https://github.com/PrakharSri18-data/Autonomous-Driving-Deep-Learning.git
• cd Autonomous-Driving-Deep-Learning

2️⃣ Setup Environment

• python -m venv venv
• venv\Scripts\activate

3️⃣ Install Dependencies

• pip install -r requirements.txt

4️⃣ Data Ingestion

• python src/part1DeepLearning/data_ingestion.py

5️⃣ Train Model

• python src/part1DeepLearning/train.py

6️⃣ Evaluate Model

• python src/part1DeepLearning/evaluation.py

7️⃣ Run Inference

• python src/part1DeepLearning/inference.py

---

📈 Results

• Successfully detects vehicles with bounding boxes
• Model evaluated using Precision & Recall
• Visual outputs saved in outputs folder

---

⚠️ Challenges

• Dataset size & preprocessing time
• Bounding box accuracy
• Class imbalance
• Real-time performance constraints

---

🧠 Key Learnings

• End-to-end ML pipeline design
• Object detection using PyTorch
• Evaluation using IoU metrics
• Real-world data analysis (Tesla autopilot)
• Combining Deep Learning + Data Science in one project

---

👤 Author

• Prakhar Srivastava
• Aspiring Data Scientist & Business Analyst | Machine Learning, Deep Learning & Generative AI Enthusiast