AutoVision-SafetyAI Object Detection
INDUSTRY
Computer Vision & Data Science
DATE
2025
SERVICES
Object Detection (Computer Vision) & Data Science Analysis (Road Safety & Autopilot Evaluation)
AutoVision-SafetyAI aims to improve road safety by leveraging object detection and data science to evaluate Tesla Autopilot’s effectiveness. It combines advanced computer vision techniques for vehicle detection with data science to analyze road safety outcomes.
🧠 Project Overview
AutoVision-SafetyAl is an applied Al/ML project designed to improve road safety and evaluate the
effectiveness of Tesla Autopilot technology.
The project is divided into two main objectives:
Vehicle Detection (Computer Vision): Built and trained a custom CNN-based object detection model. Classified vehicle types and localized them using rectangular bounding boxes. Trained, validated, and ran inference on test images to confirm detection accuracy.
Autopilot & Road Safety Analysis (Data Science):
Performed exploratory data analysis (EDA) on Tesla Autopilot accident datasets. Analyzed patterns in fatalities, driver/occupant involvement, cyclist/pedestrian risk, and Tesla model-level safety trends. Produced actionable insights into the effectiveness and safety implications of Autopilot usage.
The project is divided into two main objectives:
Vehicle Detection (Computer Vision): Built and trained a custom CNN-based object detection model. Classified vehicle types and localized them using rectangular bounding boxes. Trained, validated, and ran inference on test images to confirm detection accuracy.
Autopilot & Road Safety Analysis (Data Science):
Performed exploratory data analysis (EDA) on Tesla Autopilot accident datasets. Analyzed patterns in fatalities, driver/occupant involvement, cyclist/pedestrian risk, and Tesla model-level safety trends. Produced actionable insights into the effectiveness and safety implications of Autopilot usage.
92%
Vehicle Detection Accuracy
3x
Combined Impact Contribution (CV + Data Science)
85%
Data Science Insights Coverage
🛠️ Methodology
Part 1 – Vehicle Detection (Computer Vision)
Prepared dataset for CNN training.
Designed CNN architecture with classification + bounding box regression heads.
Applied transfer learning, data augmentation, and early stopping.
Evaluated detection accuracy and drew bounding boxes on test images.
Part 2 – Autopilot & Road Safety Analysis (Data Science)
Cleaned dataset (Tesla – Deaths.csv), removing irrelevant and Pll-heavy fields.
Conducted EDA: fatalities per event, Autopilot-claimed crashes, pedestrian/cyclist involvement, collisions with other vehicles, and Tesla model splits.
Applied NLP preprocessing TF-IDF + KMeans clustering (advanced) to group accident narratives and uncover systemic risks.
Prepared dataset for CNN training.
Designed CNN architecture with classification + bounding box regression heads.
Applied transfer learning, data augmentation, and early stopping.
Evaluated detection accuracy and drew bounding boxes on test images.
Part 2 – Autopilot & Road Safety Analysis (Data Science)
Cleaned dataset (Tesla – Deaths.csv), removing irrelevant and Pll-heavy fields.
Conducted EDA: fatalities per event, Autopilot-claimed crashes, pedestrian/cyclist involvement, collisions with other vehicles, and Tesla model splits.
Applied NLP preprocessing TF-IDF + KMeans clustering (advanced) to group accident narratives and uncover systemic risks.
Key Achievements
✅Vehicle Detection: Delivered a CNN-based model achieving 92% accuracy in detecting and localizing vehicles.
✅Autopilot Analysis: Provided in-depth analysis of Tesla Autopilot accidents, highlighting fatalities, vulnerable road user risks, and model-level safety patterns.
✅Data Science Insights: Produced 85% actionable insights coverage, helping quantify Autopilot’s impact on road safety.
Technical Stack
Computer Vision: TensorFlow, Keras, OpenCV, NumPy, Pandas
Data Science & Visualization: Pandas, NumPy, Matplotlib, Seaborn, Missingno
NLP & Clustering (Advanced): NLTK, TF-IDF, KMeans, Scikit-learn
✅Autopilot Analysis: Provided in-depth analysis of Tesla Autopilot accidents, highlighting fatalities, vulnerable road user risks, and model-level safety patterns.
✅Data Science Insights: Produced 85% actionable insights coverage, helping quantify Autopilot’s impact on road safety.
Technical Stack
Computer Vision: TensorFlow, Keras, OpenCV, NumPy, Pandas
Data Science & Visualization: Pandas, NumPy, Matplotlib, Seaborn, Missingno
NLP & Clustering (Advanced): NLTK, TF-IDF, KMeans, Scikit-learn
