ElshadaiK / Advancing-Transportation-Safety-and-Behavior-Research-through-AI

Project Summary: Driving Behavior Analysis and Risk Assessment in Simulated Environments

Overview

This project focuses on the development and implementation of a comprehensive system for analyzing, quantifying, and predicting driver behavior, as well as assessing risks in simulated driving environments. Utilizing a combination of machine learning techniques and kinematic data analysis, the project aims to enhance understanding of driving safety and performance.

Key Components and Processes

1. Data Preprocessing:

   • Involved cleansing and normalization of raw data from a driving simulator, which included metrics like velocity, acceleration, braking, steering, and lane deviation.
   • Synchronization with video data for a cohesive analysis.

2. Driver Risk Assessment:

   • Developed a model to quantify risky driving behavior based on parameters like speed, acceleration, and lane deviations.
   • Utilized percentile-based thresholds to flag high-risk incidents.
   • The RandomForestRegressor model predicted a RiskScore indicating the overall risk level of the driver's behavior. However, the model's accuracy suggested overfitting, likely due to the direct relationship between features and the computed RiskScore.

3. Environmental Risk Quantification:

   • Used a pre-trained YOLO model for object detection in video data.
   • Identified and scored potential hazards like other vehicles and pedestrians to quantify environmental risks.

4. Object Annotation in Video Data:

   • Annotated video frames with detected hazards, providing visual cues for analysis and training purposes.

5. Machine Learning Model Development and Validation:

   • Apart from the RandomForestRegressor for driver risk, the project employed cross-validation and other evaluation metrics to ensure model reliability and generalization.

6. Finalizing and Reporting:

   • Compiled comprehensive documentation for each component of the project.
   • Analyzed and interpreted results for practical applications like driver training and simulation enhancement.

Outcomes and Conclusions

• The project successfully demonstrates a multifaceted approach to risk assessment in driving, integrating kinematic data and video analysis.
• The driver risk assessment model, despite its overfitting issue, provides valuable insights into risky behaviors, underscoring the need for careful feature selection in predictive modeling.
• The environmental risk assessment through object detection highlights potential external hazards, contributing to a more rounded understanding of driving safety.
• Annotations in video data serve as a powerful tool for visualization and education, enhancing the interpretability of risk factors in simulated environments.

Future Directions

• Addressing the overfitting issue in the driver risk model through feature engineering and external validation.
• Enhancing the object detection framework to include more nuanced categories of hazards and integrating it with the kinematic data for a more comprehensive risk analysis.
• Expanding the model to adapt to different types of driving scenarios and environments for broader applicability.

Implications

This project sets the stage for advanced driving safety analysis, offering methodologies that can be extended to real-world applications. It provides a foundation for improving driver training programs, enhancing driving simulations, and potentially influencing the development of safety features in autonomous driving systems.