A Framework for Optimizing Deep Learning-Based Lane Detection and Steering for Autonomous Driving
2024

Optimizing Lane Detection and Steering for Self-Driving Cars

Sample size: 5248 publication Evidence: moderate

Author Information

Author(s): Yordanov Daniel, Chakraborty Ashim, Hasan Md Mahmudul, Cirstea Silvia

Primary Institution: Anglia Ruskin University

Hypothesis

Can a novel framework improve lane detection and steering in autonomous vehicles using deep learning?

Conclusion

The study developed a framework that enables autonomous vehicles to stay within lane boundaries with an autonomy of 77% in low challenge conditions and 66% in more complex scenarios.

Supporting Evidence

  • The framework allows for automated data collection, reducing human error.
  • The CNN model achieved 77% autonomy in low challenge conditions.
  • Data augmentation techniques were employed to enhance the training dataset.
  • The model was evaluated using real-world driving footage from Comma.ai.

Takeaway

This study created a computer program that helps self-driving cars learn how to stay in their lanes, even on tricky roads.

Methodology

A virtual environment was created in Unity3D to generate diverse driving scenarios, and a convolutional neural network was trained on the collected data to predict steering angles.

Potential Biases

The model may be biased towards straight driving due to the imbalance in the training dataset.

Limitations

The model struggles with roads that have sharp turns and ambiguous lane markings.

Digital Object Identifier (DOI)

10.3390/s24248099

Want to read the original?

Access the complete publication on the publisher's website

View Original Publication