Multilayer Perceptron-Based Error Compensation for Automatic On-the-Fly Camera Orientation Estimation Using a Single...

Area of Science:

• Computer Vision
• Robotics
• Machine Learning

Background:

• Accurate camera orientation is crucial for autonomous vehicle navigation.
• Existing methods often struggle with real-time performance and accuracy under varying conditions.
• Cameras with zero roll angle are common in automotive applications.

Purpose of the Study:

• To develop and evaluate a novel multilayer perceptron (MLP) error compensation method for real-time camera orientation estimation.
• To enhance the accuracy of estimating pitch and yaw angles using a single vanishing point and road lane lines.
• To validate the method's effectiveness in both simulated and real-world driving scenarios.

Main Methods:

• Utilized a multilayer perceptron (MLP) for error compensation in camera orientation estimation.
• Employed a single vanishing point and road lane lines as primary inputs.
• Integrated two Kalman filter models with image point (u, v) and derived angle inputs.
• Focused on cameras with a 0° roll angle, typical for autonomous vehicles.

Main Results:

• The proposed MLP method significantly improved the accuracy of camera orientation estimations.
• Performance metrics (avgE, minE, maxE, ssE, Stdev) demonstrated superior results compared to existing techniques.
• The system showed consistent accuracy across diverse scenarios in both simulator and real-vehicle tests.

Conclusions:

• The developed MLP error compensation method offers a robust and precise solution for real-time camera orientation estimation.
• The approach is adaptable and accurate, showing significant promise for enhancing autonomous vehicle systems.
• This method provides a reliable foundation for advanced driver-assistance systems and fully autonomous driving.