Camera orientation estimation using voting approach on the Gaussian sphere for in-vehicle camera.

Area of Science:

• Computer Vision
• Robotics
• Automotive Engineering

Background:

• Vehicle camera calibration is essential for Advanced Driver Assistance Systems (ADAS).
• Accurate camera orientation is critical for ADAS functionality and safety.
• Existing methods may lack robustness in real-world driving conditions.

Purpose of the Study:

• To propose a novel and accurate method for estimating the orientation of vehicle-mounted cameras.
• To enhance the stability and speed of camera calibration for ADAS applications.
• To leverage imaging geometry through orthogonal vanishing points for orientation estimation.

Main Methods:

• Detection of orthogonal vanishing points based on imaging geometry and driving environment characteristics.
• Utilizing projected lines on a Gaussian sphere and plane normal extraction.
• Employing linear Hough transform for optical axis vanishing point estimation and circular histograms for subsequent vanishing points.
• Sequential estimation and voting for three vanishing points to improve accuracy and stability.

Main Results:

• The proposed method accurately estimates vehicle camera orientation in normal driving situations.
• Sequential estimation of three vanishing points enhances both accuracy and stability.
• Rapid orientation estimation is achieved by converting voting spaces to 2D planes.

Conclusions:

• The novel vanishing point detection method provides a robust solution for vehicle camera calibration.
• This technique contributes to the advancement of reliable ADAS by ensuring precise camera orientation.
• The method offers a practical and efficient approach for real-time camera orientation estimation in automotive applications.