A Multi-Step CNN-Based Estimation of Aircraft Landing Gear Angles.

Area of Science:

• Aerospace Engineering
• Computer Vision
• Machine Learning

Background:

• Condition monitoring of aircraft landing gear is crucial for safe landings.
• Manual observation methods for landing gear angles are subjective and lack precision.
• Advancements in deep learning and pose estimation offer potential for objective measurement.

Purpose of the Study:

• To develop an objective method for measuring aircraft landing gear angles from monocular RGB images.
• To overcome limitations of traditional manual observation techniques.
• To leverage deep learning for precise pose estimation of landing gear.

Main Methods:

• Utilizes a monocular camera and a Computer-Aided Design (CAD) aircraft model.
• Employs target detection models to identify key landing gear points in 2D images.
• Applies a vector field network for pose estimation and calculates angles without depth information.
• Improves the vector field loss function and uses synthetic datasets for validation.

Main Results:

• The proposed algorithm accurately measures landing gear angles in two-dimensional images.
• Achieves a mean error of less than 5 degrees on a light-varying dataset.
• Demonstrates the validity of the algorithm through experiments with synthetic datasets.

Conclusions:

• The developed method provides an objective and accurate approach to aircraft landing gear angle measurement.
• This technique can significantly contribute to enhanced aircraft condition monitoring and safety.
• The algorithm shows promise for real-world applications in aviation maintenance and safety.