MATE: Machine Learning for Adaptive Calibration Template Detection.

Area of Science:

• Computer Vision
• Machine Learning
• Robotics

Background:

• Camera calibration is crucial for accurately mapping 3D world points to 2D image points.
• A key challenge in camera calibration is reliably detecting feature correspondences, such as chessboard corners.
• Existing methods for feature detection often involve high computational and implementation complexity.

Purpose of the Study:

• To develop a novel, efficient, and robust method for detecting chessboard corners using deep learning.
• To improve upon the speed and adaptability of traditional camera calibration techniques.
• To create a generalized solution that handles various image degradations and distortions.

Main Methods:

• A convolutional neural network (CNN) was designed and trained on a large dataset of chessboard images.
• The CNN was explicitly trained to be robust against noisy inputs and significant lens distortions.
• The method leverages the generalized nature of deep learning for feature template extraction.

Main Results:

• The proposed CNN method achieves accuracy comparable to existing techniques for chessboard corner detection.
• The deep learning approach offers significantly improved execution times compared to traditional methods.
• The network demonstrates robustness against lens distortions, sensor noise, and perspective deformations.

Conclusions:

• Deep learning, specifically CNNs, provides an effective and efficient solution for automatic chessboard corner detection in camera calibration.
• The proposed method generalizes existing solutions and offers enhanced adaptability for specific applications.
• This approach is robust to various image degradations, making it suitable for complex real-world scenarios, including multi-camera systems.