Parallel feature based calibration method for a trinocular vision sensor.

Area of Science:

• Computer Vision
• Robotics
• Metrology

Background:

• Trinocular vision systems require accurate calibration for reliable 3D reconstruction.
• Existing calibration methods can be sensitive to noise and require precise feature correspondences.
• Self-calibration of multi-camera systems remains a challenge.

Purpose of the Study:

• To present a novel and robust method for calibrating trinocular vision sensors.
• To leverage the trifocal tensor and line correspondences for improved accuracy.
• To enable precise distance measurements and facilitate self-calibration.

Main Methods:

• Utilizing a planar target with parallel lines for feature extraction.
• Calculating the trifocal tensor from line correspondences across three image planes.
• Deriving essential and fundamental matrices for pairwise camera calibration.
• Employing singular value decomposition for rotation and translation matrix estimation.
• Performing image rectification to eliminate perspective distortion.

Main Results:

• Achieved precise calibration with a root mean square error of 0.029 mm over a 250x250 mm field of view.
• Demonstrated robustness by not requiring precise point-to-point correspondences.
• Showcased the trifocal tensor's utility as a constraint for feature matching.
• Validated the method's effectiveness through experimental results.

Conclusions:

• The proposed method offers a precise and efficient approach to trinocular vision sensor calibration.
• The use of parallel lines simplifies the calibration process and enhances accuracy.
• This method provides a viable pathway for self-calibration in trinocular vision systems, applicable to natural scenes.