Effective Data-Driven Calibration for a Galvanometric Laser Scanning System Using Binocular Stereo Vision.

Area of Science:

• Robotics and Control Systems
• Machine Learning Applications
• Optical Engineering

Background:

• Galvanometric laser scanning (GLS) systems require precise calibration for accurate performance.
• Traditional model-driven calibration methods are often complex and time-consuming.
• Existing data-driven methods typically address specific tasks like laser triangulation.

Purpose of the Study:

• To develop a novel, accurate, and stable calibration method for GLS systems.
• To leverage machine learning for a more efficient and versatile calibration solution.
• To provide a unified approach for both laser projection and 3D reconstruction tasks.

Main Methods:

• A single-hidden layer feedforward neural network (SLFN) was employed to model the GLS system.
• Extreme learning machine (ELM) was used for efficient closed-form parameter solving of the SLFN.
• Training data was acquired using a moving mechanism and a binocular stereo system.

Main Results:

• The proposed SLFN-ELM method demonstrated high accuracy and stability compared to traditional calibration.
• The number of hidden neurons in the SLFN can be broadly selected without compromising generalization performance.
• The method achieved significantly less training time due to direct space vector output.

Conclusions:

• The machine learning-based GLS calibration offers a simpler, more accurate, and stable alternative to model-driven approaches.
• The proposed method provides a versatile solution applicable to both laser projection and 3D reconstruction.
• The findings suggest that SLFNs trained with ELM are effective for complex system calibration.