Data Calibration of Binocular Vision Pulse Acquisition System Based on CNN+GRU

To address potential calibration errors that may occur during the acquisition of pulse images based on binocular vision, as well as inaccuracies in three-dimensional(3D) reconstruction, a data calibration model based on Convolutional Neural Networks and Gated Recurrent Units (CNN+GRU) is proposed as a method of calibrating the pulse wave data acquired by the binocular vision pulse acquisition system. The CNN is used to retrieve the error characteristics of the pulse wave and the GRU is used to learn the time-dependent characteristics of the monitoring data and the occurrence of data drift. The precise data produced by the laser profile sensor serves as a reference for calibration. Finally, the fusion module is used to collect the calibration data. The results of both the calibration and ablation experiments show that the calibration model and its component modules are effective.