Research on Kinect Calibration and Depth Error Compensation Based on BP Neural Network

Abstract

In order to improve the depth measurement accuracy of Kinect and reduce the depth error, we applied BP neural network to the calibration of Kinect depth camera in the research. The corner information of the target image is extracted as the training set, and the dual neural network is used to calibrate the data information in different directions according to the imaging rules of pixels. The results of neural network training prediction are used to improve the calibration accuracy further. Then, based on the improvement of calibration accuracy, error analysis, and error compensation are performed for the situation where the depth error in Kinect measurement is significant. An error compensation model is established by using the error symmetry of the Kinect depth measurement, and the depth information is compensated by the optical path difference, which reduces the order of parameters and improves the measurement accuracy. Experimental results show that this method can effectively improve the calibration and depth measurement accuracy of Kinect cameras.