Alignment parameter calibration for IMU using the Taguchi method for image deblurring

Abstract

Inertial measurement units (IMUs) utilized in smartphones can be used to detect camera motion during exposure, in order to improve image quality degraded with blur through long hand-held exposure. Based on the captured camera motion, blur in images can be removed when an appropriate deblurring filter is used. However, two research issues have not been addressed: (a) the calibration of alignment parameters for the IMU has not been addressed. When inappropriate alignment parameters are used for the IMU, the camera motion would not be captured accurately and the deblurring effectiveness can be downgraded. (b) Also selection of an appropriate deblurring filter correlated with the image quality has still not been addressed. Without the use of an appropriate deblurring filter, the image quality could not be optimal. This paper proposes a systematic method, namely the Taguchi method, which is a robust and systematic approach for designing reliable and high-precision devices, in order to perform the alignment parameter calibration for the IMU and filter selection. The Taguchi method conducts a small number of systematic experiments based on orthogonal arrays. It studies the impact of the alignment parameters and appropriate deblurring filter, which attempts to perform an effective deblurring. Several widely adopted image quality metrics are used to evaluate the deblurred images generated by the proposed Taguchi method. Experimental results show that the quality of deblurred images achieved by the proposed Taguchi method is better than those obtained by deblurring methods which are not involved with the alignment parameter calibration and filter selection. Also, much less computational effort is required by the Taguchi method when comparing with the commonly used optimization methods for determining alignment parameters and deblurring filter.