Prior information-based motion blur image restoration method for the low-frequency sinusoidal vibration measurements with machine vision

Abstract

The sinusoidal vibration is the basic unit of motion control and measurement, which can constitute the arbitrary plane and space motions and is essential to the low-frequency vibration calibration. At present, the machine vision (MV) method has been widely applied to measure the low-frequency vibration, which can promote the improvement of low-frequency vibration calibration and metrology traceability. However, the MV method is always inevitably affected by the motion blur which decreases its measurement accuracy. In this study, a novel prior information-based motion blur image restoration method is investigated, which applies the known sinusoidal information and the Lucy-Richardson (LR) method with Non-Local (NL)-Means to eliminate the motion blur and improve the measurement accuracy. Meanwhile, the objective evaluation indexes and edge detection method are adopted to verify the restoration effect. Several simulations and actual experiment show that the investigated method can significantly improve the quality of the sinusoidal motion blur images, and further make the accuracies of the edge detection method and MV method increase 0.3 pixel and 0.3%, respectively.