Video Stabilization-Based elimination of unintended jitter and vibration amplification in centrifugal pumps

Abstract

To address challenges in non-contact visual sensing monitoring technology for large mechanical systems—specifically, video source instability due to camera jitter leading to measurement distortion, and the difficulty of directly observing small vibration amplitudes—we propose a method for eliminating unintended jitter and amplifying vibrations in centrifugal pumps based on SIFT-RANSAC-EPBVM. The method combines the Scale-Invariant Feature Transform (SIFT) algorithm with the Random Sample Consensus (RANSAC) algorithm to eliminate mismatched feature points. By establishing an affine transformation matrix between each video frame and the initial frame, feature points are mapped into the coordinate system of the initial frame. The Enhanced Phase-Based Video Motion (EPBVM) algorithm is then employed to amplify and display minute vibration signals, with computational complexity reduced by decreasing image size during the decomposition and reconstruction stages of the video frames. Experimental results demonstrate that the proposed method significantly improves the accuracy of vibration signal extraction: video matching accuracy increases from 92.15% to 100%, and the mean and standard deviation of the Difference of Inter-frame Transformation Fidelity (DITF) are reduced by 30%–40%. Additionally, notable improvements are observed in video amplification quality, processing time, and resistance to environmental noise.