Robust technology of industrial video surveillance system based on ORB algorithm

Abstract

A robust video processing technology that integrates adaptive Kalman filtering and optimized ORB algorithm is proposed to address the problems of image blur and feature matching failure caused by mechanical vibration and extreme lighting changes in industrial video surveillance. By dynamically adjusting the covariance matrix of KF process noise, the synergistic suppression of high-frequency vibration and pulse type illumination noise can be achieved. The experimental results show that optimizing the ORB algorithm enhances the invariance of feature point direction and scale through the Gaussian difference pyramid and grayscale centroid method, and the success rate of feature point matching is improved by 63.1% compared to the traditional method. In extreme lighting changes and dynamic scenes, the algorithm reduces image pixel offset by 56%–67% and increases peak signal-to-noise ratio by an average of 32%–34%, significantly better than mainstream methods such as histogram equalization and contrast-limited adaptive histogram equalization. By simplifying the calculation process of ORB feature descriptors, the algorithm’s processing speed increases by 57% compared to the traditional SIFT method, meeting the timeliness requirements of industrial monitoring. In four typical industrial scenarios, the pixel matching accuracy reaches 85.23%, 78.34%, and 83.56% in rotation, scale, and lighting variation scenarios. Compared with gyroscope technology and traditional ORB algorithm, the peak signal-to-noise ratio of the proposed method is improved by an average of 2.56 dB and 1.67 dB, respectively. The above results indicate that the proposed method has better image denoising and robustness effects in real industrial environments than traditional video image processing techniques and is of great significance for maintaining the clarity of video surveillance images in harsh industrial environments.