Research on Sensitive Frame Recognition Algorithm for Home Monitoring Video Based on Mixture of Gaussians Background Modeling
Yiting Yin, Jingxu Zhang, Liping Zhu, Guodong Li · 2023
Surveillance videos have become ubiquitous in our daily lives, and to mitigate the risk of information leakage during the upload process to the cloud, privacy protection is of utmost importance. This paper proposes an application scenario based on home surveillance videos, mainly using foreground-background separation technology to protect the privacy area in surveillance videos. In home surveillance videos, privacy information mainly refers to people in motion. The paper improves the traditional mixture of Gaussian model by replacing Gaussian distribution with t-distribution for background modeling and identifies moving targets in the video through foreground-background separation detection. Morphological operations are applied to the results of target detection to eliminate noise and ensure the accuracy of subsequent threshold judgment. Then, by setting appropriate thresholds, video frames with dense human activity are detected and considered as sensitive information frames. Finally, the sensitive information frames are extracted from the video stream and encrypted to protect the privacy information. Through simulation experiments, the improved mixture of Gaussian background modeling algorithm can effectively separate foreground and background in surveillance videos, improve the accuracy of moving target detection, and accurately identify sensitive information frames. By adjusting the size of the threshold, the number of sensitive information frames that need to be encrypted can be flexibly controlled, which enhances the encryption level of the video, effectively protects users' privacy information, and reduces the risk of information leakage.