Leakage Detection in Surveillance Videos: A Robust Algorithm Combining Pixel-Wise and Region-Wise Approaches

In the process of extracting potential leakage candidates, the algorithm first establishes a background model. Based on this established background, it then extracts dynamic potential leakage foreground candidates. The algorithm utilizes a combination of pixel-wise and region-wise approaches to detect potential leakage candidates. The pixel-wise approach identifies potential leakage candidates by analyzing the differences between the current frame and the background model at each pixel location. The region-wise approach identifies potential leakage candidates by analyzing the spatial and temporal characteristics of the foreground regions.

Furthermore, to reduce false positives, the algorithm employs a series of post-processing techniques, including morphological operations and temporal filtering. The morphological operations are used to remove small isolated regions and fill in gaps within the detected foreground regions. The temporal filtering is used to remove foreground regions that persist for too short or too long a period of time.

Overall, the proposed algorithm demonstrates promising results in detecting potential leakage candidates in surveillance videos. Its combination of pixel-wise and region-wise approaches, along with post-processing techniques, provides a robust and effective solution for identifying potential leakage candidates.