Summary: | 碩士 === 元智大學 === 資訊工程學系 === 106 === Modern video surveillance benefits greatly from advanced wireless imaging sensors and cloud data storage; thus, considerable video footage can be generated every second.
Surveillance videos have thus become one of the largest sources of unstructured data. Because many surveillance videos are continuously and quickly produced, using such multiscenario videos for detecting moving objects is a challenging task for users of conventional moving object detection methods. This thesis presents a novel model that harnesses both sparsity and low rankness with contextual regularization to detect moving objects for multiscenario surveillance data. For our model, we not only consider moving objects as the contiguous outlier detection problem by utilizing the low-rank constraint with contextual regularization, but also construct backgrounds for multiple scenarios by using dictionary learning-based sparse representation, which ensures that our model works effectively for multiscenario videos. Quantitative and qualitative assessments indicated that the proposed model
significantly outperformed existing methods and also achieved substantially more robustness performance than did existing state-of-the-art moving object-detection
methods.
|