PROJECT TITLE :

Moving Object Detection in Complex Scene Using Spatiotemporal Structured-Sparse RPCA

ABSTRACT:

The detection of moving objects is an essential part of many computer vision applications. RPCA-based approaches (robust principal component analysis) have frequently been used for this objective. Because of this, many approaches suffer when they encounter dynamic background scenes, camera jitter and moving objects that are difficult to spot. As a result, moving objects lose their spatiotemporal structure in the sparse component because of an underlying assumption that the pieces are independent. RPCA technique for moving object detection, we offer a spatiotemporal structured sparse RPCA algorithm with spatial and temporal regularisation in the form of graph Laplacians to solve this problem. Each Laplacian is a multi-feature graph that is created using the input matrix's superpixels. In order to minimise the RPCA objective function, we enforce the sparse component to operate as eigenvectors of the spatial and temporal graph Laplacians Spatiotemporal subspace structure is incorporated into the sparse component of this constraint. Thus, a novel goal function for discriminating moving objects from complex backdrops is obtained. A linearized alternating direction method of multipliers-based batch optimization is used to solve the given objective function. In addition, we offer a real-time online optimization technique. With the use of six publicly available datasets, we were able to test both batch and online methods. Our results show that the proposed algorithms outperform current state-of-the-art approaches in terms of performance.