Multi-Modal Fusion for Moving Object Detection in Static and Complex Backgrounds

Moving object detection from video sequences remains a focal point of research. To address the limitations evident in current methodologies, a synthesis of optical flow method and salient object fusion algorithm has been applied. Utilising the Graph-based Visual Saliency (GBVS) algorithm, significant target region signals from both static and dynamic images can be obtained. This technique captures valuable image target information, highlighting conspicuous targets within dynamic visuals. Concurrently, target signals can be isolated employing the Harmony Search (HS) algorithm, enhancing the accuracy in identifying moving objects. A weighted fusion of the extracted salient regions by the GBVS algorithm and the moving objects identified by the HS algorithm was executed in this study. This amalgamation demonstrates efficacy in extracting static objects in rudimentary environments and complex backgrounds alike. MATLAB simulation experiments have indicated that such a multi-modal fusion not only diminishes background noise but also proficiently isolates the entirety of the target. Building on traditional frame difference and background difference methods and considering the properties of the field programmable gate array (FPGA) alongside off-chip synchronous dynamic memory's access control prerequisites, adaptations for these algorithms were conceived using FPGA logic units.