Dynamic SceneNet Spatio-Temporal Interactive Network for Dynamic Video Foreground Detection

Abstract

This study introduces Dynamic SceneNet, a new Spatio-Temporal Interactive Network developed to tackle the issues of dynamic video foreground recognition. Proposed model uses sophisticated deep learning techniques to learn interactive spatiotemporal characteristics, resulting in a novel solution for accurate and adaptable foreground recognition in dynamic scenarios. Video foreground detection (VFD) is a crucial pre-processing step for accurate target tracking and recognition. Creating a reliable detection network is tough because to interference from shadows, changing backgrounds, and camera jitter. Convolution neural networks have shown to be reliable in several sectors due to their great feature extraction capabilities. This research proposes an interactive spatiotemporal feature learning network (ISFLN) for VFD.. CDnet2014, INO, and AICD datasets corroborate that the introduced ISFLN model surpasses state-of-the-art methodologies in video foreground detection, thanks to its prowess in feature enhancement, interactive multi-scale feature exploration, and deep learning techniques.