REINFORCING QUERY-BASED SURVEILLANCE SYNOPSIS SECURITY THROUGH DISTRIBUTED COMPUTING

초록

Video synopsis facilitates the efficient analysis of surveillance footage by condensing extended video content into shorter segments and reorganizing moving objects temporally based on a predefined objective function. Various approaches have been developed to address challenges in video synopsis, such as object collisions, maintaining chronological order, reducing computational time, and managing scene complexity by grouping activities based on user-defined queries. However, many query-based detection techniques often overlook specific tasks like detecting intrusions, such as unauthorized entry, identifying individuals wearing unauthorized clothing, or detecting person substitution. This research extends one of the conventional query-based approaches by characterizing each object tube through its motion and visual features to enhance the detection of such intrusions. Moreover, it implements query-based video synopsis generation using a distributed architecture that utilizes multiple servers to boost performance. While the distributed model effectively mitigates and improves the query-based computational performance, it introduces potential security risks, including object tube substitution during cross-device transfers and the possibility of tube loss. The proposed system contributes by building a distributed system architecture that reduces computational complexity and provides a comprehensive solution that balances between efficiency, security, and reliability in handling surveillance video synopsis. It offers a two-layer security model; the first layer embeds a watermark within each object tube to safeguard against substitution during processing. The second layer serializes each object tube, ensuring no loss occurs during transmission, thus preserving the integrity of the entire process. This novel approach integrates distributed computing, watermarking, and serialization to enhance both the efficiency and security of the video synopsis process, providing a robust solution for large-scale surveillance video analysis. The simulation results indicate that the security module effectively extracts the watermark with a reasonable PSNR and identifies any tube loss. Additionally, the analysis reveals that utilizing a distributed system greatly enhances performance, which is critical for real-time applications. Specifically, the proposed distributed model reduces computation time by approximately 94% compared to sequential execution when implemented across twenty machines. © Little Lion Scientific.

키워드