The detection of video deepfake is an emerging task in computer vision that holds significant importance for internet content security. As deepfake technology advances and the proliferation of fake videos becomes more rampant, research on video deepfake detection techniques is particularly crucial. This paper begins by introducing representative algorithms for video deepfake and its detection, followed by a brief analysis. It then summarizes commonly used datasets, as well as related evaluation systems. Finally, it discusses the main challenges and future directions of video deepfake detection technologies.
The necessity for robust and secure data processing frameworks has been brought to light by the development of Internet of Things (IoT) devices. Federated learning (FL) has gained traction as a viable method for utilizing distributed intelligence while protecting data privacy, owing to the widespread use of IoT devices. In our previous study, we developed a method that leverages federated averaging (FA) and distributed intelligence by creating a Federated Virtual Sensors for IoT (FVSI) platform to facilitate remote data processing while maintaining localized data. However, improving the privacy and resilience of the FVSI architecture has become essential as IoT systems become more vulnerable to hostile manipulations and privacy intrusions. This paper delves into the security vulnerabilities inherent in FL systems, particularly within IoT networks. By including differential privacy (DP) strategies and creating strong defenses against adversarial assaults on FL models, this study aims to expand the FVSI framework. The project intends to prevent privacy breaches in shared data among FL models by implementing advanced DP techniques. We’ll use strategies like noise addition and privacy-preserving data aggregation to ensure privacy and data utility are balanced. Through extensive validation, we demonstrate that these enhancements maintain high model performance across various IoT scenarios while preventing data breaches. Our results illustrate notable gains in FVSI’s resilience, paving the way for secure, trustworthy IoT deployments.
In recent years, the topography of affected areas has undergone dramatic changes due to the increasing frequency of large-scale floods and landslides associated with heavy rainfall. Large-scale floods not only induce landslides and mudslides but also cause river breaches and widespread flooding, leading to significant landform modifications. In this context, accurate topographical information is essential for effective recovery planning and the prevention of disasters. Therefore, the automation and labor-saving measures for surveying operations have been advanced by the deployment of robots and drones. In addition, amphibious robots can flexibly conduct surveying and environmental monitoring in diverse topographical conditions. In this paper, we propose a two-step optimization method for wide-area monitoring using amphibious robot. We optimize the waypoint locations to maximize the monitoring area. Then, we optimize the travel route to minimize the electric energy consumption. We carried out simulations to evaluate the proposed method. The simulation results show that the proposed method increases the monitoring area and minimizes the robot’s electric energy consumption.
This paper proposes a new extended framework for web VR tours of Point Cloud Data that enables network collaborations. The authors have already developed frameworks for web VR tours in navigation mode using 360º videos and in walk-through mode using 360º images. Also, the authors developed a framework using OpenVSLAM for web VR tours using 360º images/videos. Furthermore, the authors extended the OpenVSLAM-based framework to enable network collaborations by adopting threejs-webrtc. Indeed, the authors have also developed a framework for web VR tours in walk-through mode using Point Cloud Data (PCD). This time, the authors also extended the framework to enable network collaborations by adopting threejs-webrtc. This paper introduces the framework for web VR tours of PCD and explains the extended framework for web collaborative VR tours of PCD.
Emotion classification systems have been investigated and developed for a long time due to their usefulness in various industrial applications. However, the estimation of emotions is a complex process. This paper proposes an emotion classification system considering Analytic Hierarchy Process (AHP). We evaluate the proposed system by comparing with Word2Vec-based emotions estimation system. The evaluation results have shown that the order of emotions in both systems are almost the same for the considered sentences. However, AHP has explainable feature and is suitable for managing individual emotions, which makes the proposed system applicable to different scenarios.
