The goal of this paper's research is to develop learning methods that promote the automatic analysis and interpretation of human and mime-gestural movement from various perspectives and using various data sources images, video, depth, mocap data, audio, and inertial sensors, for example. Deep neural models are used as well as supervised classification and semi-supervised feature learning modeling temporal dependencies, and their effectiveness in a set of tasks that are fundamental, such as detection, classification, and parameter estimation, is demonstrated as well as user verification. A method for identifying and classifying human actions and gestures based on utilizing multi-dimensional and multi-modal deep learning from visual signals (for example, live stream, depth, and motion - based data). A training strategy that uses, first, individual modalities must be carefully initialized, followed by gradual fusion (called ModDrop) to learn correlations between modalities while preserving the uniqueness of each modality specific representation. In addition, the suggested ModDrop training approach assures that the classifier detect has weak inputs for one or maybe more channels, enabling these to make valid predictions from any amount of data points accessible modalities. In this paper, inertial sensors (such as accelerometers and gyroscopes) embedded in mobile devices collect data are also used.

Two-handed dynamic gesture recognition represents a fundamental component of sign language interpretation involving the modeling of temporal dependencies and inter-hand coordination. In this task, a major challenge is modeling asymmetric motion patterns, as well as bidirectional and long-range temporal dependencies. Most existing frameworks rely on early fusion strategies that merge joints, keypoints or landmarks from both hands in early processing stages, primarily to reduce model complexity and enforce a unified representation. In this work, a novel dual-stream BiLSTM–Transformer model architecture is proposed for two-handed dynamic sign language recognition, where parallel encoders process the trajectories of each hand independently. To capture spatial and temporal dependencies for each hand, an attention-based cross-hand fusion mechanism is employed, with hand landmarks extracted by the MediaPipe Hands framework as a preprocessing step to enable real-time CPU-based inference. Experimental evaluation conducted on custom Romanian Sign Language dynamic gesture datasets indicates that the proposed dual-stream-based system outperforms single-handed baselines, achieving improvements in high recognition accuracy for asymmetric gestures and consistent performance gains for synchronized two-handed gestures. The proposed architecture represents an efficient and lightweight solution suitable for real-time sign language recognition and interpretation.

Due to their vast potential, Unmanned Aerial Vehicles (UAVs) are increasingly being utilized in various applications. To prevent in-flight failures and loss of control, implementing Internet of Things (IoT)-based Predictive Maintenance (PdM) systems is crucial. However, data collected from PdM systems often contains outliers, which can significantly degrade the accuracy and performance of predictive models. In this paper, we present a comparative performance analysis of several outlier detection methods, namely K-Nearest Neighbors (KNN), Autoencoder (AE), and Isolation Forest (IForest). The datasets used to evaluate these methods were acquired from a UAV predictive maintenance system designed to estimate the Remaining Useful Life (RUL) of Li-ion batteries and detect faults in Brushless DC (BLDC) motors. Ultimately, this study aims to determine the most effective outlier detection method for UAV predictive maintenance datasets.