This paper presents a robust and efficient mmWave radar-based human activity recognition (HAR) framework optimized for practical real-time indoor deployment. Addressing computational inefficiencies and limited recognition scopes in existing systems, the framework introduces two core contributions: Multi-class Spatio-Temporal Network (MuST-Net), a lightweight, multi-class network, and an online detection process for enhanced temporal stability. MuST-Net utilizes a hybrid 2D convolutional neural network and temporal convolutional network architecture to recognize seven distinct classes, significantly broadening the system’s recognition repertoire. The online detection process implements a novel sliding-window post-processing chain that employs an activity-buffering mechanism, which maintains temporal continuity and effectively suppresses spurious detections at activity boundaries. Experimental results demonstrate the superior performance of our unified framework, attaining over 98.6% accuracy for multi-class classification by MuST-Net and achieving at least 97% accuracy for activity detection and a crucial 100% recall for fall detection. Robustness is validated across three distinct indoor environments and nine subjects—with two of the three sites entirely unseen during training—confirming strong generalization under installation, environment, and subject variations.

Atacr-net: adaptive temporal alignment and contrastive refinement network for skeleton-based action recognition

Accurate identification of dangerous driving behaviors is critical for accident prevention and occupant protection. However, most existing in-vehicle driver monitoring systems rely primarily on facial or head motion analysis, which fails to capture full-body driving behaviors and raises privacy concerns due to dependence on RGB or near-infrared imaging. In addition, these systems often exhibit limited robustness under low-light conditions. To address these limitations, this study proposes a comprehensive depth-based framework for in-vehicle 3D human pose estimation and dangerous driving posture recognition. First, a large-scale dual-view 3D pose dataset encompassing ten typical driving behaviors is constructed using a Time-of-Flight (ToF) camera. Based on this dataset, we develop a lightweight end-to-end pipeline in which an anchor-based regression model estimates the 3D poses of 16 driver keypoints, followed by an enhanced ST-GCN++ architecture for skeleton-based action recognition. By integrating pose estimation with graph-based temporal modeling, the proposed method effectively distinguishes visually similar hazardous behaviors. To facilitate real-world deployment, the algorithm is further integrated into a software system that enables closed-loop pose monitoring and hierarchical intervention. Experimental results verify that the proposed method achieves 96.02% accuracy in 3D pose estimation and 98.0% accuracy in behavior recognition. With a computational cost of only 1.49 G FLOPs and an inference latency of 0.0375 s per sample, the system achieves real-time performance (27-28 FPS) on an automotive embedded platform, making it well suited for practical in-vehicle safety applications.

This study introduces a lightweight Human Action Recognition (HAR) model designed for computational efficiency and real-world applications. Faced with the challenge of processing large scale video data, the proposed approach strategically selects only the most informative keyframes, thereby, significantly reducing data redundancy. The model leverages a high-performing pre-trained DaViT backbone for feature extraction, combined with a Temporal Transformer that effectively captures both spatial details and temporal dynamics from sparse keyframes. The proposed method reduces the high computational cost associated with traditional architectures such as 3D CNNs, LSTMs that process every single frame. To validate its practical utility, the proposed model was deployed on a quadruped robot, establishing an efficient inference pipeline in which the robot captures video and performs on-device action recognition. The proposed method demonstrates a significant step towards applying complex HAR tasks in resource-constrained, robotic environments.

Abstract The demand for accurate on-device pattern recognition in edge applications intensifies, yet existing approaches struggle to reconcile accuracy with computational constraints. To address this critical challenge, a resource-aware hierarchical network based on multi-spectral fusion and interpretable modules, namely the Hierarchical Parallel Pseudo-Image Enhanced Fusion Network (HPPI-Net), is proposed to enable real-time, on-device Human Activity Recognition (HAR) tasks. Deployed on ARM Cortex-M4 MCU for low-power real-time inference, HPPI-Net achieves 96.70% accuracy while utilizing only 22.3 KiB of RAM and 439.5 KiB of ROM after optimization. HPPI-Net employs a two-layer architecture: the first layer extracts preliminary features using Fast Fourier Transform (FFT) spectrograms, while the second layer selectively activates either a dedicated module for stationary activity recognition or a parallel LSTM-MobileNet network (PLMN) for dynamic states. PLMN fuses FFT, Wavelet, and Gabor spectrograms through three parallel LSTM encoders and refines the concatenated features with Efficient Channel Attention (ECA) and Depthwise Separable Convolution (DSC), thereby offering channel-level interpretability while substantially reducing multiply–accumulate operations. Compared to MobileNetV3, HPPI-Net notably increases accuracy by 1.22% and significantly reduces RAM usage by 71.2% and ROM usage by 42.1%. These results demonstrate that HPPI-Net realizes a favorable accuracy-efficiency trade-off and provides explainable predictions, establishing a practical solution for wearable, industrial, and smart-home HAR on memory-constrained edge platforms.

Traumatic brain injury (TBI) is a major cause of blood-brain barrier (BBB) disruption and neurological dysfunction, wherein endothelial dysfunction plays a critical pathogenic role. As a member of the G protein-coupled receptor family, sphingosine-1-phosphate receptor 2 (S1PR2) is known to regulate vascular homeostasis; however, its specific role in protecting the blood-brain barrier following TBI remains unclear. This study aims to elucidate the mechanism by which S1PR2 maintains blood-brain barrier integrity and to evaluate the therapeutic potential of S1PR2 inhibition after TBI. A mouse model of TBI was established using controlled cortical impact, while human umbilical vein endothelial cells (HUVECs) were subjected to oxygen-glucose deprivation/reoxygenation (OGD/R) to mimic ischemia-reperfusion injury in vitro. We employed shRNA technology to knock down S1PR2 expression and utilized single-cell RNA sequencing (dataset GSE269748) to characterize cell type-specific expression profiles. Endothelial function, blood-brain barrier permeability, inflammatory responses, and cell apoptosis were assessed using tube formation assays, transendothelial electrical resistance (TER) analysis, Western blotting, immunofluorescence, qPCR, ELISA, Evans blue staining, and brain water content measurements. Behavioral tests including open field test and novel object recognition test were used to evaluate the recovery of neurological function. At the same time, the PI3K-AKT pathway was interfered by S1PR2 knockdown mediated by AAV virus and pharmacological inhibitor (JTE-013/LY94002) or activator (Cyn). Single-cell analysis revealed that S1PR2 is specifically expressed in endothelial cells and is significantly upregulated following TBI. In vitro, S1PR2 knockdown counteracted the OGD/R-induced reduction in tube formation capacity and the elevation in transendothelial electrical resistance, and restored the expression of tight junction proteins Occludin and ZO-1. RNA-seq and KEGG enrichment analysis suggested that PI3K-AKT pathway was the key downstream target of S1PR2. In vivo experiments demonstrated that S1PR2 expression peaked at 72h post-TBI and colocalized with CD31, while the ratios of p-PI3K/PI3K and p-AKT/AKT were markedly reduced. Intervention targeting S1PR2 significantly enhanced locomotor activity and novel object recognition, reduced brain lesion area, suppressed neuronal apoptosis and inflammatory cytokine levels, and restored BBB integrity in TBI mice. Mechanismally, activation of PI3K-AKT pathway could mimic the protective effect of S1PR2 knockdown, whereas inhibition of this pathway negated the improvements in BBB integrity and neurological function induced by S1PR2 knockdown. Endothelial S1PR2 is a critical regulator of vascular homeostasis after TBI. Inhibition of Endothelial S1PR2 preserves blood-brain barrier integrity, mitigates neuroinflammation and apoptosis, and promotes neurological recovery through activation of the PI3K-AKT signaling pathway, thereby offering a promising new strategy for targeted TBI therapy.

A versatile colorimetric array sensor based on FeCu-based prussian blue analogue heterostructures for efficient discrimination of tetracycline antibiotics and rapid identification of urinary tract infection pathogens.

The field of sensor-based human activity recognition (HAR) mainly uses posture, motion and context data of Inertial Measurement Units (IMUs) to identify daily activities. Despite the advancements in learning-based methods, it is challenging to perform information fusion from the temporal perspective due to the complexities in fusing heterogeneous sensor data and establishing long-term context correlations. This paper proposes a novel triple spectral fusion framework tailored for HAR. First, we develop an adaptive complementary filtering technique for noise suppression and organize each IMU's sensors into posture and motion modality nodes. Given that IMU nodes form a dynamic heterogeneous graph, we then apply adaptive filtering within the graph Fourier domain to merge both homogeneous and heterogeneous node information. Furthermore, an adaptive wavelet frequency selection approach is implemented to suppress context redundancy and shorten the length of features. This approach enhances both timestamp-based graph aggregation and the correlation of long-term contexts. Our framework uses adaptive filtering in the Fourier, graph Fourier, and wavelet domains, enabling effective multi-sensor fusion and context correlation. Extensive experiments on ten benchmark datasets demonstrate the superior performance of our framework. Project page: https://github.com/crocodilegogogo/TSF-TPAMI2026.

Robot therapy, with PARO being a prominent, has become an effective intervention in elderly care, where caregivers play an important role in patient engagement during therapy. This study analyzes caregiver behavior to reveal behavioral parameters associated with caregiving skill in delivering robot therapy using PARO. The analysis parameters included physical activity of caregivers and patients during therapy, facial gaze, eye level between caregiver and patient, and PARO face orientation. Using a vision-based activity recognition system, observational data were collected from 24 novice caregivers (12 trained, 12 untrained) and 6 elderly participants, with statistical analysis (Mann-Whitney U tests) confirming significant differences in interaction behavior between trained and untrained caregivers. The results showed that trained caregivers demonstrated specific technical skills, including holding PARO longer during introductions, delaying the initial placement of PARO, sustaining facial gaze, maintaining longer eye contact, and dynamically adjusting PARO's position based on patient cues. Although trained caregivers tended to maintain a smaller eye-level difference with patients on average, this difference was not statistically significant. These findings indicate that formal caregiving education is associated with systematic differences in interaction behavior, even at a pre-professional stage. The study contributes a quantitative framework for identifying foundational behavioral markers related to caregiving skill in robot therapy, rather than evaluating clinical efficacy.

Laparoscopy is one of the most widely used surgical techniques in clinical practice. However, its practice is associated with medium- and long-term musculoskeletal disorders in surgeons. In this context, robot-assisted surgery has emerged as a promising approach for mitigating ergonomic constraints while enhancing control and precision during laparoscope manipulation. Despite these advances, existing research predominantly focuses on robotic control strategies, whereas the study of human–robot interaction in the operating room remains comparatively underexplored. This paper presents a proof-of-concept framework for workspace-aware posture adaptation in collaborative surgical robotics. The proposed approach combines vision-based human activity recognition with reinforcement learning to control the shoulder–elbow–wrist redundant angle of a seven-degree-of-freedom manipulator holding a laparoscope. Based on the detected interaction context, the system distinguishes between controlling, observing, cutting, and blocked states. During the observation and cutting phases, the controller allows the robot’s posture to be reconfigured so that it tilts away from the human operator while maintaining the position of the laparoscope; when the surgeon moves away, the robot gradually returns to its default configuration. Two reward formulations, dense and fuzzy, are compared. Real-world experiments show that both approaches learn the desired reflexive behavior, while the fuzzy reward yields improved training stability and more consistent real-system performance, increasing workspace availability around the surgeon.

Recent systems for position and movement monitoring are increasingly enhanced by the integration of advanced AI techniques. Besides traditional analytical and algorithmic approaches, which are still applied in areas such as signal processing, sensor fusion, and kinematic modeling, there is a growing body of research that leverages AI-based methods to improve accuracy, robustness, and real-time decision-making capabilities. Artificial neural networks and deep learning methods are more and more often used for tasks such as predicting movement trajectories, detecting position anomalies, and approximating complex motion patterns. The main aim of this work is to provide the main contributions of the recent publications to the current state of the field. Key trends, challenges, and prospects for their future development are also highlighted. Initial statistical analysis was conducted based on responses to queries formulated for searching engines of leading online databases since 2006. Next, the retrieved articles from the last 6 years were subjected to a more detailed analysis. They were divided into thematic areas, including models for human pose estimation; systems for motion detection and tracking, with special attention to human movement; and, eventually, more specialized applications such as action recognition, autonomous driving, motion analysis, and surveillance. The architectures of the created models, the methods for parameter tuning or training, the input datasets used, and the result evaluation metrics were classified. Finally, some more general conclusions were drawn.

HighlightsWhat are the main findings?Indoor human activities can be inferred from indoor air quality measurements.A gated hierarchical model separates activity detection from classification, achieving high activity detention recall and conditional classification accuracy.What are the implications of the main findings?Indoor pollutant dynamics contain sufficient information to enable unobtrusive activity recognition based on environmental sensing only.The proposed framework supports intelligent control applications, such as adaptive ventilation and health-aware indoor environmental control.Recent studies have shown that common household activities produce characteristic patterns in indoor air pollutants, enabling activity inference using environmental measurements alone. However, pollutant-based approaches are usually formulated as flat multi-class classification problems, even though indoor environments are dominated by long baseline periods with no emission-generating activity, leading to false alarms and unstable predictions. This work proposes a gated hierarchical inference framework for recognizing activities from indoor air quality data. A first-stage gate detects whether a time window contains activity-induced pollutant dynamics, while a second-stage classifier conditionally identifies the specific activity only when activity relevance is detected. Multivariate time-series measurements of particulate matter, volatile organic compounds, nitrogen oxides, carbon dioxide, temperature and relative humidity were collected using a portable monitoring system during controlled household cooking and cleaning experiments. Temporal windows were processed using recurrent neural network models in both stages. By separating activity detection from activity identification, the proposed method aligns inference with the physical generation of indoor pollutant signals and improves robustness in baseline-dominated monitoring scenarios while maintaining reliable discrimination among activities. The framework supports unobtrusive activity recognition and enables applications in exposure-aware monitoring and intelligent indoor environmental management.

Organizational well-being and workplace happiness are essential pillars for productivity and organizational sustainability. This study presents an integrative literature review of theoretical models on workplace well-being, aiming to provide a holistic framework applicable to organizational context. Contributions from positive psychology, human talent management, and organizational culture were analyzed, highlighting the critical role of mental, emotional and social health as core components of well-being. Findings indicate that sustained practices of active participation, recognition, effective communication, and emotional management enhance job satisfaction, intrinsic motivation and organizational performance. Based on this synthesis, an intervention model is proposed to foster healthy, inclusive and resilient work environments. Overall, the study demonstrates that integrating well-being as strategic organizational focus not only promotes human development and strengthens organizational culture but also contributes to achieving sustainable outcomes and consolidating competitive, people centered organizations.

ASR-GCN: Adaptive spatial information reconstruction GCN for skeleton-based action recognition.

Cyclopropyl as a versatile tool in the development of kinase-targeted therapeutics.

Investigating Shared Mechanisms of Arrestin Protein Activation and Cargo Recognition (Abstract ID: 235956)

STTA-SlowFast: Spatial Temporal Triplet Attention SlowFast for pilot action recognition

FMFNet: A Faster Multimodal Fusion Network for action recognition via efficient modality compensation

TE-STGCN: Topology enhanced spatio-temporal graph convolutional network for skeleton-based action recognition

Multimodal alignment of event and text streams in spiking neural networks for human action recognition