Toward Elderly-Care-Oriented Smart Home HAR via Sensor-Aligned Activity Taxonomy

Edge-Based ADL Recognition Using Room-Specialized Mixture-of-Experts

From Inertial to Ambient: A Systematic Review of Sensor Modalities and Data Acquisition Strategies in Human Activity Recognition

- Human Activity Recognition (HAR) is a rapidly advancing research domain with transformative applications across healthcare, smart homes, rehabilitation, elderly monitoring, sports analytics, and context-aware computing. The performance and deployability of HAR systems are fundamentally determined by the sensor modalities through which human motion and behavioral data are captured. This systematic review provides a comprehensive examination of the full spectrum of sensor modalities employed in HAR research spanning wearable inertial sensors, physiological sensors, vision-based sensors, depth cameras, ambient sensing systems, and emerging multimodal fusion frameworks tracing, technical characteristics, application domains, and comparative strengths and limitations. The review further examines data acquisition protocols, benchmark datasets, and the persistent challenges of inter-subject variability, sensor placement sensitivity, privacy constraints, and computational efficiency that continue to shape the field. Emerging directions including federated sensing, edge-optimized data acquisition, self-supervised learning from unlabeled sensor streams, and privacy-preserving modalities are critically evaluated as promising pathways toward robust, scalable, and ethically responsible HAR deployment. By synthesizing evidence across more than a decade of HAR sensor research, this review provides a structured reference for researchers and practitioners designing next-generation activity recognition systems, and identifies the most consequential open challenges and future directions for the field.

Privacy-Preserving Ambient Sensing for Activities of Daily Living: Multimodal Radar–Thermal Human Activity Recognition and Smart Plug Appliance Recognition

Continuous monitoring of Activities of daily living (ADLs) requires sensing systems that are privacy-preserving, low-power, and robust to environmental variation. Ambient sensing technologies provide an alternative to RGB video and wearable devices, but individual sensing modalities exhibit characteristic limitations. Sparse mmWave radar provides strong motion sensitivity but limited posture detail, low-resolution thermal sensing preserves posture-related spatial information, and smart plug telemetry captures only appliance-mediated behavioural interaction. To address these limitations, this paper proposes a layered multimodal ambient-sensing framework comprising a sparse-track 24-GHz FMCW radar, a 32×24 low-resolution thermal sensor, and a Moko smart plug. It experimentally evaluates a radar–thermal HAR branch together with a separate smart plug appliance-recognition branch. The framework proposes three streams to enable continuous non-wearable monitoring while maintaining redundancy and reduced privacy exposure for intelligent-building and ambient assisted living environments. Radar and thermal streams are jointly evaluated on binary motion and four-class posture and activity recognition tasks collected across multiple environmental configurations using recording-grouped cross-validation, while the appliance stream is evaluated using per-plug telemetry from residential-grade appliances. The radar–thermal streams use a single-subject, fixed-placement dataset of binary-motion windows and four-class posture and motion windows collected across six furniture configurations. The separate intrusive load monitoring stream utilises smart plugs to classify appliances. Regarding binary motion recognition, radar (F1,Transformer=0.882±0.034) and thermal (F1,XGBoost=0.870±0.069) pipelines achieved similar macro F1 performance. On the four-class posture and activity recognition task, thermal features (F1,thermal=0.775±0.053) substantially outperformed radar (F1,radar=0.609±0.110). Weighted late fusion produced only modest descriptive gains. Separately, smart plug telemetry demonstrated strong appliance recognition performance using lightweight tree-based models suitable for constrained edge deployment. The results support a scoped redundancy argument. Sparse track-level radar carries gross motion, while low-resolution thermal sensing carries posture. The smart plug appliance monitoring extends the framework toward appliance-mediated instrumental activity of daily living (IADL) monitoring, with lightweight tree-based models achieving strong recognition performance under constrained edge deployment conditions. The findings support a layered multimodal sensing architecture for privacy-preserving ADL monitoring, where radar contributes motion-sensitive coverage, thermal sensing contributes posture-aware spatial context, and smart plug telemetry contributes appliance-level behavioural evidence within intelligent healthcare and ambient assisted living environments.

A Semi-Supervised CPC-Transformer Approach for Human Activity Recognition Under Label Scarcity

A semi-supervised approach that integrates Contrastive Predictive Coding (CPC) with a hybrid BiGRU-Transformer architecture is introduced, thereby enabling comprehensive temporal modeling for human activity recognition in smart-home environments.

Event-Driven Multimodal Sensing and Computing for Context-Aware Home Monitoring Using Stereo Vision and Dietary Event Anchoring

Real-world home monitoring requires sensing systems that can capture daily behaviour without continuous raw-video retention or excessive user burden. However, domestic environments present irregular activity timing, fragmented human presence, asynchronous multimodal events, and privacy-sensitive data management. This study proposes an event-driven multimodal sensing and computing framework for context-aware home monitoring using stereo vision and dietary event anchoring. The framework integrates stereo RGB-based three-dimensional human motion sensing, dining-zone-triggered meal image acquisition, runtime event orchestration, timestamp-based cross-modal synchronization, privacy-aware local storage, and large-language-model-assisted dietary context interpretation. Instead of continuously recording all sensor streams, the system activates and organizes sensing through human presence detection, debounce logic, cooldown-based session control, and dining-zone occupancy events. Meal-related events are used as contextual anchors to associate motion sessions and dietary observations into synchronized behavioural episodes. The prototype was deployed for 11 consecutive days in a real kitchen–dining environment, with the stabilized real-time monitoring phase evaluated from 11 to 14 February 2026. During this phase, the system generated 26 event-driven motion sessions and 51,165 captured pose frames, of which 25,925 were valid. Sustained active sessions accounted for 30.8% of all sessions but contributed 81.5% of captured pose frames, indicating that event-driven orchestration concentrated motion data within behaviourally meaningful activity windows. Eight meal-related records were obtained, seven of which overlapped with motion sessions, resulting in 87.5% meal-event overlap coverage. Structured pose outputs required approximately 550 kB/min, corresponding to about 33 MB/h of recorded pose data. LLM-assisted meal-image interpretation achieved a mean absolute percentage error of 25.44%, supporting its use for coarse dietary-context description rather than precise nutritional quantification. However, this result is interpreted only as evidence for coarse dietary-context description and not as validation of a precise nutritional or clinical dietary assessment method. These results demonstrate the system-level feasibility of transforming irregular domestic observations into structured, temporally indexed, and privacy-aware multimodal behavioural records for future home monitoring applications.

RAG-HAR+: Towards Cost-Efficient LLM-Based Human Activity Recognition for Edge Deployment

This work introduces RAG-HAR+, a retrieval-first and cost-optimized extension that strengthens retrieval while reducing dependence on LLM-based inference, and extends the RAG-HAR mobile prototype to demonstrate the practical feasibility of retrieval-first, LLM-assisted HAR in mobile sensing scenarios.