Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

Integrating Temporal Context into Streaming Data for Human Activity Recognition in Smart Home

Marina Vicini, Martin Rudorfer, Zhuangzhuang Dai, Luis J. Manso

TL;DR

This study targets aging-in-place HAR in smart homes using PIR and door sensors, addressing the underutilization of temporal context. It introduces Temporal Mutual Information with day-based segmentation, along with an expanded feature space that includes cyclical time encoding and a location-change flag, and computes per-segment MI weights offline. Evaluations on four CASAS datasets show improvements in accuracy and weighted F1 over state-of-the-art baselines, particularly in low-data regimes, demonstrating the value of temporal context for robust activity recognition. The work suggests practical benefits for telecare deployments with fewer sensors, enhancing timely intervention while maintaining privacy and low intrusion.

Abstract

With the global population ageing, it is crucial to enable individuals to live independently and safely in their homes. Using ubiquitous sensors such as Passive InfraRed sensors (PIR) and door sensors is drawing increasing interest for monitoring daily activities and facilitating preventative healthcare interventions for the elderly. Human Activity Recognition (HAR) from passive sensors mostly relies on traditional machine learning and includes data segmentation, feature extraction, and classification. While techniques like Sensor Weighting Mutual Information (SWMI) capture spatial context in a feature vector, effectively leveraging temporal information remains a challenge. We tackle this by clustering activities into morning, afternoon, and night, and encoding them into the feature weighting method calculating distinct mutual information matrices. We further propose to extend the feature vector by incorporating time of day and day of week as cyclical temporal features, as well as adding a feature to track the user's location. The experiments show improved accuracy and F1-score over existing state-of-the-art methods in three out of four real-world datasets, with highest gains in a low-data regime. These results highlight the potential of our approach for developing effective smart home solutions to support ageing in place.

Integrating Temporal Context into Streaming Data for Human Activity Recognition in Smart Home

TL;DR

This study targets aging-in-place HAR in smart homes using PIR and door sensors, addressing the underutilization of temporal context. It introduces Temporal Mutual Information with day-based segmentation, along with an expanded feature space that includes cyclical time encoding and a location-change flag, and computes per-segment MI weights offline. Evaluations on four CASAS datasets show improvements in accuracy and weighted F1 over state-of-the-art baselines, particularly in low-data regimes, demonstrating the value of temporal context for robust activity recognition. The work suggests practical benefits for telecare deployments with fewer sensors, enhancing timely intervention while maintaining privacy and low intrusion.

Abstract

With the global population ageing, it is crucial to enable individuals to live independently and safely in their homes. Using ubiquitous sensors such as Passive InfraRed sensors (PIR) and door sensors is drawing increasing interest for monitoring daily activities and facilitating preventative healthcare interventions for the elderly. Human Activity Recognition (HAR) from passive sensors mostly relies on traditional machine learning and includes data segmentation, feature extraction, and classification. While techniques like Sensor Weighting Mutual Information (SWMI) capture spatial context in a feature vector, effectively leveraging temporal information remains a challenge. We tackle this by clustering activities into morning, afternoon, and night, and encoding them into the feature weighting method calculating distinct mutual information matrices. We further propose to extend the feature vector by incorporating time of day and day of week as cyclical temporal features, as well as adding a feature to track the user's location. The experiments show improved accuracy and F1-score over existing state-of-the-art methods in three out of four real-world datasets, with highest gains in a low-data regime. These results highlight the potential of our approach for developing effective smart home solutions to support ageing in place.
Paper Structure (16 sections, 5 equations, 4 figures, 3 tables)

This paper contains 16 sections, 5 equations, 4 figures, 3 tables.

Table of Contents

  1. Introduction
  2. Related Works
  3. Data Segmentation
  4. Feature Extraction
  5. Classification
  6. Method
  7. Temporal Mutual Information
  8. Temporal Segmentation of the Day:
  9. Refined MI Calculation:
  10. Expanded Feature Space
  11. Cyclical Time Encoding:
  12. Location Change Feature:
  13. Experiments and Results
  14. Evaluation Procedure
  15. Results and Discussion
  16. ...and 1 more sections

Figures (4)

  • Figure 1: Data is acquired through devices in the users' smart homes. After segmenting the data, features are extracted and weighted with temporal Mutual Information matrices. Lastly, a multi-class classifier is employed. Activities can be monitored by service providers or family.
  • Figure 2: Different data segmentation techniques. Colours represent different sensors.
  • Figure 3: Visualization of the day of the week and hour of the day through sine and cosine transformations to capture their cyclical nature. This ensures that "Monday" is close to both "Sunday" and "Tuesday" in the feature space.
  • Figure 4: Normalized confusion matrix of test set results for the SWMI method with cyclic features and location change feature, for TM006 (top-left), TM007 (top-right), TM008 (bottom-left), and Aruba (bottom-right). The labels are: 0) Bathing, 1) Bed to toilet, 2) Cook / Meal Preparation, 3) Eat, 4) Enter home, 5) Leave home, 6) Personal Hygiene, 7) Relax, 8) Sleep, 9) Take Medicine, 10) Wash Dishes, 11) Work and 12) Other.