Improving Safety for Disabled and Elderly Individuals: A Multimodal Classification Approach Based on Support Vector Machine for Alert Systems Within Smart Homes
Parsa Parsafar, Payam Qaderi Baban, Adel Nasiri · IEEE Instrumentation & Measurement Magazine · 2024
This article investigates the detection of disabled and elderly individuals' activity within smart homes by monitoring residents' behavior in the home environment. The ultimate objective is to construct a comprehensive behavioral model by leveraging data stored in the building management system's database, frequently gathered through the sensors outlined in the paper. This model will be employed for an alarm and notification system, fulfilling the safety of the residents. More precisely, our advanced intelligent system is adept at discerning the specific activities of a resident among seven distinct conditions. These conditions encompass a range of daily actions, including toilet or bathroom use, eating, resting, sleeping, communication, dressing and danger. For instance, the proposal is exploited to track sleeping habits, which provide an expressive, yet indirect, health-related indicator. The system achieves this by analyzing temporal data gathered from both the resident's body and various sensors strategically placed throughout the home and the pre-established behavioral model. The discussion revolves around employing well-suited sensors on the residents' bodies to monitor the vital signs, accurate data processing techniques within the proposed system implemented in the building, and also an efficient and accurate behavior classification. Extensive research was conducted in three distinct stages. The paper focuses on the selection of sensors and processing techniques, with a particular emphasis on the activities performed by disabled and elderly individuals in smart homes. The collected data was classified and processed utilizing Support Vector Machine (SVM) algorithms. Lastly, the optimal kernel and SVM training rule were determined through the implementation of the Particle Swarm Optimization (PSO) algorithm, resulting in the formation of the final pattern for storage in the building management system's database.