A Hybrid Model for Human Activity Recognition to Support Tracking of Inhabitants and Activities in a Simulated Multi- Inhabitant Smart Environment
Year: 2023
Type: article
Abstract: Smart Environments (SEs) employed as testbeds consist of sensors and applications formulated to develop a comfortable and safe environment for their inhabitants. Despite the increase in quantities of SE, the accessibility of these spaces for researchers has become even more challenging due to the challenges of sensors, features, algorithms, and implementation. The recognition of Activities of Daily Living is the most important theme and input to a variety of SE applications (e.g., healthcare and home security). Recognising activity from human motion patterns and sensors is difficult owing to the complexity of human life in SEs, whether by single or numerous occupants. Another significant obstacle to overcome is lowering the cost of maintaining and installing sensors in SE. Despite their abundance, these sensors are difficult to place and costly for most people. In addition to problems with sensors and their placement, it is challenging to optimise their configuration. This study adopted a simulation tool, the Smart Environment Simulator (SESim), to generate Human Activity data using multiple script-regulated inhabitants to address this obstacle. Sensor readings are recorded and saved in the dataset which is evaluated in the model for Human Activity Recognition (HAR) suggested in this work while the multiple inhabitants perform activities in the SE. A proof-of-concept Human Activity simulation system, the Extended Smart Environment Simulator (eSESim) was developed and used for dataset generation. The dataset generated was evaluated using experimental models. This study proposes a hybrid model for HAR to support Inhabitant Tracking and Activity Status determination in a SE by investigating Machine Learning and Deep Learning models. The first strategy of this study is based on recognising Human Activities utilising sensors (motion, temperature, presence, light, and interaction), Machine Learning, and deep neural networks to compare findings and quantify the impact of iii each on learning accuracy. The second approach is based on monitoring inhabitants in the Smart Environment to determine the most accurate classifier suitable for Inhabitant Tracking and identification. Once this is established, the next phase is to track Activity Status via clustering on the SESIM Dataset. Each component of this study’s methodology underwent extensive testing using different Machine Learning and Deep Learning techniques, with the outcomes discussed. The results from this study show that the Hybrid Ensemble model for HAR is suitable for improving accuracy in Human Activity classification. Secondly, the results indicate that HAR is possible using a Synthetic Dataset to support inhabitant and activity status tracking. Thirdly this study gives an insight into how Machine Learning techniques can be used to track inhabitants and their Activity Status in a SE.
Author: Musharu, TN
Language: English
Subfield: Computer Vision and Pattern Recognition
Field: Computer Science
Domain: Physical Sciences
Open Access status: green
Funder: Nelson Mandela University
OpenAlex ID: W7220761960
