Computing Reviews
Today's Issue Hot Topics Search Browse Recommended My Account Log In
Review Help
Optimizing the configuration of an heterogeneous architecture of sensors for activity recognition, using the extended belief rule-based inference methodology
Espinilla M., Medina J., Calzada A., Liu J., Martínez L., Nugent C. Microprocessors & Microsystems52  381-390,2017.Type:Article
Date Reviewed: May 17 2018

The aging populations in many countries are on the verge of increasing, and the incidence of common diseases related to this group, such as dementia, seem to by rising day by day. Hence, there is a growing need to have sensor-based activity recognition for smart electronic devices that have the ability to record information with high in-processing capabilities for computing, and at the same time consider low power consumption. Such activity recognition aims to recognize more people with a series of observations of activities with multiple conditions. In such environments, it is not enough to deploy the appropriate sensors “to collect, store, and process information, [but also] to classify activities from sensor data through the use of computational activity models.” Such sensor-based activities have been divided into data-driven (DDA) and knowledge-driven (KDA) approaches.

DDA is based on machine learning, whereas KDA is based on prior knowledge available through knowledge engineering and knowledge management techniques. Hence, in DDA, “a training process is carried out [in order] to build an activity model, which is followed by testing processes to evaluate the generalization of the model in classifying unseen activities.” Since KDA uses prior knowledge, it is easy to get started with clear and logically sound classifiers. The disadvantage of KDA is having weak uncertainties and temporal information with static and competitive classifiers. DDA, due to the training process, requires large datasets for training and learning; besides, it suffers from scarcity or the cold start problem. As a result, the feature selection methods to provide a way to select a subset of relevant features to generate a classifier or to select a particular model from datasets obtained from real environments by having several sensors working together at the same time are the real challenge in this kind of research. Some studies have been proposed based on the hybrid extended belief rule-based (E-BRB) inference methodology approach known as RIMER+ and use evidential reasoning. Such approaches focus on feature selection methods to optimize the subsets of initial sensors for activity recognition, as well as to reduce the computational complexity involved in such studies. Some studies used 39 sensors and others used 59 sensors to collect data in a real environment. This particular research is called R-DRAH, similar to the adaptation of RIMER+ for activity recognition while tolerating some sensor failures during the activity for smart environments. The proposed method is evaluated against the popular DDA classifier selection approach.

The main aim of this research is to deploy low-cost sensors in a smart environment and, from the relevant data obtained from the sensors, to select a set of features to generate classifiers as well as to assist in obtaining better insight into the classification problem. The study used 14 sensors that collected datasets containing 245 observations for about 28 days in an apartment. Two optimizations were selected, where the first method chose ten sensors and the second one chose seven sensors and removed the rest. When compared with other studies, the proposed study’s classifiers had higher performance and an accuracy mean rate of 96.33 percent. If the authors increased the number of sensors and the collected datasets, would it produce the same result? Would it hold the same accuracy rate? That is yet to be seen.

Reviewer:  J. Arul Review #: CR146037 (1807-0401)
Bookmark and Share
  Featured Reviewer  
Would you recommend this review?
Other reviews under "Sensors": Date
Sensors for mobile robots
Everett H., A. K. Peters, Ltd., Natick, MA, 1995. Type: Book (9781568810485)
Sep 1 1996
Target detection and tracking by bionanosensor networks
Okaie Y., Nakano T., Hara T., Nishio S., Springer International Publishing, New York, NY, 2016.  68, Type: Book
May 31 2017
Sensor deployment strategy for detection of targets traversing a region
Clouqueur T., Phipatanasuphorn V., Ramanathan P., Saluja K. Mobile Networks and Applications 8(4): 453-461, 2003. Type: Article
Dec 11 2003

E-Mail This Printer-Friendly
Send Your Comments
Contact Us
Reproduction in whole or in part without permission is prohibited.   Copyright 2004™
Terms of Use
| Privacy Policy