Ontology-Based Framework for the Automatic Recognition of Activities of Daily Living Using Class Expression Learning Techniques

Abstract

The miniaturization and price reduction of sensors have encouraged the proliferation of smart environments, in which multitudinous sensors detect and describe the activities carried out by inhabitants. In this context, the recognition of activities of daily living has represented one of the most developed research areas in recent years. Its objective is to determine what daily activity is developed by the inhabitants of a smart environment. In this field, many proposals have been presented in the literature, many of them being based on ad hoc ontologies to formalize logical rules, which hinders their reuse in other contexts. In this work, we propose the use of class expression learning (CEL), an ontology-based data mining technique, for the recognition of ADL. This technique is based on combining the entities in the ontology, trying to find the expressions that best describe those activities. As far as we know, it is the first time that this technique is applied to this problem. To evaluate the performance of CEL for the automatic recognition of activities, we have first developed a framework that is able to convert many of the available datasets to all the ontology models we have found in the literature for dealing with ADL. Two different CEL algorithms have been employed for the recognition of eighteen activities in two different datasets. Although all the available ontologies in the literature are focused on the description of the context of the activities, the results show that the sequence of the events produced by the sensors is more relevant for their automatic recognition, in general terms.

Human Activity Recognition Models in Ontology Networks

“…With this experimental setup, we stress again that we do not want to compare our activity recognition models with other techniques validated in the literature, e.g., upper ontologies for ADL [25] or MLN [28], but rather we want to emphasise the unique traits of the architectural aspects of Arianna + . At the same time, we do not aim to investigate the recognition performance for specific activities, e.g., meal preparation.…”

“…For instance, in [20]- [22] activities are described with a combination of probabilistic models and semantic constraints. Other contextualisation techniques exploit a hybrid solution via the augmentation of ontologies with specific ML-based algorithms, which reason in terms of activity patterns [23], [24], or infer concepts through a learning approach [25]. In [26], an ontology is used to identify features useful for training human activity models, whereas for the same purpose a probabilistic ontology is used in [20].…”

“…Indeed, finding the conceptualisation that best describes the context to support the recognition of human activities is far from trivial. Several ontology-based representations have been evaluated in [25] on the basis of the context they define, e.g., sensor events, sensor hierarchies, human postures or locations. In [22], the context is defined in terms of affordances, and it is based on relevant knowledge about space, time and human attitudes.…”

A Novel Action Recognition Framework Based on Deep-Learning and Genetic Algorithms

“…This framework provides an infrastructure in which knowledge, organized in conceptual spaces, can be semantically discovered, shared, and queried across some applications. Salguero et al [23] also proposed an ontology-based data mining framework for the recognition of the daily activities of humans. This framework is based on combining the entities in the ontology and tries to find the expressions which describe daily living activities.…”

SemImput: Bridging Semantic Imputation with Deep Learning for Complex Human Activity Recognition

“…Due to the involvement of several sensors, data transmission problems among wireless sensors lead to segments of data being missed represented by Null. For this reason, we analyzed the data and performed the required imputation in order to complement the missing segments of data [37,39]. Lastly, we tested SemImput framework on the UCI-ADL dataset, which was collected while monitoring 10 different ADLs [40] using passive infrared, reed switches, and float sensors.…”