Human Activity Recognition in Smart-Home Environments for Health-Care Applications

Abstract

With a growing population of elderly people, the number of subjects at risk of cognitive disorders is rapidly increasing. Many research groups are studying pervasive solutions to continuously and unobtrusively monitor fragile subjects in their homes, reducing health-care costs and supporting the medical diagnosis. Clinicians are interested in monitoring several behavioral aspects for a wide variety of applications: early diagnosis, emergency monitoring, assessment of cognitive disorders, etcetera. Among the several behavioral aspects of interest, anomalous behaviors while performing activities of daily living (ADLs) are of great importance. Indeed, these anomalies can be indicators of serious cognitive diseases like Mild Cognitive Impairment. The recognition of such abnormal behaviors relies on robust and accurate ADLs recognition systems. Moreover, in order to enable unobtrusive and privacy-aware monitoring, environmental sensors in charge of unobtrusively capturing the interaction of the subject with the home infrastructure should be preferred. This thesis presents several contributions on this topic. The major ones are two novel hybrid ADLs recognition algorithms. The former is supervised while the latter is unsupervised. Preliminary results, which still need to be confirmed, show that the recognition rate of the unsupervised method is comparable to the one obtained by the supervised one, with the great advantage of not requiring the acquisition of an annotated dataset. Beyond ADLs recognition, other contributions on smart sensing and anomaly recognition are presented. Regarding unobtrusive sensing, we propose a machine learning technique to detect fine-grained manipulations performed by the inhabitant on household objects instrumented with tiny accelerometer sensors. Finally, a novel rule-based framework for the recognition of fine-grained abnormal behaviors is presented. Experimental results on several datasets show the effectiveness of all the proposed techniques.ii

Comparative Analysis of Two Methods in Fine-Grained Activity Recognition for Ambient Assisted Living

“…The activity recognition process in smart homes involves multiple steps Civitarese (2019), typically broken down into data collection, data preprocessing, feature extraction, and activity classification, as we can see in Figure 3. Initially, raw data is gathered from various sensors placed in the environment or worn by the user.…”

“…The key challenge in developing these technologies is related to the precision of activity recognition Civitarese (2019). Most current research Mehr and Polat (2019) focuses on monitoring high-level activities, such as eating or sleeping.…”

Human Activity Recognition Based on an Efficient Neural Architecture Search Framework Using Evolutionary Multi-Objective Surrogate-Assisted Algorithms

“…HAR is an essential task in computer vision, which aims at analyzing the long-term behavior pattern and recognizing the specific type of behavior from the input data [1]. It has been widely used in virtual reality [2], automatic driving [3], and smart home environment monitoring [4]. Especially with the significant rise in the population over 65, the personal expenditure on medical service and long-term care are increasing sharply due to the awakening of health consciousness [5], HAR also plays a vital role in helping the elderly living alone, and the youth who need medical care [6].…”

Super-Resolution Image Reconstruction Based on Self-Calibrated Convolutional GAN

“…Single image super-resolution reconstruction (SISRR) is an important research direction in the field of computer vision. Super-resolution reconstruction aims to reconstruct the corresponding high-resolution image from the observed low-resolution image to obtain richer image detail information, which has been utilized in medical imaging [1], [2], monitoring equipment [3] and other fields [4].…”