Progress in ambient assisted systems for independent living by the elderly
Sensor fusion for activity recognition
Sayuti et al. (2014) discussed the trade-offs between measurements delay and throughput in a case study utilizing the lightweight priority scheduling scheme for activity monitoring from a distributed sensor system. The findings showed that the proposed scheme presented promising solution that supports decision making for Ambient Assistive Living (AAL) system in a real setting. The creation of an AAL environment not only embeds sensors to acquire information, it processes this information and interacts with the subject for enhanced quality of life.
There are several types of monitors which can be used to gather data of physical activities. In medical practice, it is common to continuously monitor patients’ biological status such as heart rates and saturation level by using wearable monitors. However, as some researchers have suggested, it is not practical to apply these wearable monitors in the community setting in order to evaluate their daily activities. Many remote monitoring projects have been developed in recent years in laboratory and community settings using non-wearable environment monitors. Xin and Herzog (2012) in their work presented a wearable monitoring system designed to achieve continuous in-house and outdoor health monitoring to support elderly people’s independence. The system acts as a health diagnosis assistant through its on-board intelligence to generate real-time reliable health condition diagnosis. The on-board decision support system continuously learns the subject’s health characteristics at certain time intervals from the attached sensor system. Hence, a dynamic decision model is continuously adapted to the subject’s health profile. The system is also able to measure deviations from the normal state and categorize whether it is a definite critical situation or a just a normal uncritical deviation.
Pirttikangas et al. (2006) studied activity identification using wearable, small-sized sensor devices. They attached these small devices to four different locations on the subject’s body. In their experiment, they collected data from 13 different subjects of both sexes performing 17 daily life activities. They extracted features from heart rate and tri-axial accelerometer sensors for different sampling times. They employed the forward-backward sequential search algorithm for important feature selection from these features.
De Miguel-Bilbao et al. (2013) illustrated a non-invasive sensor system that consists of action sensors and presence sensors for monitoring daily life activities, as well as the configuration of the monitored homes and users. The post processing stage for activity monitoring is independent from the home topology monitoring process. The system extracted parameters can be considered as long-term monitoring data aiming at detecting and validating daily activities, enabling early detection of physical and cognitive dysfunctions (Rashidi et al. 2011). The monitoring of household activity method can help to improve global geriatric evaluation and enhance the possibility for a better remote monitoring system of elderly people in their homes. This knowledge can support design and manufacture of biomedical sensors that are small, reliable, sensitive, and inexpensive (Agoulmine et al. 2011).
Shuai et al. (2010) focused on including activity duration into the learning of inhabitants’ daily living activities and behaviour patterns in a smart home environment. They applied a probabilistic learning algorithm to study multi-inhabitants in the same smart home environment. They predicted both inhabitants and their ADL model utilizing the activity carried out and the people who are performing it through experiments performed in a smart kitchen laboratory. The experimental results for activity identification demonstrated high accuracy compared to unreliable results that are obtained with no activity duration information in the model. Their approach also provides a great opportunity for identifying drifts in long-term activity monitoring as an early stage detection of deteriorating situation.
Language-based programming and interaction approach provides support for developers to freely express the global behaviour of a smart home application as one logical entity. The high-level language eases the implementation efforts for the application developer. By structuring the application development into different high-level models, developers can simplify application maintenance and customization due to changing user requirements or changes in the monitored living environment. In this way, people are directed to use rules for describing the required behaviour within a smart home environment. Consequently, by providing a rule-based modelling language the gap between the user-based application development and the actual system implementation can be reduced (Bischoff et al. 2007).
Algase et al. (2003) investigated reliable measures that are suitable enough to identify the wandering behaviour. Most of the studies they researched for wandering behaviour were relying on simple classification of subject’s state as wandering or not-wandering based upon personal caregiver judgments, which doesn’t have clear consistent assessment. They found that unplanned ambulation is a key element across all methods used for wandering behaviour identification. They studied different types of sensors used to wandering behaviour identification. They found that the StepWatch device outperform all other devices as it is always able to identify wandering behaviour correctly. The StepWatch always produced the best estimate for the subject’s wandering time spent, whereas other tested devices in the study were oversensitive to normal movement and produced substantial overestimates.
Wireless body attached sensor devices and smart phones were utilized to monitor the health condition of elderly people in a recent study by Bose (2013). These body attached devices offered remote sensing for the elderly vital signs for health condition assessment anytime and anywhere. Moreover, it supported creating customized solution for each subject according to his individual health condition requirements. If the system detects an emergency situation or deteriorating conditions, the smart phone will alert pre-assigned supervisors or the elderly person’s family or neighbours through text messages or making a phone call with predefined condition description voice message. In some cases, it even alerts the ambulance service with detailed report for the subject condition and location. Moreover, the system features some unique functions to support the elderly person’s daily life basic requirements, such as regular medication reminder, medical guidance, etc. However, he highlighted in his work that there is still need for innovations required in the Wireless Sensor Networks (WSN) field to enable such technologies to reach reliable and confidence application in this domain.
The work illustrated by Arai (2014), concerned with vital signs monitoring such as blood pressure, body temperature, pulse rate, bless, location/attitude, and consciousness using wearable distributed sensor network for the purpose of rescue of elderly people who will be in vital need for support in evacuation condition from a disaster location. Experimental results show that all of the vital signs as well as location and attitude identification of the elderly persons were correctly monitored with the proposed sensor networks. Moreover, it was clear that there is no specific correlation between pulse rate and the subject age, there is no specific calorie consumption that can be linked to age, EEG signal can be linked to eyes movement to predict psychological state, and there is clear difference between healthy person and patient with dementia disease. Finally, they found that there are links between blood pressure and physical/psychological stress (Arai 2014).
Phua et al. (2009) illustrated studied memory and problem-solving abilities to produce what then they called Erroneous-Plan Recognition (EPR), aiming to identify imperfections or faults in specific plans implementation by memory problems’ patients. Several challenges faced the researchers that are related to the correct definition of a plan within daily living activities, the choice of the activities to be monitored, the type of sensors required to recognize these activities, and the activities recognition technique to be used. In this study, they used independent sequential error detection layers to identify specific errors in the plan implementation. Their obtained result indicated that error data can be separated effectively. This study gave examples of how the suggested EPR system can work well with Deterministic Finite-State Automata (DFSA) technique for identifying error probabilities.
Lauriks et al. (2007) provided detailed analysis for the state of the art in information and communication technologies (ICT) that can be applied in solving unmet needs by elderly people. They categorized these needs as tailored information system requirement, customized disease support requirement, social interaction requirement, health condition monitoring requirement, and observed safety requirement. ICT solutions targeting memory problems demonstrate that people with memory diseases are able to use simple electronic equipment with enough confidence. Instrumental ICT-based systems targeting social activities could be simply implemented via the use of mobile phones or entertaining robotic platforms. GPS-based tracking devices proved their ability to enhance feeling of safety. However, more studies regarding these ICT solutions in simulated daily life situations are required before going to commercialized implementation for elderly people daily life support. The final step after sensor data fusion is the activity recognition algorithms used to characterize the activities performed by the elderly.