Algorithms
From Capsil Wiki
With the rapid improvement in hardware and software capabilities of wearable and ambient sensors, the challenge moved from that of obtaining sensor data to that of analysing large quantities of data and providing a context aware autonomous system, leading to truly 'personalised' health care. Thus, this section will focus on the common data analysis algorithms for pervasive healthcare systems:
Contents |
On-node Data Processing
This refers to algorithms that can be implemented directly on a sensor node, allowing the optimisation of resources in terms of energy and communications. These techniques are also essential if abnormal events were to be detected on the sensor node, and alerts were to be generated at that level.
Sensor Fusion
The use of multiple sensors with information fusion has the several main advantages compared to single sensor systems. These include improved signal to noise ratios, enhanced robustness and reliability in the event of sensor failure, integration of independent features and prior knowledge, reducing uncertainty and improved resolution, precision, confidence and hypothesis discrimination.
Context Aware and Autonomic Sensing
The contextual information in Body Sensor Networks is mainly focused on the user's activity, physiological status and the surrounding physical environment. Understanding the context in which the user performs his/her activities is essential in comprehending the activities themselves and their relationship to prior and future activities, as well as environmental changes. Context aware sensing and autonomic sensing are quite linked. The latter referring to networks that can autonomically configure, optimise, manage, heal, protect, adapt, scale and integrate.
Data Mining and Trend Analysis
With large amounts of data typically obtained from Body Sensor Networks, efficient data-mining is essential to allow important patterns to be recognised, errors in the data highlighted and trends to be noted. This section will cover approaches that have been successfully used to provide pattern recognition in Body Sensor Networks.
Falls Detection Algorithms
A variety of data analysis techniques incorporating value algorithms have been applied to Falls detection domain. Many of the approaches focus on improving the selectivity of Falls. The issue of false positives is the most significant issues limiting the reliability of body worn Falls detectors. Many efforts have focused on improving the classification base solely on the sensor by using a variety of mathematical techniques such as thresholding using support data with sensor data typically accelerometer based to achieve greater selectivity. Alternatively supporting data sources have been utilized to improve selectivity such as the inclusion of sound as additional source to a classifier to improve the accuracy of the falls detection.
