Context Aware and Autonomic Sensing

From Capsil Wiki

From a data analysis point of view, context can be regarded as different levels of detail linked to physical and perceptual representations. A user's cognitive activities are normally described at an abstract level, whereas the recognition of the physical status of a person is usually more descriptive and data-driven. Context aware sensing generally aims at answering some of the following questions:

• Who: the identity of people in an environment
• What: the activities and interactions in the current environment
• Where: environment and location details
• When: timestamps and temporal relationships between events
• Why: States of a person, relationships between events and environmental factors.

From a sensing perspective, most of the context aware applications are based on motion sensors, physiological sensors (such as heart rate and respiratory sensors), visual sensors (cameras, using off-line and real-time processing) and ambient sensors that provide environmental information.

The following table summarises some recent applications, more ideas on context aware sensing can be found in Thiemjarus et al.:

Link	Sensors used	Context awareness techniques	Purpose of study
Bao and Intille, 2006	Bi-axial accelerometer on hip, waist and ankle	Decision tree classifiers/Naive Bayes	Activity Recognition from user annotated data
Ermes et al. 2008	Accelerometers on wrist and waist, ECG sensor, respiration sensor, magnetometers	Decision tree and an MLP (Multilayer perceptron)	Detection of daily activities and sports indoors and outdoors
Krause et al. 2006	SenSay Prototype including a SenseWear arm-band, a headset and a back-pack	KSOM (Kohonen's Self Organising Map)	Identify context states without supervision based on wearable sensors for a context aware mobile phone.
Lo et al. 2007	An ear worn activity recognition sensor	Naive Bayes classifier based on Gaussian class models	Activity recognition in a pervasive home environment
Loosli et al. 2005	EMG, Blood volume pressure, skin conductivity and respiration sensors	One class SVM and rupture detection	Context change detection
Maurer et al. 2006	multisensor platform (eWatch) worn on different positions	LDA then decision tree classifiers	Analysing the tradeoff between recognition accuracy and computing complexity
Subramanya et al. 2006	Combining GPS measurements with accelerometers, microphones, light and temperature sensors	Dynamic bayesian Network for an activity model showing locations and environment	Estimating both activity and spatial context over time

From a methodology point of view, the following list summarises some of the widely used techniques for context aware sensing:

• Hidden Markov Models (HMMs) and their variants, including coupled HMMs, abstract HMMs and variable-length HMMs. A detailed webpage linking to several recent papers as well as code and methods is available.
• Conditional Random Fields: CFRs model the conditional probability of the label sequence rather than the joint probability of both labels and observations (as in HMMs). As above, a detailed [webpage] with resources on CRF is available.
• Bayesian network approaches, which are graphical models that encodes probabilistic relationships among variables of interest. A website summarising methods and applications of Bayesian networks is available.
• Discriminative classifiers: aiming to discriminate between activities without actually modelling activity classes or behaviour variability. This includes several classifier types, such as Gaussian Mixture Models (GMMs), Multi-layer perceptrons (MLP), Radial-basis functions (RBFs), K-nearest neighbours (KNN) and K-means classifiers. A summary webpage with links to several toolboxes and applications is available.

Autonomic Sensing

The eight defining characteristics of an autonomic system are defined (the IBM website) as:

1. Self-management: An autonomic computing system needs to "know itself" - its components must also possess a system identity. Since a "system" can exist at many levels, an autonomic system will need detailed knowledge of its components, current status, ultimate capacity, and all connections to other systems to govern itself. It will need to know the extent of its "owned" resources, those it can borrow or lend, and those that can be shared or should be isolated.
2. Self-configuration: An autonomic computing system must configure and reconfigure itself under varying (and in the future, even unpredictable) conditions. System configuration or "setup" must occur automatically, as well as dynamic adjustments to that configuration to best handle changing environments.
3. Self-optimisation: An autonomic computing system never settles for the status quo - it always looks for ways to optimize its workings. It will monitor its constituent parts and fine-tune workflow to achieve predetermined system goals.
4. Self-healing: An autonomic computing system must perform something akin to healing - it must be able to recover from routine and extraordinary events that might cause some of its parts to malfunction. It must be able to discover problems or potential problems, then find an alternate way of using resources or reconfiguring the system to keep functioning smoothly.
5. Self-protection: A virtual world is no less dangerous than the physical one, so an autonomic computing system must be an expert in self-protection. It must detect, identify and protect itself against various types of attacks to maintain overall system security and integrity.
6. Self-adaptation: An autonomic computing system must know its environment and the context surrounding its activity, and act accordingly. It will find and generate rules for how best to interact with neighboring systems. It will tap available resources, even negotiate the use by other systems of its underutilized elements, changing both itself and its environment in the process -- in a word, adapting.
7. Self-integration: An autonomic computing system cannot exist in a hermetic environment. While independent in its ability to manage itself, it must function in a heterogeneous world and implement open standards -- in other words, an autonomic computing system cannot, by definition, be a proprietary solution.
8. Self-scaling: An autonomic computing system will anticipate the optimized resources needed while keeping its complexity hidden. It must marshal I/T resources to shrink the gap between the business or personal goals of the user, and the I/T implementation necessary to achieve those goals -- without involving the user in that implementation.

These areas are essential research issues for Body Sensor Networks and merit a detailed discussion. However, from a data-processing point of view, collaborative information processing (leading to self-optimisation) is of relevance to data-processing in autonomic sensing and will be explained further in the following section.

Collaborative Information Processing

In addition to considerations of single-platform signal processing, the networked information processing is further constrained by application requirements on energy efficiency, network latency, and fault tolerance. Wireless Sensor Networks (WSNs) are severely constrained in computation and communication capabilities due to the cost and size of available sensors. On the other hand, autonomic computing (AC) offers a promising solution to manage large-scale computing systems without human intervention. Thus, the last decade has witnessed a rising interest in the use of Autonomic computing techniques for sensor networks. A preliminary power aware self-configuring and self optimising sensor is given in Kang et al. The simulation results confirm that the proposed power-aware scheme prolongs the network lifetime and balances the energy insensor nodes. A special issue of IEEE Signal Processing Magazine focussed on collaborative signal and information processing in micro-sensor networks was published in 2002.

Zhao et al. use collaborative information processing to provide a solution for a sensor network tracking an object. In particular, tracking was used as a canonical problem to expose important constraints in designing, scaling, and deploying sensor networks. Results from simulations and experimental implementations demonstrate information based approaches can be scalable and make efficient use of scarce sensing and communication resources.

Beam-forming is one of the simple, yet important applications of collaborative information processing. In beam-forming, the signals obtained at different receivers are used to cancel out noise and re-obtain the original signal sent by the source. In addition to the re-construction of the source signal and the localisation of the source, beam-forming can help in source separation as different sources can be separated based on their locations.

IPSN (Information processing in sensor networks conference), since 2001, has presented a forum where several relevant papers were discussed, such as D'Costa et al., Liu et al. and Whitehouse et al.