Martin Berchtold

ActiServ: Activity Recognition Service for mobile phones

Smart phones have become a powerful platform for wearable context recognition. We present a service-based recognition architecture which creates an evolving classification system using feedback from the user community. The approach utilizes classifiers based on fuzzy inference systems which use live annotation to personalize the classifier instance on the device. Our recognition system is designed for everyday use: it allows flexible placement of the device (no assumed or fixed position), requires only minimal personalization effort from the user (1–3 minutes per activity) and is capable of detecting a high number of activities. The components of the service are shown in an evaluation scenario, in which recognition rates up to 97% can be achieved for ten activity classes.

An extensible modular recognition concept that makes activity recognition practical

In mobile and ubiquitous computing, there is a strong need for supporting different users with different interests, needs, and demands. Activity recognition systems for context aware computing applications usually employ highly optimized off-line learning methods. In such systems, a new classifier can only be added if the whole recognition system is redesigned. For many applications that is not a practical approach. To be open for new users and applications, we propose an extensible recognition system with a modular structure. We will show that such an approach can produce almost the same accuracy compared to a system that has been generally trained (only 2 percentage points lower). Our modular classifier system allows the addition of new classifier modules. These modules use Recurrent Fuzzy Inference Systems (RFIS) as mapping functions, that not only deliver a classification, but also an uncertainty value describing the reliability of the classification. Based on the uncertainty value we are able to boost recognition rates. A genetic algorithm search enables the modular combination.

Recognizing Group Activities Using Wearable Sensors

Pervasive computing envisions implicit interaction between people and their intelligent environments instead of between individuals and their devices, inevitably leading to groups of individuals interacting with the same intelligent environment. These environments must be aware of user contexts and activities, as well as the contexts and activities of groups of users. Here an application for in-network group activity recognition using only mobile devices and their sensors is presented. Different data abstraction levels for recognition were investigated in terms of recognition rates, power consumption and wireless communication volumes for the devices involved. The results indicate that using locally extracted features for global, multi-user activity recognition is advantageous (10% reduction in energy consumption, theoretically no loss in recognition rates). Using locally classified single-user activities incurred a 47% loss in recognition capabilities, making it unattractive. Local clustering of sensor data indicates potential for group activity recognition with room for improvement (40% reduction in energy consumed, though 20% loss of recognition abilities).

Increased Robustness in Context Detection and Reasoning Using Uncertainty Measures: Concept and Application

This paper reports on a novel recurrent fuzzy classification method for robust detection of context activities in an environment using either single or distributed sensors. It also introduces a classification of system architectures for uncertainty calculation in general. Our proposed novel method utilizes uncertainty measures for improvement of detection, fusion and aggregation of context knowledge. Uncertainty measurement calculations are based on our novel recurrent fuzzy system. We applied the method in a real application to recognize various applause (and non applause) situations, e.g. during a conference. Measurements were taken from mobile phone sensors (microphone, accel. if available) and acceleration sensory attached to a board marker. We show that we are able to improve robustness of detection using our novel recurrent fuzzy classifier in combination with uncertainty measures by ~30% on average. We also show that the use of multiple phones and distributed recognition in most cases allows to achieve a recognition rate between 90% and 100%.

Using a Context Quality Measure for Improving Smart Appliances

Many ubicomp appliances require the recognition of context. Existing context systems do not provide information about the quality of the context recognized to the appliance at runtime. In this paper we propose the first context quality system which gives quantitative measures, the context quality measure (CQM), in real time. The CQM can be used by application to improve decision quality when interpreting context values. Our Fuzzy Inference System based approach considers the context detection algorithm as a black-box. It is therefore able to give generalized independet context quality measures and is applicable as an add-on to any context recognition system. A first practical implementation shows a gain of 33% in context detection quality in tested application scenarios.

AwarePen - Classification Probability and Fuzziness in a Context Aware Application

Fuzzy inference has been proven a candidate technology for context recognition systems. In comparison to probability theory, its advantage is its more natural mapping of phenomena of the real world as context. This paper reports on our experience with building and using monolithic fuzzy-based systems (a TSK-FIS) to recognize real-world events and to classify these events into several categories. It will also report on some drawbacks of this approach that we have found. To overcome these drawbacks a novel concept is proposed in this paper. The concept incorporates fuzzy-based approaches with probabilistic methods, and separates the monolithic fuzzy-based system into several modules. The core advantage of the concept lays in the separation of detection complexity into distinct modules, each of them using fuzzy-based inference for context classification. Separation of detection functionality is supported by an automatic process using transition probabilities between context classifiers to optimize detection quality for the resulting detection system. This way our approach incorporates the advantages of fuzzy-based and probabilistic systems. This paper will show results of experiments of an existing system using a monolithic FIS approach, and reports on advantages when using a modular approach.

Matrix Routing -- An Interference Range Insensitive Routing Protocol for Wireless Sensor Networks

Interference ranges can dramatically affect the throughput in wireless sensor networks. While the transmission range defines the maximum physical range of a radio signal the interference range determines the area in which other nodes will be prevented from successful receiving or transmitting signals. In this paper we present an initial self organizing routing protocol for wireless sensor networks, named Matrix Routing, which is maximally insensitive even to high interference disturbances. Matrix routing is predictable, proactive but not table driven, needs minimum hardware and computational power and does not require transmission of routing packets. The protocol is characterized by zero overhearing costs and minimal idle listening. The paper shows a proof of concept, evaluates potential of our algorithm and discusses strength, limitations and application areas.

An experiment in hierarchical recognition of group activities using wearable sensors

Pervasive computing envisions implicit interaction between people and their intelligent environments instead of individual devices, inevitably leading to groups of individuals interacting with the same intelligent environment. These environments must therefore be aware not only of user contexts and activities, but the contexts and activities of groups of users as well. This poster will demonstrate an experiment conducted towards understanding hierarchical multi-user group activity recognition using wearable sensors. The experiment will explore different data abstraction levels in terms of recognition rates, power consumption and wireless communication volumes for the devices involved.

Gath-Geva specification and genetic generalization of Takagi-Sugeno-Kang fuzzy models

This paper introduces a fuzzy inference system, based on the Takagi-Sugeno-Kang model, to achieve efficient and reliable classification in the domain of ubiquitous computing, and in particular for smart or context-aware, sensor-augmented devices. As these are typically deployed in unpredictable environments and have a large amount of correlated sensor data, we propose to use a Gath-Geva clustering specification as well as a genetic algorithm approach to improve the models robustness. Experiments on data from such a sensor-augmented device show that accuracy is boosted from 83% to 97% with these optimizations under normal conditions, and for more. challenging data from 54% to 79%.

Pluggable real world interfaces Physically enabled code deployment for networked sensors

In this paper we present a novel abstraction and deployment process using real world interfaces, which reflect the realities of pervasive software development. Pluggable real world interfaces support ldquoplugpsilanpsilaplayrdquo deployment for sensor-augmented hardware and provide an object-oriented encapsulation of high-level contextual interfaces. The architecture adds an additional object based abstraction layer between the sensor subsystem (delivering e.g. cues) and the application (delivering the situation context). Component abstraction layers are implemented as code that comes with physical components, e.g. a chair, and provides the functionality for detecting context bundled with the sensory hardware. The approach will lead to pluggable real world interfaces: The functionality of an appliance will be composed from the functionality of its components - just like a meeting room will be composed from many chairs. This paper will present concept, architecture and a first implementation based on a Java run-time system for very tiny, very low-power embedded sensor nodes.