Dawud Gordon

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.

Energy-Efficient Activity Recognition Using Prediction

Energy storage is quickly becoming the limiting factor in mobile pervasive technology. For intelligent wearable applications to be practical, methods for low power activity recognition must be embedded in mobile devices. We present a novel method for activity recognition which leverages the predictability of human behavior to conserve energy. The novel algorithm accomplishes this by quantifying activity-sensor dependencies, and using prediction methods to identify likely future activities. Sensors are then identified which can be temporarily turned off at little or no recognition cost. The approach is implemented and simulated using an activity recognition data set, revealing that large savings in energy are possible at very low cost (e.g. 84% energy savings for a loss of 1.2 pp in recognition).

Investigation of Context Prediction Accuracy for Different Context Abstraction Levels

Context prediction is the task of inferring information about the progression of an observed context time series based on its previous behaviour. Prediction methods can be applied at several abstraction levels in the context processing chain. In a theoretical analysis as well as by means of experiments we show that the nature of the input data, the quality of the output, and finally the flow of processing operations used to make a prediction, are correlated. A comprehensive discussion of basic concepts in context prediction domains and a study on the effects of the context abstraction level on the context prediction accuracy in context prediction scenarios is provided. We develop a set of formulae that link scenario-dependent parameters to a probability for the context prediction accuracy. It is demonstrated that the results achieved in our theoretical analysis can also be confirmed in simulations as well as in experimental studies.

A novel micro-vibration sensor for activity recognition: Potential and limitations

This paper researches the potential of a novel ball switch as a wearable vibration sensor for activity recognition. The ball switch is available as a commercial, off-the-shelf sensor and is unique among such sensors due to its miniaturized design and the low mass of the ball. We present a detailed analysis of the physical properties of the sensor as well as a recommendation for circuit design, sampling method and a feature generation algorithm for activity recognition. The analysis reveals that it is sensitive to vibrations between 1.5 kHz and 8 kHz, where the acceleration sensor is responsive below 1.6 kHz. Furthermore, the ball switch is substantially cheaper (3x), smaller (2x) and uses less power (50x) than an accelerometer based system, but delivers less information. We also present the results of a case study in activity recognition done in parallel with an acceleration sensor using 5 subjects and 8 different activities. It shows that the ball switch can increase recognition rates when added to an accelerometer-based system, demonstrating that it can sample activity-pertinent information which an accelerometer can not. We conclude that this ball switch can be used to recognize high-frequency activity components and effectively improve recognition rates while representing a very low cost sensor in terms of price, device size and power consumption.

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).

Dinam: A wireless sensor network concept and platform for rapid development

Dinam is a novel approach to simplified rapid prototyping of wireless sensor network applications as well as an according WSN platform. As opposed to the traditional mote-based development archetype, dinam proposes combining the development steps into a single continuous, fluid process that is completely integrated into the node. The dinam concept sensor node integrates all development tools, source code and other data into the sensor node system. It is claimed that this concept will greatly reduce the amount of effort required to develop wireless sensor network applications by removing the overhead of installation, iterative development steps and complexity of the development process. In order to confirm or refute this claim, a first prototype for educational purposes is developed and presented which implements the dinam approach. The development of applications is evaluated in terms of time required for a specific scenario with a user study. The results presented here indicate that an integrated instruction and development period of 10 minutes is sufficient for simple applications using the dinam approach.

Group affiliation detection using model divergence for wearable devices

Methods for recognizing group affiliations using mobile devices have been proposed using centralized instances to aggregate and evaluate data. However centralized systems do not scale well and fail when the network is congested. We present a method for distributed, peer-to-peer (P2P) recognition of group affiliations in multi-group environments, using the divergence of mobile phone sensor data distributions as an indicator of similarity. The method assesses pairwise similarity between individuals using model parameters instead of sensor observations, and then interprets that information in a distributed manner. An experiment was conducted with 10 individuals in different group configurations to compare P2P and conventional centralized approaches. Although the output of the proposed method fluctuates, we can still correctly detect 93% of group affiliations by applying a filter. We foresee applications in mobile social networking, life logging, smart environments, crowd situations and possibly crowd emergencies.

Using prediction to conserve energy in recognition on mobile devices

As devices are expected to be aware of their environment, the challenge becomes how to accommodate these abilities with the power constraints which plague modern mobile devices. We present a framework for an embedded approach to context recognition which reduces power consumption. This is accomplished by identifying class-sensor dependencies, and using prediction methods to identify likely future classes, thereby identifying sensors which can be temporarily turned off. Different methods for prediction, as well as integration with several classifiers is analyzed and the methods are evaluated in terms of computational load and loss in quality of context. The results indicate that the amount of energy which can be saved is dependent on two variables (the acceptable loss in quality of recognition, and the number of most likely classes which should be accounted for), and two scenario-dependent properties (predictability of the context sequences and size of the context-sensor dependency sets).

jNode: A sensor network platform that supports distributed inertial kinematic monitoring

Because of the intrinsic advantages of wireless inertial motion tracking, standalone devices that integrate inertial motion units with wireless networking capabilities have gained much interest in recent years. Several platforms, both commercially available and academic, have been proposed to balance the challenges of a small form-factor, power consumption, accuracy and processing speed. Applications include ambulatory monitoring to support healthcare, sport activity analysis, recognizing human group behaviour, navigation support for humans, robots and unmanned vehicles, but also in structural monitoring of large buildings. This paper provides an analysis of the current state-of-the-art platforms in wireless inertial motion tracking and presents a novel open-source and open-hardware hybrid tracking platform that is extensible, low-power, flexible enough to be used for both short- and long-term monitoring and based on a firmware that allows it to be easily adapted after being deployed.

Global peer-to-peer classification in mobile ad-hoc networks: a requirements analysis

This paper examines global context classification in peer-to-peer ad-hoc mobile wireless networks (P2P-MANETs). To begin, circumstances are presented in which such systems would be required to classify a global context. These circumstances are expounded upon by presenting concrete scenarios from which a set of requirements are derived. Using these requirements, related work is evaluated for applicability, indicating no adequate solutions. Algorithmic approaches are proposed, and analysis results in a benchmark as well as bounds for distribution of processing load, memory consumption and message passing in P2P-MANETs.