As mobile phones advance in functionality and capability, they are being used for more than just communication. Increasingly, these devices are being employed as instruments for introspection into habits and situations of individuals and communities. Many of the applications enabled by this new use of mobile phones rely on contextual information. The focus of this work is on one dimension of context, the transportation mode of an individual when outside. We create a convenient (no specific position and orientation setting) classification system that uses a mobile phone with a built-in GPS receiver and an accelerometer. The transportation modes identified include whether an individual is stationary, walking, running, biking, or in motorized transport. The overall classification system consists of a decision tree followed by a first-order discrete Hidden Markov Model and achieves an accuracy level of 93.6p when tested on a dataset obtained from sixteen individuals.

/pdf/using-mobile-phones-to-determine-transportation-modes-2pw1qd0lx7.pdf

Inferring high-level behavior from

Using mobile phones to determine transportation modes

Accurate recognition and tracking of human activities is an important goal of ubiquitous computing. Recent advances in the development of multi-modal wearable sensors enable us to gather rich datasets of human activities. However, the problem of automatically identifying the most useful features for modeling such activities remains largely unsolved. In this paper we present a hybrid approach to recognizing activities, which combines boosting to discriminatively select useful features and learn an ensemble of static classifiers to recognize different activities, with hidden Markov models (HMMs) to capture the temporal regularities and smoothness of activities. We tested the activity recognition system using over 12 hours of wearable-sensor data collected by volunteers in natural unconstrained environments. The models succeeded in identifying a small set of maximally informative features, and were able identify ten different human activities with an accuracy of 95%.

/pdf/a-hybrid-discriminative-generative-approach-for-modeling-akp5bg1joc.pdf

A hybrid discriminative/generative approach for modeling human activities

In this paper, we introduce a compact system for fusing location data with data from simple, low-cost, non-location sensors to infer a user's place and situational context. Specifically, the system senses location with a GSM cell phone and a WiFi-enabled mobile device (each running Place Lab), and collects additional sensor data using a 2” x 1” sensor board that contains a set of common sensors (e.g. accelerometers, barometric pressure sensors) and is attached to the mobile device. Our chief contribution is a multi-sensor system design that provides indoor-outdoor location information, and which models the capabilities and form factor of future cell phones. With two basic examples, we demonstrate that even using fairly primitive sensor processing and fusion algorithms we can leverage the synergy between our location and non-location sensors to unlock new possibilities for mobile context inference. We conclude by discussing directions for future work.

Mobile context inference using low-cost sensors

Predicting a driver's route would be useful for warning a driver of upcoming road hazards, informing about traffic situations, and serving relevant advertising. There are many clues to a driver's future route, including their past behavior and likely destination. Another clue is the driver's turn choices at a sequence of intersections. Strung together, the turn choices form a route. This paper develops a basic algorithm, and variations, to predict the aggregate turn behavior of drivers at intersections. Given an intersection with a few different turn options, including the option to continue straight ahead, our goal is to infer the proportion of drivers who will take each option. For ground truth, we use raw turn counts gathered for government traffic studies by our local municipality at 40 different intersections. Our basic turn prediction algorithm is based on the assumption that drivers tend to choose roads that offer them more destination options. This matches our intuition that turning onto a short, dead-end road is relatively rare compared to turning onto a highway "on" ramp. The best performing algorithm predicts turn proportions with a median error of 0.142.

background or methods

/pdf/where-will-they-turn-predicting-turn-proportions-at-kw8gvach3c.pdf

Where will they turn: predicting turn proportions at intersections

Many countries are facing the problem of caring economically for their ageing population. One approach to this is the development of environments which possess ambient intelligence that will provide care for older people while assisting them with their everyday life activities. One function that is required in care provision is spotting abnormal behaviours as this might be an indicator of a problem requiring attention from a carer. Agent technology can be employed to detect abnormalities by first learning a normal set of personal behaviours and then detecting deviations from these. This paper presents a novel connectionist embedded agent architecture that combines the use of unobtrusive and relatively simple sensors and employs a constructive algorithm with temporal capabilities which is able to recognize different high level activities (such as “sleeping”, “working at computer”, “eating”), and identify abnormal behaviours. The network is trained in an online mode, and is able to adapt and expand itself as new data are made available over time or it can add new output nodes to represent new classes or accommodate the abnormal instances. The developed connectionist approach is not computationally demanding and hence it can be integrated into the limited processor-power embedded computing platforms used in intelligent domestic environments.

Inferring high-level behavior from

Citations

Using mobile phones to determine transportation modes

A hybrid discriminative/generative approach for modeling human activities

Mobile context inference using low-cost sensors

Where will they turn: predicting turn proportions at intersections

A neural network agent based approach to activity detection in AmI environments

Related Papers (1)

Inferring High-Level Behavior from Low-Level Sensors