Julia Seiter

RoomSense: an indoor positioning system for smartphones using active sound probing

We present RoomSense, a new method for indoor positioning using smartphones on two resolution levels: rooms and within-rooms positions. Our technique is based on active sound fingerprinting and needs no infrastructure. Rooms and within-rooms positions are characterized by impulse response measurements. Using acoustic features of the impulse response and pattern classification, an estimation of the position is performed. An evaluation study was conducted to analyse the localization performance of RoomSense. Impulse responses of 67 within-rooms positions from 20 rooms were recorded with the hardware of a smartphone. In total 5360 impulse response measurements were collected. Our evaluation study showed that RoomSense achieves a room-level accuracy of > 98% and a within-rooms positions accuracy of > 96%. Additionally, the implementation of RoomSense as an Android App is presented in detail. The RoomSense App enables to identify an indoor location within one second.

Discovery of activity composites using topic models

In this work we investigate unsupervised activity discovery approaches using three topic modelź(TM) approaches, based on Latent Dirichlet Allocationź(LDA), n -gram TMź(NTM), and correlated TMź(CTM). While LDA structures activity primitives, NTM adds primitive sequence information, and CTM exploits co-occurring topics. We use an activity composite/primitive abstraction and analyze three public datasets with different properties that affect the discovery, including primitive rate, activity composite specificity, primitive sequence similarity, and composite-instance ratio. We compare the activity composite discovery performance among the TM approaches and against a baseline using k -means clustering. We provide guidelines for method and optimal TM parameter selection, depending on data properties and activity primitive noise. Results indicate that TMs can outperform k -means clustering up to 17%, when composite specificity is low. LDA-based TMs showed higher robustness against noise compared to other TMs and k -means.

Monitor and understand pilgrims: data collection using smartphones and wearable devices

Each year, millions of people visit the sacred sites in Makkah and Madinah. Even though the Hajj pilgrimage is one of the biggest annual events in the world, with many of the pilgrims reporting it as a life-changing experience, quite a little is done to objectively monitor the pilgrims and to understand from the crowd and from the individual point of view what makes this event so special. We present a data collection phase of 8 days of pilgrimage in April 2013 with 41 pilgrims carrying Android smartphones and 10 pilgrims wearing two physiological sensors, namely chest belts and wrist-worn devices. We describe the data recording itself, and emphasize the problems raised and the challenges faced during the study. We provide the best practices for performing solid and efficient user studies in such a difficult environment, and give first insights towards measuring important aspects of the Hajj pilgrimage such as recognition of activities and stages, analysis of group behavior, detection of stressful situations and health monitoring of pilgrims in general.

Can Smartphones Help with Running Technique

Abstract Running is one of the most popular sports for the masses. However, not every runner might run properly. Incorrect running technique decreases movement efficiency and increases the risk of injury. In this work, we present the development of a smartphone application to provide feedback on running technique on the example of arm carriage. Recognition algorithms were developed in a preliminary study with 10 participants. Investigating sensor positions and modalities, we found that a single IMU on the upper arm yielded an accuracy of 80.73% for the assessment of arm movement. We implemented our approach as a smartphone application and found that runners improved their arm movement using our application within a user study including 23 participants. Results from questionnaires revealed high user acceptance (average rating of 8 from 10 possible points).

Daily life activity routine discovery in hemiparetic rehabilitation patients using topic models.

BACKGROUND Monitoring natural behavior and activity routines of hemiparetic rehabilitation patients across the day can provide valuable progress information for therapists and patients and contribute to an optimized rehabilitation process. In particular, continuous patient monitoring could add type, frequency and duration of daily life activity routines and hence complement standard clinical scores that are assessed for particular tasks only. Machine learning methods have been applied to infer activity routines from sensor data. However, supervised methods require activity annotations to build recognition models and thus require extensive patient supervision. Discovery methods, including topic models could provide patient routine information and deal with variability in activity and movement performance across patients. Topic models have been used to discover characteristic activity routine patterns of healthy individuals using activity primitives recognized from supervised sensor data. Yet, the applicability of topic models for hemiparetic rehabilitation patients and techniques to derive activity primitives without supervision needs to be addressed. OBJECTIVES We investigate, 1) whether a topic model-based activity routine discovery framework can infer activity routines of rehabilitation patients from wearable motion sensor data. 2) We compare the performance of our topic model-based activity routine discovery using rule-based and clustering-based activity vocabulary. METHODS We analyze the activity routine discovery in a dataset recorded with 11 hemiparetic rehabilitation patients during up to ten full recording days per individual in an ambulatory daycare rehabilitation center using wearable motion sensors attached to both wrists and the non-affected thigh. We introduce and compare rule-based and clustering-based activity vocabulary to process statistical and frequency acceleration features to activity words. Activity words were used for activity routine pattern discovery using topic models based on Latent Dirichlet Allocation. Discovered activity routine patterns were then mapped to six categorized activity routines. RESULTS Using the rule-based approach, activity routines could be discovered with an average accuracy of 76% across all patients. The rule-based approach outperformed clustering by 10% and showed less confusions for predicted activity routines. CONCLUSION Topic models are suitable to discover daily life activity routines in hemiparetic rehabilitation patients without trained classifiers and activity annotations. Activity routines show characteristic patterns regarding activity primitives including body and extremity postures and movement. A patient-independent rule set can be derived. Including expert knowledge supports successful activity routine discovery over completely data-driven clustering.

Joint segmentation and activity discovery using semantic and temporal priors

We introduce a hierarchical nonparametric topic modeling approach to infer activity routines from context sensor data streams based on a distance dependent Chinese restaurant process (ddCRP). Our approach does not require labeled data at any stage. Neither does our approach depend on time-invariant sliding windows to sample context word statistics. Our activity discovery approach builds on the idea that context words occurring within one activity are semantically similar, whereas context words of different activities are less similar. Context word streams are segmented into supersamples and then semantic and temporal features are obtained to construct a segmentation prior that relates supersamples via its context words. Our hierarchical model uses the segmentation prior and ddCRP to group supersamples and the Chinese restaurant process (CRP) to discover activities. We evaluate our approach using the Opportunity dataset that contains activities of daily living. Besides being nonparametric, our ddCRP based model outperforms both, classic parametric latent Dirichlet allocation (LDA) and the nonparametric Chinese restaurant franchise (CRF). We conclude that ddCRP+CRP is an adequate approach for fully unsupervised activity discovery from context sensor data.

Activity monitoring in daily life as an outcome measure for surgical pain relief intervention using smartphones

We investigate the potential of a smartphone to measure a patients change in physical activity before and after a surgical pain relief intervention. We show feasibility for our smartphone system providing physical activity from acceleration, barometer and location data to measure the interventions outcome. In a single-case study, we monitored a pain patient carrying the smartphone before and after a surgical intervention over 26 days. Results indicate significant changes before and after intervention, particularly in physical activity in the home environment.

Informative Yet Unrevealing: Semantic Obfuscation for Location Based Services

The preservation of geo-privacy is a critical consideration for location-based service (LBS) providers. Unfortunately, a trade-off typically exists between the quality of location-based services and revealing of private information (e.g. geo-coordinates) to obtain such services. In this work, we develop semantic obfuscation methods, which allow a trusted third-party to convert revealing geo-coordinates into highly anonymous semantic features. Following, LBS providers can operate directly via location semantics to deliver the necessary services. Using a large-scale travel survey dataset, we evaluate our obfuscation approach while considering a common user-intention prediction problem. Our results demonstrate that our approach is capable of significantly obfuscating user location while maintaining LBS quality. On average, we show that the k-anonymity measure increases by 15.22 times while the quality of prediction drops only 3.24%.

Learning functional compositions of urban spaces with crowd-augmented travel survey data

Regions in urban environments often afford a mixture of different utilities. Their identification allows urban planners to leverage important insights on the emerging functional dynamics of cities. With the increasing availability of human mobility data and other forms of online digital breadcrumbs, we can now characterize urban regions with multi-source features. In this work, we form a comprehensive view of urban regions by fusing features depicting their temporal, spatial, and demographic aspects. Aggregating 47K explicitly stated trip purposes into their respective destination regions, we obtain multi-dimensional ground-truths on the functionalities of urban spaces. Given fused features and training labels, we can perform supervised learning, via multi-output regression, to estimate the functional composition of urban spaces. With 14 functional dimensions, our approach using crowd-augmented travel survey predictors delivers a mean absolute error of 3.9, approximately half of the error resulting from a mean-based straw man approach (mean absolute error of 7.9). Clustering estimated regional functionalities, we find highly coherent cluster assignments (adjusted Rand Index of 0.81) compared to clustering directly on regional functionality labels. Finally, we provide an illustrative case-study where clustering of estimated region functionalities can be used to intuitively differentiate prototypical spatial neighbourhoods of a large metropolitan.

Sensor technology for ice hockey and skating

Sensor technology that is unobtrusively integrated into the clothing and equipment of an athlete can support the training of sport activities and monitor the athletes progress. In this paper, we propose two wearable systems that support ice hockey players in the training of skating and shooting. These assistants measure the motions of players and compare them with reference executions of the same activities by professional players. A third system that we introduce monitors the player;s activities during a hockey game and creates a match report for objective performance measurement. For each of the three proposed applications, we present a prototype setup that we evaluate with amateur and professional players. The main findings are i) that with a skate-worn motion sensor and user-dependent training, eight skating motions can be spotted with an accuracy above 90%, ii) that stick-integrated sensors enable the measurement of relevant shot features, which differentiate professional from amateur athletes, and iii) that it is possible to spot important ice hockey activities in the signals of body-worn motion sensors worn during a game.