Probability Distributions.- Linear Models for Regression.- Linear Models for Classification.- Neural Networks.- Kernel Methods.- Sparse Kernel Machines.- Graphical Models.- Mixture Models and EM.- Approximate Inference.- Sampling Methods.- Continuous Latent Variables.- Sequential Data.- Combining Models.

Pattern Recognition and Machine Learning

How the mind works

Cognitive engineering

The widespread presence of motion sensors on users' personal mobile devices has spawned a growing research interest in human activity recognition (HAR). However, when deployed at a large-scale, e.g., on multiple devices, the performance of a HAR system is often significantly lower than in reported research results. This is due to variations in training and test device hardware and their operating system characteristics among others. In this paper, we systematically investigate sensor-, device- and workload-specific heterogeneities using 36 smartphones and smartwatches, consisting of 13 different device models from four manufacturers. Furthermore, we conduct experiments with nine users and investigate popular feature representation and classification techniques in HAR research. Our results indicate that on-device sensor and sensor handling heterogeneities impair HAR performances significantly. Moreover, the impairments vary significantly across devices and depends on the type of recognition technique used. We systematically evaluate the effect of mobile sensing heterogeneities on HAR and propose a novel clustering-based mitigation technique suitable for large-scale deployment of HAR, where heterogeneity of devices and their usage scenarios are intrinsic.

Smart Devices are Different: Assessing and MitigatingMobile Sensing Heterogeneities for Activity Recognition

We present novel accelerometer-based techniques for accurate and fine-grained detection of transportation modes on smartphones. The primary contributions of our work are an improved algorithm for estimating the gravity component of accelerometer measurements, a novel set of accelerometer features that are able to capture key characteristics of vehicular movement patterns, and a hierarchical decomposition of the detection task. We evaluate our approach using over 150 hours of transportation data, which has been collected from 4 different countries and 16 individuals. Results of the evaluation demonstrate that our approach is able to improve transportation mode detection by over 20% compared to current accelerometer-based systems, while at the same time improving generalization and robustness of the detection. The main performance improvements are obtained for motorised transportation modalities, which currently represent the main challenge for smartphone-based transportation mode detection.

/pdf/accelerometer-based-transportation-mode-detection-on-1gic4c2v8y.pdf

Accelerometer-based transportation mode detection on smartphones

Mobile applications for activity monitoring are regarded as a high potential field for efficient improvement of health care solutions. The measurement of physical activity within every-day conditions should be as easy as using an automatic weighing machine. Up to now physical activity monitoring required special sensor devices and are not suitable for an every day usage. Movement pattern recognition based on acceleration data enables the usage of standard mobile phones for measurement of physical activity. Now, just by carrying a standard phone in a pocket, the device provides information about the type, intensity and duration of the performed activity. Within the project DiaTrace, we developed the method and algorithm to detect activities like walking, jumping, running, cycling or car driving. Based on activity measurement, this application also calculates the consumed calories over the day, shares activity progress with friends or family and might deliver details about different kinds of transportation during a business trip. The DiaTrace application can easily used today by standard phones which are already equipped with the required sensors.

/pdf/activity-recognition-for-everyday-life-on-mobile-phones-4pqdapxci2.pdf

Activity Recognition for Everyday Life on Mobile Phones

In this paper, we present an evaluation of vibrotactile onbody feedback for the purpose of pedestrian navigation. For this specific task, many researchers already provide different approaches such as vibrating belts, wristbands or shoes. Still, there are issues left that have to be considered, such as which body position is most suitable, what kind of vibration patterns are easy to interpret, and how applicable are vibrotactile feedback systems in real scenarios. To find answers, we reconstructed prototypes commonly found in literature and continued to further evaluate different foot-related designs. On the one hand, we learned that vibrotactile feedback at the foot reduces visual attention and thus also potentially reduces stress. However, on the other hand, we found that urban space can be very diverse, and ambiguous and therefore a vibrotactile system cannot completely replace common path finding systems for pedestrians. Rather, we envision such a system to be applied complementary as an assistive technology.

Exploring vibrotactile feedback on the body and foot for the purpose of pedestrian navigation

EarFieldSensing (EarFS) is a novel input method for mobile and wearable computing using facial expressions. Facial muscle movements induce both electric field changes and physical deformations, which are detectable with electrodes placed inside the ear canal. The chosen ear-plug form factor is rather unobtrusive and allows for facial gesture recognition while utilizing the close proximity to the face. We collected 25 facial-related gestures and used them to compare the performance levels of several electric sensing technologies (EMG, CS, EFS, EarFS) with varying electrode setups. Our developed wearable fine-tuned electric field sensing employs differential amplification to effectively cancel out environmental noise while still being sensitive towards small facial-movement-related electric field changes and artifacts from ear canal deformations. By comparing a mobile with a stationary scenario, we found that EarFS continues to perform better in a mobile scenario. Quantitative results show EarFS to be capable of detecting a set of 5 facial gestures with a precision of 90% while sitting and 85.2% while walking. We provide detailed instructions to enable replication of our low-cost sensing device. Applying it to different positions of our body will also allow to sense a variety of other gestures and activities.

EarFieldSensing: A Novel In-Ear Electric Field Sensing to Enrich Wearable Gesture Input through Facial Expressions

The availability of new hardware allows new interaction and visualization metaphors for work environment and private life. Smart watches with integrated acceleration sensors and vibration feedback enable a new kind of interaction. In complement with a matrix display and connectivity to the Internet, smart watches provide assistance in ambient surroundings. This work presents challenges and opportunities of smart watches and introduces a concept of non-obtrusive interaction with smart watches by using physical activity recognition. Hereby the paper presents a prototype application of maintenance assistance and shows possible applications for ambient assistant living environments. In combination of information visualization, permanent sensing and notification, the smart watch becomes the next technology step in the row of smart devices for home-consumer and industrial users.

Ambient interaction by smart watches

Since the human body is a living organism, it emits various life signs which can be traced with an action potential sensitive electromyography, but also with motion sensitive sensors such as typical inertial sensors. In this paper, we present a possibility to recognize the heart rate (HR), respiration rate (RR), and the muscular microvibrations (MV) by an accelerometer worn on the wrist. We compare our seismocardiography (SCG) / ballistocardiography (BCG) approach to commonly used measuring methods. In conclusion, our study confirmed that SCG/BCD with a wrist-worn accelerometer also provides accurate vital parameters. While the recognized RR deviated slightly from the ground truth (SD=16.61%), the detection of HR is non-significantly different (SD=1.63%) to the gold standard.

A study on measuring heart- and respiration-rate via wrist-worn accelerometer-based seismocardiography (SCG) in comparison to commonly applied technologies

Bodo Urban

Papers

Activity Recognition for Everyday Life on Mobile Phones

Exploring vibrotactile feedback on the body and foot for the purpose of pedestrian navigation

EarFieldSensing: A Novel In-Ear Electric Field Sensing to Enrich Wearable Gesture Input through Facial Expressions

Ambient interaction by smart watches

A study on measuring heart- and respiration-rate via wrist-worn accelerometer-based seismocardiography (SCG) in comparison to commonly applied technologies