Eugen Berlin

Detecting leisure activities with dense motif discovery

This paper proposes an activity inference system that has been designed for deployment in mood disorder research, which aims at accurately and efficiently recognizing selected leisure activities in week-long continuous data. The approach to achieve this relies on an unobtrusive and wrist-worn data logger, in combination with a custom data mining tool that performs early data abstraction and dense motif discovery to collect evidence for activities. After presenting the system design, a feasibility study on weeks of continuous inertial data from 6 participants investigates both accuracy and execution speed of each of the abstraction and detection steps. Results show that our method is able to detect target activities in a large data set with a comparable precision and recall to more conventional approaches, in approximately the time it takes to download and visualize the logs from the sensor.

Beyond activity recognition: skill assessment from accelerometer data

The next generation of human activity recognition applications in ubiquitous computing scenarios focuses on assessing the quality of activities, which goes beyond mere identification of activities of interest. Objective quality assessments are often difficult to achieve, hard to quantify, and typically require domain specific background information that bias the overall judgement and limit generalisation. In this paper we propose a framework for skill assessment in activity recognition that enables automatic quality analysis of human activities. Our approach is based on a hierarchical rule induction technique that effectively abstracts from noise-prone activity data and assesses activity data at different temporal contexts. Our approach requires minimal domain specific knowledge about the activities of interest, which makes it largely generalisable. By means of an extensive case study we demonstrate the effectiveness of the proposed framework in the context of dexterity training of 15 medical students engaging in 50 attempts of surgical activities.

Coming to grips with the objects we grasp: detecting interactions with efficient wrist-worn sensors

The use of a wrist-worn sensor that is able to read nearby RFID tags and the wearers gestures has been suggested frequently as a way to both detect the objects we interact with and to identify the interaction. Making such a prototype feasible for longer-term deployments is far from solved however, as plenty of challenges remain in the hardware, embedded algorithms, and the overall design of such a bracelet-like device. This paper presents several of the challenges that emerged during the development of a functioning prototype that is able to sense interaction data for several days. We focus in particular on RFID tag reading range optimization, efficient data logging methods, meaningful evaluation techniques, and long-term deployments.

Enhancing Accelerometer-Based Activity Recognition with Capacitive Proximity Sensing

Activity recognition with a wearable accelerometer is a common investigated research topic and enables the detection of basic activities like sitting, walking or standing. Recent work in this area adds different sensing modalities to the inertial data to collect more information of the user’s environment to boost activity recognition for more challenging activities. This work presents a sensor prototype consisting of an accelerometer and a capacitive proximity sensor that senses the user’s activities based on the combined sensor values. We show that our proposed approach of combining both modalities significantly improves the recognition rate for detecting activities of daily living.

When Else Did This Happen? Efficient Subsequence Representation and Matching for Wearable Activity Data

In long-term activity recognition, large sets of inertial sensor data need to be analyzed in which physical actions of the sensor’s wearer are captured non-stop for weeks to months. These massive time sequences often burden the processing, and especially any post-analysis of the data. We propose a method that approximates and matches accelerometer time series, that is fast on large data sets, well-suited to human acceleration data, and efficient to log on the sensors. Experiments show that approximation and matching are faster than traditional methods, while remaining competitive in recognition of motion patterns.

Enabling Efficient Time Series Analysis for Wearable Activity Data

Long-term activity recognition relies on wearable sensors that log the physical actions of the wearer, so that these can be analyzed afterwards. Recent progress in this field has made it feasible to log high-resolution inertial data, resulting in increasingly large data sets. We propose the use of piecewise linear approximation techniques to facilitate this analysis. This paper presents a modified version of SWAB to approximate human inertial data as efficiently as possible, together with a matching algorithm to query for similar subsequences in large activity logs. We show that our proposed algorithms are faster on human acceleration streams than the traditional ones while being comparable in accuracy to spot similar actions, benefitting post-analysis of human activity data.

Towards Benchmarked Sleep Detection with Wrist-Worn Sensing Units

The monitoring of sleep by quantifying sleeping time and quality is pivotal in many preventive health care scenarios. A substantial amount of wearable sensing products have been introduced to the market for just this reason, detecting whether the user is either sleeping or awake. Assessing these devices for their accuracy in estimating sleep is a daunting task, as their hardware design tends to be different and many are closed-source systems that have not been clinically tested. In this paper, we present a challenging benchmark dataset from an open source wrist-worn data logger that contains relatively high-frequent (100Hz) 3D inertial data from 42 sleep lab patients, along with their data from clinical polysomnography. We analyse this dataset with two traditional approaches for detecting sleep and wake states and propose a new algorithm specifically for 3D acceleration data, which operates on a principle of Estimation of Stationary Sleep-segments (ESS). Results show that all three methods generally over-estimate for sleep, with our method performing slightly better (almost 79% overall median accuracy) than the traditional activity count-based methods.

Low-power lessons from designing a wearable logger for long-term deployments

The advent of a range of wearable products for monitoring ones healthcare and fitness has pushed decades of research into the market over the past years. These units record motion and detect common physical activities to assist the wearer in monitoring fitness, general state of health, and sleeping trends. Most of the detection algorithms on board of these devices however are closed-source and the devices do not allow the recording of raw inertial data. This paper presents a project that, faced by these limitations of commercial wearable products, set out to create an open-source recording platform for activity recognition research that (1) is sufficiently power-efficient, and (2) remains small and comfortable enough to wear, to be able to record raw inertial data for extended periods of time. We study especially, via high-resolution power profiling, several trade-offs present in the choice for the basic hardware components of our prototype, and contribute with three key design areas that have had a significant impact on our prototype design.

Trainspotting: Combining fast features to enable detection on resource-constrained sensing devices

This paper focuses on spotting and classifying complex and sporadic phenomena directly on a sensor node, whereby a relatively long sequence of sensor samples needs to be considered at a time. Using fast feature extraction from streaming data that can be implemented on the sensor nodes, we show that on-sensor event classification can be achieved. This approach is of particular interest for wireless sensor networks as it promises to reduce wireless traffic significantly, as only events need to be transmitted instead of potentially large chunks of inertial data. The presented approach characterizes the essence of an events signal by combining several simple features on low-cost MEMS inertial data. Using a scenario and real data from vibration signatures generated by passing trains, we show how with this approach the classification of passing trains is possible on miniature nodes placed near the railroad tracks. Experiments show that, at the cost of slightly more local processing, the chosen features produce good train type classification with up to 90% of trains correctly identified.

Assessing activity recognition feedback in long-term psychology trials

The physical activities we perform throughout our daily lives tell a great deal about our goals, routines, and behavior, and as such, have been known for a while to be a key indicator for psychiatric disorders. This paper focuses on the use of a wrist-watch with integrated inertial sensors. The algorithms that deal with the data from these sensors can automatically detect the activities that the patient performed from characteristic motion patterns. Such a system can be deployed for several weeks continuously and can thus provide the consulting psychiatrist an insight in their patients behavior and changes thereof. Since these algorithms will never be flawless, however, a remaining question is how we can support the psychiatrist in assigning confidence to these automatic detections. To this end, we present a study where visualizations at three levels from a detection algorithm are used as feedback, and examine which of these are the most helpful in conveying what activities the patient has performed. Results show that just visualizing the classifiers output performs the best, but that users confidence in these automated predictions can be boosted significantly by visualizing earlier pre-processing steps.