As a substitute for communication with a human driver, additional communication cues for AVs have been proposed [1]. To analyze their effect on traffic flow, preceding studies captured pedestrians´ crossing decision in an unnatural manner, e.g., via data collection devices (e.g., [2]). In this paper, we explored the applicability of two optical tracking systems to capture crossing behavior more naturally in real-world conditions. In a Wizard-of-Oz study, N = 35 participants encountered a simulated AV. They were instructed to show natural crossing behavior, which was recorded with a light barrier and a SMARTTRACK3/IF. Results showed that both optical tracking systems can be used to capture pedestrians´ crossing behavior while interacting with AVs. The light barrier recorded the movement more often validly. Still, the correlation between recordings of both tracking systems was high which indicated a high convergence of measurement results.

Nowadays, fine-grained Human Activity Recognition (HAR) has become extremely interesting among researchers due to its applications in fields such as healthcare, security, sports, and smart environments. In this paper, we briefly describe the current State of the Art approaches in fine-grained human activity recognition. We also discuss the characteristics, complexities, and scarcity of inertial datasets related to fine-grained and coarse-grained activities. To mitigate this scarcity, we present our inertial dataset, consisting of 17 participants performing 4 fine-grained tasks through interaction with an Inertial Measurement Unit (IMU) sensor embedded in a solid object. Next, we test the most commonly used machine learning classifiers (e.g., kNN, XGboost) on the collected dataset and showcase the results. Finally, we demonstrate the necessity of a new approach to deal with the recognition of fine-grained activities and we state our future research directions in this context.

The automatic, sensor-based assessment of challenging behavior of persons with dementia is an important task to support the selection of interventions. However, predicting behaviors like apathy and agitation is challenging due to the large inter- and intra-patient variability. Goal of this paper is to improve the recognition performance by making use of the observation that patients tend to show specific behaviors at certain times of the day or week. We propose to identify such segments of similar behavior via clustering the distributions of annotations of the time segments. All time segments within a cluster then consist of similar behaviors and thus indicate a behavioral predisposition (BPD). We utilize BPDs by training a classifier for each BPD. Empirically, we demonstrate that when the BPD per time segment is known, activity recognition performance can be substantially improved.

During the last decades, disease-related disabilities, primarily caused by stroke have increased worldwide. Neurorehabilitation exercise therapy plays a vital role in the recovery of such disabilities. However, due to global demographic changes and the increasing number of stroke patients, therapists are facing difficulties in coping with the demand. Consequently, the necessity for appropriate technical support to help the therapists for providing helpful progress feedback to the patients is becoming evident. So far, such technological systems are not yet available for clinical usage. Moreover, there is still a lack of research demonstrating the possibility of pursuing the therapeutic exercises by the patients themselves at their homes using non-invasive commodity sensors. In this work, we design a system pipeline containing commodity cameras by which the patients would be able to record their exercises at home; we also evaluate and analyze the acquired data using an off-the-shelf machine-learning framework. The medical experts can utilize our system to monitor the patients’ progress over the prescribed duration of the therapy. Here, rather than using specialized sensors with the body to acquire the movement information of the body joints, which some of the existing works use, we use a machine-learning framework to acquire this information. Our evaluation process demonstrates the situations in which these activities can be reliably acquired with commodity RGB cameras; moreover, the challenging aspects of the acquisition which can affect the accuracy of recognition of the framework are discussed and analyzed.

For many use cases, such as supervised machine learning, labeled data is needed. However, to collect information for labels in real-life contexts, scientists are confronted with the challenge of gathering labeled data over an extended period. Labeling this data can become problematic, as constant supervision, similar to a laboratory setting, is neither feasible nor desired. Therefore, participants of such studies have to label their data themselves via appropriate apps on a smartphone. Nevertheless, this process can become very obtrusive in daily life and might even influence the results, especially studies regarding emotions. For example, in studies where participants need to indicate their stress levels frequently, labels get missed in situations where it would be inappropriate to take the phone. Consequently, missing these labels presents a significant problem. This paper aims to provide an unobtrusive solution to labeling data in real-world studies. We recorded a dataset consisting of five gestures and data from daily life. Thereby, we provide a set of predefined gestures that can be distinguished from other everyday life activities by using accelerometer and gyroscope sensors of wearable devices on the wrist. The use of predefined hand gestures for labeling data can therefore serve as an additional tool for the labeling process. Two machine learning approaches were compared and achieved promising results with Matthews Correlation Coefficients of up to 0.789 for a Random Forest and up to 0.835 for a Convolutional Neural Network.

Currently, the increasing use of mobile technologies is present in the field of activities of daily living (ADL), which demands better design of Mobile Human-Computer Interaction (MobileHCI) in Ambient Assisted Living (AAL) environments. People in AAL environments interact by means of electronic devices (e.g., sensors, actuators, mobile phones, tablets, wearable technologies) to accomplish their ADL. On this matter, a categorization of human activity monitoring approaches is proposed, from which existing solutions are identified on Pervasive Healthcare Evolution, approaches on the categorization for the detection of ADL, and technologies regarding the maintenance and monitoring of individuals at home. Moreover, an overview of evaluation models on the extent of performance of ADL is presented.

Our feet are not just used to walk upright, feet can also reveal important context information on our physical constitution and context. In this research, we drove a multisensory approach to uncover the hidden information our feet hold. We instrumented a shoe with an accelerometer, a gyroscope, FSRs, a microphone, and a humidity sensor, while we drove a conventional machine learning approach. Our results show that we are capable to extract information on body posture (e.g., standing, sitting, kneeling), ambulation activity (e.g., walking, jumping, climbing stairs), gait abnormalities (e.g., overpronation, supination), and terrain (e.g., lawn, gravel, sand). These insights may show a new direction for the future of smart shoes.

Nowadays, many organisations steadily have to face new challenges due to an increasing competition, new technologies and manpower shortage. While dealing with this growth of challenges and confronting employees with higher demands, organizations have to pay attention to employees' personal resources that are connected to their well-being and health. Human energy is a psychological construct connected to different states of experiencing e.g. vigor and vitality, or otherwise fatigue. The subjective experience of human energy varies during the working day. In order to promote employees' health and well-being, an objective measure to determine human energy levels is needed. In this paper, we report on first insights from a pilot study with 12 healthy participants investigating if human’s glucose concentration can serve as an objective measure of human energy. We analyze the possible interplay between subjective human energy perception during work and sensor-based blood sugar levels, assessed by a continuous glucose monitoring (CGM) system in healthy adults.

Smart support systems for the recognition of Activities of Daily Living (ADLs) can help elderly people live independently for longer improving their standard of living. Many machine learning approaches have been proposed lately for Human Activity Recognition (HAR), including elaborated networks that contain convolutional, recurrent, and attentive layers. The ubiquity of wearable devices has provided an increasing amount of time-series data that can be used for such applications in an unobtrusive manner. But there are not many studies on the performance of the attention-based Transformer model in HAR, especially not for complex activities such as ADLs. This work implements and evaluates the novel self-attention Transformer model for the classification of ADLs and compares it to the already well-established approach of recurrent Long-Short Term Memory (LSTM) networks. The proposed method is a two-level hierarchical model, in which atomic activities are initially recognized in the first step and their probability scores are extracted and utilized for the Transformer-based classification of seven more complex ADLs in the second step. The Transformer is used at the second step to classify seven ADLs. Our results show that the Transformer model outperforms LSTM networks, while relying only on attention-mechanism to depict global dependencies between input and output without the need to use any recurrence. The proposed model was tested using two different segment lengths, indicating its effectiveness in learning long-range dependencies of shorter actions in complex activities.

Studies have shown that cognitive function and stamina are vulnerable to aging. Susceptibility to such age-related decline can be related to many factors, including education, literacy, occupation, and engagement in leisure activities. In this paper, we present the design and development of a novel interactive application to exercise older adults’ cognitive function using state-of-the-art natural language processing (NLP). Installed on mobile devices, the app encourages cognitive health in older adults through continuously personalized, inquiry-based information acquisition. Learning tasks are designed to run multimodal pieces of media content aligned with users’ interests followed by multiple-choice questions, their answers, and related distractors dynamically generated by a series of Bidirectional Encoder Representations from Transformers (BERT) enhanced by Convolutional Neural Network (CNN) layers. A usability study was carried out on senior citizens living in the researchers’ community, aged 65-92. The evaluation consisted of sessions, roughly 30 minutes each, over three weeks. User performance measures of information recall represented by accuracy score and response time, and mental workload measures composed of mental demand, temporal demand, user-performance, effort, and frustration were collected. Findings show the viability of attaining a positive user experience and engagement for older adults with inquiry-based information acquisition training personalized by users’ interests, interactions, and language analysis. The work lends insights and potential avenues for improving accessibility and engagement in tasks that stimulate older adults’ cognitive function using mobile devices.

Activity recognition is an increasingly important feature in mobile consumer devices that enables the design of context- aware applications and the customisation of user experiences. Recent deep learning-based recognisers demonstrate promis- ing performances for hand activity recognition – e. g., wash- ing hands – or human activity recognition – e. g., cycling – when exploiting accelerometer data from smartwatches or Magnetic Angular Rate and Gravity (MARG) data from body-worn sensors. However, the robustness and the gen- eralisation capabilities of deep learning models can be an issue, since the consumers’ locomotion data is typically not available at training time. In this paper, we propose for the first time a pipeline of pre-trained deep feature extrac- tors and online learning-capable shallow classifiers – such as passive-aggressive classifiers, random forests, and support vector classifiers – for the personalisation of mobile activity recognition systems. Our suggested solution could enable efficient on-device online learning for personalisation, thereby contributing to users’ privacy preservation. We report a series of experiments exploring the potential and the limitations of this approach to increase user-specific performance given only a limited number of training samples. Our results are encouraging and show (i) that a single additional data point per activity class (in a 25-class problem) from a new user can reduce recognition errors relatively by up to 20 %, and (ii) that with an increasing number of training samples our personalised models outperform comparable approaches in the literature, achieving accuracies of up to 91.4 %.

Electric field sensors are used in a variety of ways to recognize different human actions and behaviors, for example, fall detection or classification of movements. However, very little is known about the number of sensors that are needed to achieve an acceptable recognition rate. Most systems just use as many sensors as possible to achieve confusion matrices with high true positive and true negative rates. In this paper, the relation of recognition rates and the size of a system composed of Passive Electric Field sensors shall be further investigated. For this purpose, several setups to recognize different human activities will be created and evaluated, each with a varying number of sensor tokens.

Working in front of the computer for long periods of time leads to exhaustion, fatigue and high strain on the eyes. Often the blinking rate is disturbed and the eyes do not form enough tears to stay moist. Computer workstations are usually equipped with a webcam, making it convenient to monitor eye blinking and easy to perform blink interventions. In this paper, we present a blink detection method that uses 3D landmarks. After presenting existing publicly available blink databases, we evaluate the accuracy of our blink detection method and discuss the results. In addition, we illustrate the scope of our eye blink detection method in the application of a non-contact stress detection software to support stress reduction in the field of occupational health management.

A major challenge in stationary care is the limited amount of time for each patient due to a large overhead being created by manual documentation efforts. Studies show that it is not uncommon for caregivers to spend more than one hour per day for documentation efforts. In this paper a novel concept for reducing the manual documentation effort by leveraging methods of human activity recognition is introduced and a corresponding dataset is published. The dataset captures different care activities like repositioning, sitting up, transfer and patient mobilization using body worn sensors in a realistic setting with multiple patients and caregivers. For evaluation of the data, two experimental setups are presented. An unsegmented case, where the duration of the care activity is unknown and a segmented case, where the beginning and the end of the activity is known beforehand. First experiments show the feasibility of recognizing care activities using different types of Neural Networks.