The ubiquitous availability of heterogeneous sensor data created by Internet-of-Things (IoT) technologies and Industry 4.0 trends drastically accelerated the development of machine learning applications. AutoML services enable users with sparse machine learning knowledge to develop AI-based applications and rapidly evaluate the feasibility of data-driven ideas. Therefore, there exists a demand for holistic, low-code, end-to-end AutoML systems, which cover all stages of the machine learning lifecycle (i.e., feature engineering, model training, evaluation, versioning, provisioning, etc.). Although there are proprietary, cost-intensive platforms addressing these issues, no open-source solutions covering these aspects are known to us. In this paper we present AutoTiM, an open-source service capable of creating and operating highly performant machine learning models without requiring domain expertise or machine learning knowledge.

Deep Learning models for mapping documents from different domains, e.g., text, images, and audio, into a common vector space, enable a seamless information retrieval between the different domains and, thus, significantly improve the user experience of many expert tools. Despite various models for multi-modal mappings presented in scientific literature, the implementation and integration remain a challenge within the industry, especially for small or medium-sized companies. Reasons are, that developing such retrieval systems for production use-cases is a non-trivial task, requiring scalable, reliable, and cost-efficient infrastructure, services as well as adequate Deep Learning models. We present a generic and flexible blueprint architecture, targeting the development of a production-ready image-text retrieval search system using Kubernetes,MLflow, Elasticsearch, and integrating state-of-the-art Deep Learning models.

In this paper, we make the case for building interdisciplinary scenarios, integrating technological, business and user perspective during the fuzzy front-end of innovation. We start from a living lab framework, underpinning the iterative process to integrate insights from all three perspectives within research projects. Then, we explain how the approach was applied within the HRADIO project. This EU funded project focuses on the development of hybrid radio applications. We demonstrate how the process of building interdisciplinary scenarios and the involvement of multiple perspectives to evaluate these scenarios, has enabled both strategic decision making (1) and improved the technical development process within the HRADIO project (2). The first focuses on the (lack of) interest in certain scenarios from a business, technology and/or user perspective, to explain: (a) how these thresholds can be overcome and (b) what the consequences are for the project. The latter explains how we moved from interdisciplinary scenarios to a first Minimum Viable Product (MVP) that could be tested in the first pilot phase, to prototypes to be tested in the second pilot phase. To validate the importance of the scenarios, we also explain how the scenarios are integrated into the exploitation timeline. Finally, and importantly, we address how the process of multidisciplinary scenario building and evaluation can be improved.

The world of linear radio broadcasting is characterized by a wide variety of stations and played content. That is why finding stations playing the preferred content is a tough task for a potential listener, especially due to the overwhelming number of offered choices. Here, recommender systems usually step in but existing content-based approaches rely on metadata and thus are constrained by the available data quality. Other approaches leverage user behavior data and thus do not exploit any domain-specific knowledge and are furthermore disadvantageous regarding privacy concerns. Therefore, we propose a new pipeline for the generation of audio-based radio station fingerprints relying on audio stream crawling and a Deep Autoencoder. We show that the proposed fingerprints are especially useful for characterizing radio stations by their audio content and thus are an excellent representation for meaningful and reliable radio station recommendations. Furthermore, the proposed modules are part of the HRADIO Communication Platform, which enables hybrid radio features to radio stations. It is released with a flexible open source license and enables especially small- and medium-sized businesses, to provide customized and high quality radio services to potential listeners.

The world of linear radio broadcasting is characterized by a wide variety of stations and played content. That is why finding stations playing the preferred content is a tough task for a potential listener, especially due to the overwhelming number of offered choices. Here, recommender systems usually step in but existing content-based approaches rely on metadata and thus are constrained by the available data quality. Therefore, we propose a new pipeline for the generation of audio-based radio station fingerprints relying on audio stream crawling and a deep autoencoder. We show that the proposed fingerprints are especially useful for characterizing radio stations by their audio content and thus are an excellent representation for meaningful and reliable radio station recommendations.

The procedural generation of data sets for empirical algorithm validation and deep learning tasks in the area of primitive-based geometry is cumbersome and time-consuming while ready-to-use data sets are rare. We propose a new and highly flexible framework based on Evolutionary Computing that is able to create primitive-based abstractions of existing triangle meshes favoring fast running times and high geometric variation over reconstruction precision. These abstractions are represented as CSG trees to widen the scope of possible applications. As part of the evaluation, we show how we successfully used the generator to create a data set for the evaluation of neural point cloud segmentation pipelines and additionally explain how to use the system to create artistic abstractions of meshes provided by publicly available triangle mesh databases.

Hybrid radio is an umbrella term for the combination of classic broadcast radio with online services enabling highly personalized and interactive content. Hybrid services heavily rely on well-maintained metadata but currently, a multitude of different data sources and models exist, each with certain aspects and different levels of quality. We propose a distributed metadata platform which harmonizes relevant metadata from a variety of data sources and makes it comfortably searchable. The distributed and open nature of the platform renders centralized aggregators obsolete and allows even smaller stations to participate in a search network which significantly increases their visibility. The capability of the platform is proven by the implementation and evaluation of a metadata-based radio station recommender system which is one of the most important hybrid radio building blocks. Finally, the platform is evaluated by a qualitative analysis which juxtaposes requirements based on pre-defined user scenarios with its technical features.

This paper presents a light-weight process for 3D reconstruction and measurement of chronic wounds using a commonly available smartphone as an image capturing device. The first stage of our measurement pipeline comprises the creation of a dense 3D point cloud using structure-from-motion (SfM). Furthermore, the wound area is segmented from the surrounding skin using dynamic thresholding in CIELAB color space and a surface is estimated to simulate the missing skin in the wound area. Together with a mesh reconstruction of the wound, the skin surface and the segmented wound is used to calculate the wound dimensions, i.e., its length, surface area and volume. We evaluate the presented pipeline using three wound phantoms, representing different stages in healing, and compare the subsequently scanned and measured wound dimensions with manually measured ones.

Visionen vom Internet of Things und der nahtlosen Einbettung der virtuellen Welt in den physischen Alltag des Menschen sind durch ubiquitare Vernetzung, stationare und mobile Computer sowie miniaturisierte Sensorik langst Realitat geworden. Zusammen mit Algorithmen des Data Minings und der kunstlichen Intelligenz werden so kontextsensitive Dienste und vernetzte Alltagsgegenstande geschaffen, welche einen immensen Mehrwert im privaten, kommerziellen und industriellen Raum schaffen. Im Rahmen dieses Szenarios vielbeachtete Forschungsgebiete sind die Erschliesung von menschlichem Kontext und die menschliche Aktivitatserkennung mithilfe von mobiler Sensorik. Wahrend es auf dem Gebiet der rein quantitativen Erkennung von menschlicher Aktivitat bereits viele Verfahren zur Vorhersage und Erkennung von Bewegungsereignissen auf Basis von Bewegungs- oder Tiefeninformationen sowie visueller Sensorik gibt, sind Konzepte zur feingranularen, automatisierten Analyse mit qualitativem Schwerpunkt bislang kaum existent. Typische Anwendungsgebiete fur diese sind zum Beispiel die Identifikation von Notfallsituationen im medizinischen Bereich oder die Erkennung von Fehlstellungen und Anomalien bei physischer, menschlicher Aktivitat. Um solche Fragestellungen aus dem Bereich der Erfassung, Erkennung und qualitativen Analyse von menschlicher Bewegungsaktivitat zu adressieren, wird in dieser Arbeit zunachst ein ganzheitliches, verteiltes Sensorsystem, welches auf Basis von Bewegungsinformationen menschliche Bewegungsaktivitat untersucht, spezifiziert. Anschliesend wird ein Vorgehen zur automatisierten und qualitativen Analyse individueller, wiederkehrender, menschlicher Bewegungsereignisse, mithilfe eines adaptiven Segmentierungsverfahrens und eines Konzepts zur Formalisierung und Diskretisierung von subjektiven Qualitatsmerkmalen in menschlichen Bewegungsablaufen, vorgestellt. Im Anschluss steht die qualitative Untersuchung von nicht vorhersehbarer, menschlicher Bewegungsaktivitat im Fokus. Hierzu werden neue Konzepte zur Segmentierung und zur generischen Projektion der physischen, menschlichen Leistung des Menschen auf diskrete Merkmalsvektoren vorgestellt. Zusammengefasst stellt die vorliegende Arbeit ein umfassendes Paket zur generischen Untersuchung von menschlicher Bewegungsaktivitat vor. Mit diesem lassen sich quantitative und qualitative Analysen von Bewegungsaktivitat fur sowohl wiederkehrende als auch fur nicht vorhersehbare, menschliche Bewegungsereignisse effizient umsetzen.

Analysis of human activity, e.g., by tracking and analyzing motion information or vital signs became lots of attention in medical as well as athletic appliances during the last years. Nonetheless, comprehensive and labeled datasets containing human motion information are only sparsely accessible to the public. Especially qualitatively labeled datasets are rare, although they are of great value for the development of concepts concerning qualitative motion assessment, e.g., to avoid injuries during athletic workouts or to optimize a training’s success. Therefore, we provide an open and qualitative as well as quantitative labeled dataset containing acceleration and rotation data of 8 different body weight exercises, conducted by 26 study participants. It encompasses more than 11,000 exercise repetitions of which we extracted 8,576 into individual segments. We believe, that due to its structure and labeling our work is suitable to serve for development, benchmarking, and validation of new concepts for human activity recognition and qualitative motion assessment (Publication notes: The dataset will be published at http://github.com/andrebert/body-weight-exercises together with this paper’s presentation on the MobiHealth conference 2017, taking place in Vienna, 14–16 November.).

Due to fast distribution of powerful, portable processing devices and wearables, the development of learning-based IoT-applications for athletic or medical usage is accelerated. But besides the offering of quantitative features, such as counting repetitions or distances, there are only a few systems which provide qualitative services, e.g., detecting malpositions to avoid injuries or to optimize training success. Therefore we present a novel, holistic, and sensor-based approach for qualitative analysis of asynchronous, non-recurrent human motion. Furthermore, we deploy it to automatically assess the difficulty level of boulder routes on basis of climbing movements. Within a comprehensive study encompassing 153 ascents of 18 climbers, we extract and examine features such as strength, endurance, and control and achieve a successful classification rate of difficulty levels of more than 98

The great success of wearables and smartphone apps for provision of extensive physical workout instructions boosts a whole industry dealing with consumer oriented sensors and sports equipment. But with these opportunities there are also new challenges emerging. The unregulated distribution of instructions about ambitious exercises enables unexperienced users to undertake demanding workouts without professional supervision which may lead to suboptimal training success or even serious injuries. We believe, that automated supervision and realtime feedback during a workout may help to solve these issues. Therefore we introduce four fundamental steps for complex human motion assessment and present SensX, a sensor-based architecture for monitoring, recording, and analyzing complex and multi-dimensional motion chains. We provide the results of our preliminary study encompassing 8 different body weight exercises, 20 participants, and more than 9,220 recorded exercise repetitions. Furthermore, insights into SensXs classification capabilities and the impact of specific sensor configurations onto the analysis process are given.

Smartphone applications designed to track human motion in combination with wearable sensors, e.g., during physical exercising, raised huge attention recently. Commonly, they provide quantitative services, such as personalized training instructions or the counting of distances. But qualitative monitoring and assessment is still missing, e.g., to detect malpositions, to prevent injuries, or to optimize training success. We address this issue by presenting a concept for qualitative as well as generic assessment of recurrent human motion by processing multi-dimensional, continuous time series tracked with motion sensors. Therefore, our segmentation procedure extracts individual events of specific length and we propose expressive features to accomplish a qualitative motion assessment by supervised classification. We verified our approach within a comprehensive study encompassing 27 athletes undertaking different body weight exercises. We are able to recognize six different exercise types with a success rate of 100% and to assess them qualitatively with an average success rate of 99.3%.

In recent years, the importance of location-based services and indoor positioning systems increased significantly for both, research and industry. Visual localization systems have the advantage of not depending on dedicated infrastructure and thus they are interesting for navigation within buildings. While there are already approaches which are using pre-recorded databases of reference images to obtain an absolute position for a given query image, suitable applications which are estimating the relative movement of pedestrians out of a first person perspective video are still missing. This paper presents a novel approach for a pedometer as well as for an activity detector using a such a first person perspective video stream of a pedestrian as input data. The system counts the number of steps and furthermore detects current activities of a user. Therefore, we analyze all video input data with the SURF algorithm in order to extract robust feature points. Especially the orientation and scaling properties of this feature points are used for an accurate measurement.

Mutual usage of vehicles as well as car sharing became more and more attractive during the last years. Especially in urban environments with limited parking possibilities and a higher risk for traffic jams, car rentals and sharing services may save time and money. But when renting a vehicle it could already be damaged (e.g., scratches or bumps inflicted by a previous user) without the damage being perceived by the service provider. In order to address such problems, we present an automated, motion-based system for impact detection, that facilitates a common smartphone as a sensor platform. The system is capable of detecting the impact segment and the point of time of an impact event on a vehicle's surface, as well as its direction of origin. With this additional specific knowledge, it may be possible to reconstruct the circumstances of an impact event, e.g., to prove possible innocence of a service's customer.

Nowadays, many daily duties being of a private as well as of a business nature are handled with the help of online services. Due to migrating formerly local desktop applications into clouds (e.g., Microsoft Office Online, etc.), services become available by logging in into a user account through a web browser. But possibilities for authenticating a user in a web browser are limited and employing a username with a password is still de facto standard, disregarding open security or usability issues. Notwithstanding new developments on that subject, there is no sufficient alternative available. In this paper, we specify the requirements for a secure, easy-to-use, and third-party-independent authentication architecture. Moreover, we present KeyPocket, a user-centric approach aligned to these requirements with the help of the user’s smartphone. Subsequently, we present its state of implementation and discuss its individual capabilities and features.

—Today’s smartphones are equipped with numerous different sensors and are capable of providing a large number of diverse services. But due to the high energy consumption of built-in sensors, as well as because of the energy consumed while analyzing gathered raw data, a bottleneck in resource supply is likely to occur during a service provision. This bottleneck leads to one of the biggest challenges regarding the develop- ments on mobile devices: the trade-off between a high short-term service performance and sustainable energy management. Furthermore, despite of numerous hardware improvements (e.g., energy saving displays, batteries with bigger capacities, etc.) this issue remains unsolved. Hence, software-based approaches can be used to optimize the resource and energy management on mobile devices according to the user’s preferences, existing context information, and the current energetic state of a device. Moreover, a holistic energy management enables the system to provide context dependent services. In this article, we present a concept for custom tailored service provision on mobile devices in combination with EMMA (Energy Management Middleware Architecture), a modular architechture for managing a mobile device’s service infrastructure in relation to its current resource state as well as to the user’s individual preferences. Additionally, we give an insight into our prototypical application, which demonstrates EMMA’s core concepts including its featured approach of individual service provision.

Mobile tagging became more and more popular in commercials, magazines, newspapers, and other applications during the last years. In context of commercials, a bar code containing the advertisers internet address is often used to refer a customer to related online content. Due to their robustness as well as their comparably high fault-tolerance in case of low quality pictures, QR-Code systems are commonly used for that task. Connected to that topic we present a special procedure for mobile tagging, which uses a distinct logo or image in order to refer to certain information instead of a QR-Code. Our procedure was optimized to work with a conventional smartphone – the only prerequisite for usage is the possession of a smartphone capable of capturing and analyzing the different logos with our smartphone application. To match the logos with related information and to determine their uniqueness we introduce a new similarity measure on basis of SURF feature points and a contour comparison.