Print
Univis
Search
 
FAU-Logo
Techn. Fakultät Willkommen am Institut für Informatik FAU-Logo

Activity Recognition

Activity Recognition
  • Insufficient physical activity is the 4th leading risk factor for mortality. Methods for assessing the individual daily life activity are of major interest in order to monitor the health status and to provide mandatory feedback about the individual quality of life. The assessment of daily life activities with self-reports induces problems like reliability, validity, and sensitivity. In recent years, small and light-weight wearable sensors provide a reliable and objective method for the assessment of daily life activities.

    The picture shows the sensor setup. With this sensor setup, we measured daily life activities. After the feature extraction, we can classify the level and type of activity. 

     

    Please cite this publication when using DaLiAc:

    Leutheuser, H.Schuldhaus, D.Eskofier, B. M. (2013) Hierarchical, multi-sensor based classification of daily life activities: comparison with state-of-the-art algorithms using a benchmark dataset, PLoS ONE, 8(10), e75196. 

    A more detailed description of the DaLiAc (Daily Life Activities) database can be found on Opens internal link in current windowhttp://www.activitynet.org

Early Event Detection
  • A considerable number of wearable system applications necessitate early event detection (EED). EED is defined as the detection of an event with as much lead time as possible. Applications include physiological (e.g., epileptic seizure or heart stroke) or biomechanical (e.g., fall movement or sports event) monitoring systems. EED for wearable systems is under-investigated in literature. Therefore, we introduce a novel EED framework for wearable systems based on hybrid Hidden Markov Models. Our study specifically targets EED based on inertial measurement unit (IMU) signals in sports. We investigate the early detection of high intensive soccer kicks, with the possible pre-kick adaptation of a soccer shoe before the shoe-ball impact in mind. We conducted a study with ten subjects and recorded 226 kicks using a custom IMU placed in a soccer shoe cavity. We evaluated our framework in terms of EED accuracy and EED latency. In conclusion, our framework delivers the required accuracy and lead times for EED of soccer kicks and can be straightforwardly adapted to other wearable system applications that necessitate EED. 

     

    Articles in Conference Proceedings
    A Framework for Early Event Detection for Wearable Systems
    Proceedings of the 2015 ACM International Symposium on Wearable Computers (The 2015 ACM International Joint Conference on Pervasive and Ubiquitous Computing), Osaka, Japan, 07.09.2015 - 11.09.2015, pp. 109-112, 2015, ISBN 978-1-4503-3578-2 (BiBTeX, Who cited this?)
Classification of Daily Life Activities

  • Hierarchical, Multi-Sensor Based Classification of Daily Life Activities: Comparison with State-of-the-Art Algorithms Using a Benchmark Dataset

    Insufficient physical activity is the 4th leading risk factor for mortality. Methods for assessing the individual daily life activity (DLA) are of major interest in order to monitor the health status and to provide mandatory feedback about the individual quality of life. The assessment of DLAs with self-reports induces problems like reliability, validity, and sensitivity. In recent years, small and light-weight inertial sensors were used to provide objective measurements of physical activity. One major research field is the classification of DLAs, e.g. walking, washing dishes or climbing stairs.

    We propose a hierarchical classifier structure that is flexible in the integration of new activities. Furthermore, the sensor fusion of accelerometer and gyroscope improves the distinction of activities like ascending/descending stairs.

    Since it is difficult to compare newly proposed methods to existing approaches, we provide an publicly available dataset (Opens external link in new windowwww.activitynet.org). The hierarchical, multi-sensor based classification system was compared to state-of-the-art algorithms using a benchmark dataset.


    Classification of Daily Life Activities by Decision Level Fusion of Inertial Sensor Data

    Since new and better sensors are frequently available on the market, an optimal activity recognition system should integrate such devices without much effort. We propose a decision level fusion technique, in which each sensor node independently classifies the performed activity. The decision of the sensors are fused and combined for a final decision.


    Towards Big Data for Activity Recognition: A Novel Database Fusion Strategy

    Usually, activity recognition algorithms are evaluated on one specific database which is limited in the number of subjects, sensors, and type of activities. A novel database fusion strategy is proposed which fuses different available databases to one large database. This fusion of databases addresses the two attributes high volume and high variety of the term "big data".


    Articles in Conference Proceedings
    9th International Conference on Body Area Networks (BODYNETS 2014), London, Great Britain, 29.09.2014-01.10.2014, pp. 97-103, 2014 (BiBTeX, Who cited this?)
    Journal Articles
    Hierarchical, Multi-Sensor Based Classification of Daily Life Activities: Comparison with State-of-the-Art Algorithms Using a Benchmark Dataset
    PLos ONE, vol. 8, no. 10, pp. e75196, 2013 (BiBTeX, Who cited this?)
    Articles in Conference Proceedings
    8th International Conference on Body Area Networks (8th International Conference on Body Area Networks (BODYNETS 2013)), Boston, USA, 30.9.2013-2.10.2013, pp. 77-82, 2013 (BiBTeX, Who cited this?)