Goal: ADM technology will release the individual (e.g. an obese patient) from logging all diet- and food-related intake details. As it is typically needed when participating in weight management programs.
Approach: On-body sensors help to detect information such as time, amount, and category of food - every time food or a drink was consumed. We use multi-modal sensory input and adapted pattern recognition techniques to spot activities related to intake, including intake gestures (movements of arms, upper body), chewing and swallowing. Challenges are related to the intrinsic complexity and variability of food intake (and hence, the activity pattern complexity) and the wearing comfort of the sensors.
Achievements: On-body sensors were found/developed that are comfortable and minimally invasive for the user - they can be easily attached to the body or integrated into clothing. A robust detection and inference framework was developed that incorporates the activity detection information (activity events) to segment intake (feeding) cycles. More...
Goal: Rehabilitation Assistants will support the individual (e.g. an orthopedics patient) with a wearable system that can track exercise performance.
Approach: We develop textile sensing and recognition solutions that are integrated into garments. Sensing approaches include novel strain-sensitive yarns, and inertial sensors. Challenges include the sensor integration into fabrics, sensor stability (over time, wearing and washing cycles) and the limited computational resources for on-body processing and pattern recognition.
Achievements: Stable sensing approaches were developed and various garment prototypes (using the different sensing solutions) have been built by some of my talented colleagues: Corinne Mattmann and Holger Harms. We developed solutions and performance evaluation schemes to recognize postures and rehabilitation exercises, supported by clinical experts: Corina Schuster, Reha Rheinfelden. More details can be found in our publications: Mattmann, Amft et al., ISWC 2007, Harms, Amft et al., BodyNets 2008, Harms, Amft et al., BodyNets 2009, Harms, Amft et al., Ubicomp 2009.
Goal: DART aims to evaluate the potential of sensor networks for distributed activity recognition and to develop a software framework (Titan) for convenient application deployment.
Approach: Sensors spread in the environment and on-body are used to recognize activity and context of a user. Each sensor node performs a local activity pattern detection and classification. These condensed activity events are subsequently used for more complex activity inference, e.g. to detect and validate a task sequence. This is a joint work with my colleague Clemens Lombriser working on Titan.
Achievements: We developed a solution to fuse events detected by different nodes belonging to the same activity and successfully evaluated simple composite activity inference schemes (network-central integrator node) in a car assembly task with 12 sensor nodes, an alphabet of 47 atomic activities and 11 composites. See Amft et al., EuroSSC 2007.
Goal: The project aims to provide a software for rapid prototyping of context recognition applications.
Approach: The CRN Toolbox is consists of (1) reusable algorithm components, (2) graphical configuration frontend, (3) flexible communication mechanisms and (4) interfaces to external tools. This project is a joint work with my colleague David Bannach, University Passau.
Achievements: The CRNT has been used in various industrial and research projects. It is available under LGPL from crnt.sourceforge.net. Please see our publications: Bannach et al., ARCS 2006, Bannach, Amft, Lukowicz, IEEE Pervasive Computing 7(2) 2008 . More...
Goal: The Parking Game was developed as demonstrator for the CRN Toolbox and to test my activity spotting recognition.
Approach: The game concept involves the player's physical activity in a 3D virtual reality game. The player weaves hand and arm gestures to signal commands to a virtual car driver, helping to park the car in an available parking spot. Gesture commands are recognised by an online implementation of my activity spotting algorithms in the CRNT. The player wears a glove with inertial sensors at the right hand. This is a joint project with David Bannach, University Passau.
Achievements: The game was initially developed and demonstrated for a Activity Recognition Workshop held at ISWC 2006. Details can be found in Bannach, Amft et al., CIG 2007. More...
Goal: The project targeted a wearable system for tracking the user's location indoors.
Approach: In our approach lights were tracked by solar cells that (as an intended side-effect) generate power. Flexible solar cells attached to the shoulders of a jacket were foreseen. An empirical and theoretic evaluation of building lighting was combined with practical tracking experiments.
Achievements: A model for the spacial resolution of lights was derived. From experiments a distance estimation accuracy of 21cm was achieved. See our publications: Randall, Amft et al., LOCA 2005, Amft et al., IFAWC 2005, and Randall, Amft et al., Personal and Ubiquitous Computing, 11(6):417–428. More...
Goal: The project investigated different sensing solutions to determine muscle contraction and consequently quantify activity intensity.
Approach: Out work concentrated on mechanical sensing of force (FSR) and fabric strain. We evaluated their feasibility for different muscles at the lower arm. This is a joint project with Paul Lukowicz and Holger Junker.
Achievements: A solution was developed to use FSR sensors to detect muscle activity from small muscles at the lower arm. Fabric strain sensors were used at the arm to track limb circumference changes. See Amft et al., BSN 2006.
