원문정보
초록
영어
Healthcare, due to the aging of western populations, requires new technologies to help assisting the needs of elders. The smart home paradigm is one of the promising new trends of research aiming to bring socially and economically viable solutions to this challenge. One of the most crucial problems in developing smart environment is activity recognition. It can be defined as the process of inferring, with various sensors, what the patient is doing and then, being able to predict what he might do in the future. We can find in the literature a lot of works on this theme, however the majority remain essentially theoretical. More specifically, they often work only on a particular component of the activity recognition process, for example by focusing only on the hardware (sensors) or solely on the high level recognition part, assuming that low level recognition already works. Furthermore, we noticed that most available recognition test platforms with an infrastructure, such as MavHome, are static and involve a complex set of sensors, which inevitably has a heavy cost. The work presented in this paper aims of providing solutions to these problems by proposing a way to implement from A to Z a complete recognition platform that works, is simple to use, inexpensive, sturdy and portable. This platform is based only on RFID tags and can be reuse everywhere to test various recognition algorithms, even directly at the patients’ home. We also present a first experimentation conducted with this platform using data mining recognition algorithms.
목차
1. Introduction
2. Selecting and Modeling the Right Activities
3. Choosing the Right Set of Sensors
3.1. RFID Technology: Using Passive or Active Tags?
3.2. Which Sub-type of Passive Tags Should-we Use?
3.3. How to Attach Tags with Objects and How Many are Needed?
4. Activity Recognition based on RFID Tags
4.1. Hardware Setup
4.2. Platform Test Layout
4.3. Software Setup
4.4. Activity Recognition Using a Data Mining Approach
4.5. Activity Record
4.6. Base Frame Elaboration
4.7. Final Frame Creation
4.8. Temporal Aspect
4.9. Use of the C4.5 Algorithm on Data
5. Experiments
5.1. Used Learning Protocol
5.2. Test Protocol
5.3. Obtained Results
6. Conclusion and Perspective
Acknowledgement
References