Context Recognition based on Object Classification for Wearable Systems

Type: Diploma thesis
Student: Marc Speck
Advisor: Thomas von Büren, Paul Lukowicz

Introduction

Context-aware systems have become a booming research area within the field of wearable computing. Regardless of the difficulty with even defining the term "context", many approaches have been taken to recognize the context of a person. However, none of these approaches has ever described the relation between the topology engendered by the objects in the environment and the current context of a person. This work investigates the potential of context recognition based on object classification.

Ultrasonic Echo-Analysis

The analysis of the echoes of ultrasonic waves has been assessed as the most promising sensing technique. A newly introduced model predicts the echoes of objects and hence allows the assessment of the limit of ultrasonic echo-analysis. Measured echoes compared with the model show similar results but the quality of the model must be further improved. Nonetheless, the measured echoes illustrate that they inherit enough information for a powerful object classification as could be shown
with neural networks.

Implementation

An implementation for a wearable system demonstrates the usability of the approach. Figure 1 depicts a simplified overview of the implemented system.

Conclusion

Ultrasonic echo-analysis has proven as a powerful approach for context recognition. Future work may include an enhancement of the model and an unobtrusive integration of the object-recognition system into a wearable system. Figure 2 shows how a final integration may be realized: the white cylinder attached to the glasses is the currently employed transducer.