Motion sensing: From single sensors to sensor networks

It is well known, that regular physical activity is an important factor for preserving the health status, the challenge is how to quantify it accurately. Similarly rehabilitation programs rely on physical exercises, where the best results can be achieved through daily training. But who monitors and evaluates exercise execution at home? Micro electro mechanical systems (MEMS) based accelerometers provide a technological solution for inexpensive monitoring systems. The required number of accelerometers within the monitoring system depends on the use case. Quantifying certain activities like walking or cycling can be achieved with only one sensor, while recognizing differences in movements requires more observations and thus a network of accelerometers. We present some typical movement related applications. For these use cases single sensor and multi-sensor systems are compared with respect to their advantages, challenges and limitations. Even signal processing requirements differ from application to application. Two approaches are explained in detail: knowledge-based and model-driven algorithms. While knowledge-based algorithms rely on feature extraction and an inference machine, which infers high level information from these features, model-driven algorithms try to describe relations between collected data and movements.