Discussion

Outcomes

The project's ambulation monitoring and fall detection system was considered successful. This outcome was the result of a number of factors, which are discussed below.

Foot sensors

The application of the Vermahide material provided by Professor Andy Russell to produce simple but effective foot-pressure sensors was one of the unexpected successes of the project. Despite their rudimentary design and implementation, these sensors provided the processing software with easily distinguishable pressure signals, without which the project would have been very difficult.

Software

The software was designed to be flexible and generic, capable of handling and processing many types of DBN, not necessarily just related to ambulation monitoring. Different input and feature extraction configurations can be specified at runtime, without the need to re-compile. These features greatly simplified the development of the numerous DBN models tested with the system (Chapter 8).

The software was also designed to easily accommodate future improvements, such as different types of data input, more feature extraction functions and improvements to the graphical user interface, such as displaying DBN beliefs as they are produced.

Problems and difficulties encountered

One of the major difficulties encountered in the project was extracting useful ambulation information from the piezoelectric accelerometers supplied by Compumedics. Although they were eventually determined to be of little use in the application of ambulation monitoring and fall detection, they did encourage the development of the software's feature extraction and signal processing subsystem.

Further development

This project includes many aspects which could benefit from further developments. Processing of ambulation data would probably benefit from the application of DBN learning software^10.1. Nicholson also suggested the addition of `meta-level' DBNs, which could be used to process the output of the models produced here at a higher level still - such as determining activity or energy levels of a subject. The software system is already capable of running multiple DBNs simultaneously, so it should only require the addition of an interface for transferring beliefs to higher-level processing.

Vermahide pressure transducers

Development of the Vermahide pressure sensors also encouraged ideas for a number of further applications for this interesting material in similar sensors. Compact sensors incorporating Vermahide could be developed to fit into a shoe [22] in order to measure pressure or detect footfalls with more accuracy than the sensors developed here. Such sensors may find applications in areas such as monitoring and diagnosis of gait disorders, analysis of running or jogging styles for athletes, and judging in sporting events^10.2.