Knowledge Acquisition from Sensor Data in an Equine Environment

Recent advances in sensor technology have led to a rapid growth in the availability of accurate, portable and low-cost sensors. In the Sport and Health Science domains, this has been used to deploy multiple sensors in a variety of situations in order to monitor participant and environmental factors of an activity or sport. As these sensors often output their data in a raw, proprietary or unstructured format, it is difficult to identify periods of interest, such as events or actions of interest to the Sport and Exercise Physiologists. In our research, we deploy multiple sensors on horses and jockeys while they engage in horse-racing training exercises. The Exercise Physiologists aim to identify events which contribute most to energy expenditure, and classify both the horse and jockey movement using basic accelerometer sensors. We propose a metadata driven approach to enriching the raw sensor data using a series of Profiles. This data then forms the basis of user defined algorithms to detect events using an Event-Condition-Action approach. We provide an Event Definition interface which is used to construct algorithms based on sensor measurements both before and after integration. The result enables the end user to express high level queries to meet their information needs.