Nowadays, there is a quest for efficient detection of anomalies in time series data. The detection of outlier regions that do not conform to a well-defined behaviour is a complex topic. The Internet-of-Things is generating massive amounts of data using the plethora of available sensing devices. Therefore, there is a need to detect anomalous segments on long term data using automatic methodologies. This information is critical to numerous applications, ranging from heart disease diagnosis to fall detection, including detailed human movement characterisation.
In this work, anomaly detection approaches will be used to characterise inertial data retrieved from human motion. Manufacturing industries rely on a set of predefined motions to optimize the production process. The employees are trained to execute the assigned tasks using methods that must prevent the development of health issues. During a complete working day, there are several planned or unplanned anomalies in the working process, which may indicate failures on the productivity and may also constitute threats to proper movement execution. Therefore, it is of utmost importance to monitor movement execution during the accomplishment of the assigned tasks, as such to ensure an equal balance between ergonomics and productivity.
This project will therefore allow developing novel algorithms for anomaly detection in time series, which although being centred on inertial repetitive human motion data can be re-used in another context where anomaly detection is required.
In this work, the aim is to develop anomaly detection algorithms applied to time series. The primary outcome consists of the possibility to detect work related anomalies in manufacturing contexts. The secondary outcome is the re-use of the developed algorithms to retrieve anomalies in human motion within AAL projects (fall and activity monitoring).
Author: Rui Varandas
Type: MSc thesis
Partner: Faculdade de Ciências e Tecnologia da Universidade Nova de Lisboa