Path Recognition with DTW in a Distributed Environment

The Internet of Things is a concept, where various devices are connected in a network and data is exchanged between them. With the help of Internet of Things applications, it is possible to access sensors remotely to collect data from the physical world. The collected data contains potential knowledge, which could be revealed by applying machine learning techniques. Due to the rapid development of Internet of Things applications, the amount of collected data increases enormously. In order to perform computations on large datasets, distributed computing technologies are used.
Recognizing people’s movements is a popular topic in the context of the Internet of Things. Movement patterns are usually sequential and continuous, and can therefore be encoded in the form of time series. Since the Dynamic-Time-Warping (DTW) is an established algorithm for processing time series data, it is chosen as a similarity measure for different movement patterns. Moreover, based on the DTW results, the movements are classified.
In this thesis, we provide an implementation for the recognition of movement patterns. The prototype is built on Apache Spark and Apache Hadoop and uses their distributed computation possibilities. In an experiment, data from probands is collected and evaluated. Finally, the algorithm performance and accuracy is measured.