2DTAX, A Framework for Clustering Two-dimensional Trajectories Data

Abstract—Searching moving object trajectories in 2dimensional space is a big challenge. Much research work has been performed on this field for many years. However, due to many factors such as sensor failures, noises, and sampling rates, it’s very difficult to design a robust and fast method to retrieve or to do clustering these data. TAX (Textual ApproXimation) is one of the methods for searching one-dimensional time series data, such as stock or electrocardiogram data. TAX has been proved to be more accurate on average than existing methods. The main idea behind TAX is to approximate time series data as a set of temporal terms to apply document retrieval methods. The main problem of applying TAX to multi-dimensional data, such as trajectory data and motion capture data, is how to extract temporal terms from multi-dimensional time series data. In this paper, we propose an unified framework, called 2D-TAX, to cluster 2-dimensional moving object trajectories employing the TAX idea. Employing the TAX idea allows us to pre-process original multi-dimensional time series data into sequence-like data structure, which becomes easier to analyze and classify. With our framework, temporal-terms of 2D data can be obtained by decomposing these data into smaller segments, and by assigning a temporal-term for segments which have similar moving pattern. Our experimental results confirmed that our 2D-TAX is more effective than existing methods in clustering trajectory data.