Efficient Algorithms to Discover Flock Patterns in Trajectories

With the ubiquitous use of location enabled devices, pattern discovery in trajectories has been receiving increasing interest. Among such patterns, we have queries related to how groups of moving objects behave over time such as discovering flocks. A flock pattern is defined as a set of moving objects that move within a predefined distance to each other for a given continuous period of time. A typical application example is surveillance, where relies on discovering flocks on very large streaming spatiotemporal data efficiently. Previous work presented a polynomial solution to the problem of finding flocks with fixed time duration. And presented as well a set of algorithms based on this solution, which are the state-of-the-art algorithms regarding this problem. In this paper, we improve those algorithms by applying the plane sweeping technique in conjunction to an inverted index. The plane sweeping accelerates the detection of groups of objects that are candidates to be a flock in a time instant and the inverted index is used to compare candidate disks across time instants quickly. Using an assortment of real-world trajectory datasets, we show that our proposed methods are very efficient. When compared with the baseline flock algorithm, our proposed methods achieved up to 46x speedup reducing the elapsed time from thousands of seconds to milliseconds.