On Splitting Raw Trajectories

With the surge of data-driven solutions for trajectory analysis operations, the need for accurate trajectory trip data has spiked. However, the available datasets are raw trajectories spanning from hours to years, not representing actual trips for downstream applications. Therefore, pre-processing steps, such as basic rules to extract trips, are needed to use the datasets. However, this paper demonstrates that the current pre-processing steps are not enough and result in low accuracy, negatively affecting the downstream applications. This paper presents an overview of an accurate and scalable algorithm for splitting raw trajectories for trip extraction. We go beyond the basic rules to introduce a realistic definition of a trip and offer two scalable heuristics over the exhaustive brute force approach of the algorithm with similar accuracy. Experimental results show that the proposed algorithm is: (a) far more accurate than the basic rules, (b) scalable when employing either of the heuristics.