A Grid-Based Approach for Measuring Similarities of Taxi Trajectories

Abstract

Similarity measurement is one of the key tasks in spatial data analysis. It has a great impact on applications i.e., position prediction, mining and analysis of social behavior pattern. Existing methods mainly focus on the exact matching of polylines which result in the trajectories. However, for the applications like travel/drive behavior analysis, even for objects passing by the same route the trajectories are not the same due to the accuracy of positioning and the fact that objects may move on different lanes of the road. Further, in most cases of spatial data mining, locations and sometimes sequences of locations on trajectories are most important, while how objects move from location to location (the exact geometries of trajectories) is of less interest. For the abovementioned situations, the existing approaches cannot work anymore. In this paper, we propose a grid aware approach to convert trajectories into sequences of codes, so that shape details of trajectories are neglected while emphasizing locations where trajectories pass through. Experiments with Shanghai Float Car Data (FCD) show that the proposed method can calculate trajectories with high similarity if these pass through the same locations. In addition, the proposed methods are very efficient since the data volume is considerably reduced when trajectories are converted into grid-codes.

gsstSIM: A high‐performance and synchronized similarity analysis method of spatiotemporal trajectory based on grid model representation

“…In response to different application requirements, scholars have proposed classic trajectory similarity analysis methods, such as Hausdorff distance (Hausdorff, 1914), discrete Fréchet distance (Eiter & Mannila, 1994), dynamic time warping (DTW) (Berndt & Clifford, 1994), longest common subsequence (LCSS) (Vlachos et al., 2002), and edit distance with real penalty (ERP) (Chen & Ng, 2004), edit distance with real sequence (EDR) (Chen et al., 2005). In recent years, scholars further improved algorithm efficiency and adapted them to more application scenarios, such as FastDTW (Salvador & Chan, 2007), one‐way distance (OWD) (Lin & Su, 2008), Symmetrized Segment‐Path Distance (SSPD) (Besse et al., 2016), C‐SIM (Mariescu‐Istodor & Fränti, 2017), and Spatial Grid Coding Distance (SGCD) (Jiao et al., 2020). However, the above methods focus on the similarity in spatial dimensions, and cannot meet the requirements in spatiotemporal computation and application in the ICT era.…”

“…Jiao et al. (2020) proposed a spatial grid‐based trajectory similarity measure SGCD (Spatial Grid Coding Distance), which is approximately 2.4 times faster than the FastDTW with a computational complexity of O ( n ). Li et al (2020) also proposed a DGGS‐based trajectory data management and analysis technology system, in which the advantages of trajectory similarity measures based on the DGGS were pointed out.…”