Grid-Based Method for GPS Route Analysis for Retrieval

Abstract

Grids are commonly used as histograms to process spatial data in order to detect frequent patterns, predict destinations, or to infer popular places. However, they have not been previously used for GPS trajectory similarity searches or retrieval in general. Instead, slower and more complicated algorithms based on individual point-pair comparison have been used. We demonstrate how a grid representation can be used to compute four different route measures: novelty, noteworthiness, similarity, and inclusion. The measures may be used in several applications such as identifying taxi fraud, automatically updating GPS navigation software, optimizing traffic, and identifying commuting patterns. We compare our proposed route similarity measure, C-SIM, to eight popular alternatives including Edit Distance on Real sequence (EDR) and Frechet distance. The proposed measure is simple to implement and we give a fast, linear time algorithm for the task. It works well under noise, changes in sampling rate, and point shifting. We demonstrate that by using the grid, a route similarity ranking can be computed in real-time on the Mopsi2014 1 route dataset, which consists of over 6,000 routes. This ranking is an extension of the most similar route search and contains an ordered list of all similar routes from the database. The real-time search is due to indexing the cell database and comes at the cost of spending 80% more memory space for the index. The methods are implemented inside the Mopsi 2 route module.