Exploration of ground truth from raw GPS data

Abstract

To enable smart transportation, a large volume of vehicular GPS trajectory data has been collected in the metropolitanscale Shanghai Grid project. The collected raw GPS data, however, suffers from various errors. Thus, it is inappropriate to use the raw GPS dataset directly for many potential smart transportation applications. Map matching, a process to align the raw GPS data onto the corresponding road network, is a commonly used technique to calibrate the raw GPS data. In practice, however, there is no ground truth data to validate the calibrated GPS data. It is necessary and desirable to have ground truth data to evaluate the effectiveness of various map matching algorithms, especially in complex environments. In this paper, we propose truthFinder, an interactive map matching system for ground truth data exploration. It incorporates traditional map matching algorithms and human intelligence in a unified manner. The accuracy of truthFinder is guaranteed by the observation that a vehicular trajectory can be correctly identified by human-labeling with the help of a period of historical GPS dataset. To the best of our knowledge, truthFinder is the first interactive map matching system trying to explore the ground truth from historical GPS trajectory data. To measure the cost of human interactions, we design a cost model that classifies and quantifies user operations. Having the guaranteed accuracy, truthFinder is evaluated in terms of operation cost. The results show that truthFinder makes the cost of map matching process up to two orders of magnitude less than the pure human-labeling approach.

Visual Traffic Jam Analysis Based on Trajectory Data

“…The map matching result is analysed by comparison to manually labelled data, with Mao et al's technique [32]. We randomly choose 500 trajectories from our dataset, containing 14,632 sampling points and match them automatically.…”

A force-directed approach for offline GPS trajectory map matching

“…The authors classify the analytical approaches used in the algorithms into 'geometrical', which use proximity-based methods, 'topological', which use the notions of connectivity between the links (one-way roads, connectivity and reachability information), 'probabilistic', which further use information about the quality or accuracy of the GPS signal (typically obtained from the GPS sensor), and 'advanced', which use more specific methods such as Kalman filters, hidden Markov chains, timing information (e.g., to predict the exiting from a tunnel) and other application-specific approximation techniques. Typically the underlying map network is known, however some researchers [4,15,19,34] have developed approximation techniques to generate an unknown underlying map or to perform map matching without reference to a known map topology by observing the clustering of trajectories.…”

Motorway Measurement Campaign to Support R&D Activities in the Field of Automated Driving Technologies

“…An in-depth treatise of RTK based ground truth generation for SLAM evaluation is given in [ 24 ], while taking uncertainties and their effect on the evaluation into account. A framework for cost-effective manual improvement of GPS ground truth quality is described by [ 25 ].…”