Mining large-scale, sparse GPS traces for map inference

Abstract

We address the problem of inferring road maps from largescale GPS traces that have relatively low resolution and sampling frequency. Unlike past published work that requires high-resolution traces with dense sampling, we focus on situations with coarse granularity data, such as that obtained from thousands of taxis in Shanghai, which transmit their location as seldom as once per minute. Such data sources can be made available inexpensively as byproducts of existing processes, rather than having to drive every road with high-quality GPS instrumentation just for map buildingand having to re-drive roads for periodic updates. Although the challenges in using opportunistic probe data are significant, successful mining algorithms could potentially enable the creation of continuously updated maps at very low cost.In this paper, we compare representative algorithms from two approaches: working with individual reported locations vs. segments between consecutive locations. We assess their trade-offs and effectiveness in both qualitative and quantitative comparisons for regions of Shanghai and Chicago.

An Automatic Road Network Construction Method Using Massive GPS Trajectory Data

“…A future research topic could be the further smoothing of the line segments. Liu et al [21] use the F-score to quantitatively measure the performance of a method inspired by information retrieval evaluation. In their method, they first calculate the precision and recall of each inferred road where recall = cv Ground and precision = cv M , in which • denotes the length of the road, M is the extracted map, Ground is the ground-truth map, and cv = M ∩ Ground denotes the matched parts.…”

“…2017, 6, 400 3 of 15 for generating road networks have been proposed in recent years [16,17]. In general, these methods can be organized into three categories [18]: (1) point clustering [19][20][21], which assumes that the input raw data consist of a set of points that are then clustered in various ways (such as by the k-means algorithm) to obtain street segments that are finally connected to form a road network; (2) incremental track insertion [10,[22][23][24][25][26], which constructs a road network by incrementally inserting trajectory data into an initially empty graph; and (3) intersection linking [27][28][29], in which the intersection vertices of the road network are first detected and then linked together by recognizing suitable road segments. Some of the representative algorithms of each category are listed in Table 1.…”

“…The algorithm developed by Davies et al [20] was found to significantly outperform other algorithms under a variety of conditions. In recent years, directed Hausdorff, path-based, shortest-path-based, and graph-sampling-based distance measures have been adopted to evaluate the quality of the topological data and other features of constructed road networks [18,21]. [29] In this paper, a digital map construction method using vehicle GPS trajectories is proposed.…”

Map inference in the face of noise and disparity

“…We use two different evaluation methods from previous work to compare our inferred maps to the ground truth map. The first method (GEO) is taken from [15], and evaluates map geometry only. Here, the connectivity of the map is ignored entirely, but every segment of both maps is taken into account.…”

Crowdsensing Maps of On-street Parking Spaces

“…from Google Maps [6]) can be a very long and tiresome process. Several projects were done on successfully inferring road maps from GPS traces using data mining algorithms [7], [8] but none of them address the question of inferring parking maps. On the other hand, the existence of these maps can be very useful in several applications.…”