The Path Inference Filter: Model-Based Low-Latency Map Matching of Probe Vehicle Data

Abstract

We consider the problem of reconstructing vehicle trajectories from sparse sequences of GPS points, for which the sampling interval is between 10 seconds and 2 minutes. We introduce a new class of algorithms, called altogether path inference filter (PIF), that maps GPS data in real time, for a variety of trade-offs and scenarios, and with a high throughput.Numerous prior approaches in map-matching can be shown to be special cases of the path inference filter presented in this article. We present an efficient procedure for automatically training the filter on new data, with or without ground truth observations. The framework is evaluated on a large San Francisco taxi dataset and is shown to improve upon the current state of the art. This filter also provides insights about driving patterns of drivers. The path inference filter has been deployed at an industrial scale inside the Mobile Millennium traffic information system, and is used to map fleets of data in San Francisco, Sacramento, Stockholm and Porto.

A hybrid HMM model for travel path inference with sparse GPS samples

“…Global optimization algorithms (Horst and Tuy 2013) process all GPS samples together to infer the most likely travel path given all input GPS samples (Hunter et al 2014;Thiagarajan et al 2009), local optimization algorithms generally construct vehicle path step by step with best choice at each step and the constructed vehicle path might not be the most likely one (Greenfeld 2002;Yu et al 2006). Local optimization algorithms are in general computationally fast but their success for path inference shows a steep drop against a sparse number of GPS observations, especially in the case of incorrect alignments at the beginning of the local optimization algorithms.…”

“…Although historical average speed computed from data is commonly used in different traffic-related problems (Gonzalez et al 2007;Li and McDonald 2002;Work et al 2008), to the best of our knowledge, most of the path inference algorithms assume constant legal speed limits according to road types and transportation modes (Chen and Bierlaire 2015;Hunter et al 2014;Lou et al 2009;Yuan et al 2010). In this study, we use average speed computed from historical speed data of a given road.…”

“…Recent studies on map matching focus not only on localization of GPS samples on a map but also on inference of the travel path (Bierlaire et al 2013;Chen and Bierlaire 2015;Hunter et al 2014;Lou et al 2009;Miwa et al 2012;Thiagarajan et al 2009). Some of these studies focus on travel routes with sparse GPS samples (Chen and Bierlaire 2015;Hunter et al 2014;Lou et al 2009;Miwa et al 2012;Thiagarajan et al 2009).…”

“…Shortage in certain roads and low number of GPS samples for different parts of roads are challenges for algorithms that try to infer travel paths that drivers follow. The issue stands out more especially in urban areas, where roads can be relatively short and vehicles can travel on many different road segments in short time intervals (Hunter et al 2014).…”

“…In this study, we propose an effective algorithm that aims to reconstruct vehicles' travel routes even in GPS samples taken every 2 min, a duration considered to be a low sample rate in many studies (Chen and Bierlaire 2015;Hunter et al 2014;Lou et al 2009;Miwa et al 2012;Thiagarajan et al 2009). For instance, if a vehicle travels at a speed of 50 km/ h, then the vehicle can travel about 1666 m between each GPS sample.…”