Lane Detection and Lane-Changing Identification with High-Resolution Data from a Swarm of Drones

Abstract

In the era of big data, new transportation-related concepts and methodologies need to be proposed to understand how congestion propagates. pNEUMA, a unique dataset that was acquired during a first-of-its-kind experiment using a swarm of drones over a dense city center, has uncovered new opportunities for revisiting and evaluating existing concepts, and new ways to describe significant traffic-related phenomena. This dataset is part of an open science initiative shared with the research community and consists of more than half a million detailed trajectories of almost every vehicle that was present in the study area. The aim of this paper is to describe the first methodological approach to how such information can be utilized to extract lane-specific information from this new kind of data and set the benchmark for possible future approaches. Specifically, we describe the methodological framework of two related algorithms: lane detection and lane-changing maneuver identification. Azimuth was the main concept utilized in this methodological framework to overcome existing issues in the literature related to identifying lane-changing maneuvers. The combination of high-quality data, clustering techniques, and detailed spatial information in the lane-detection algorithm indicated it was an effective tool without the need for complex computational effort. Moreover, high-resolution data together with modern time-series analysis tools for lane-changing identification, showed that high-accuracy predictive algorithms can be obtained. The accuracy of both tools was over 95%. Challenging scenarios are identified for future studies and to further improve the tools.