Identifying Lane Changes Automatically using the GPS Sensor of Portable Devices

Abstract

Automatic lane change identification can be used in warning systems that alert the driver when drifting off-lane, or in lane-specific navigation systems. Furthermore, they may help road authorities plan better road network design and infrastructure. Though lane-departure warning systems are already commercially available, these systems rely on on-board vehicle sensors. The availability of such systems can be increased if it became available to owners of smartphones or navigation devices. In this thesis, I propose and test several methods that rely solely on GPS data originating from portable devices (two smartphones, a GPS-equipped GoPro Max, and a USB GPS receiver), recorded during test rides on a Dutch highway (42.8 km) with a total of 64 lane changes. The methods rely on observing changes in the lateral offset between the GPS trajectories and the road geometry. The resulting identification accuracy of the best performing algorithm was achieved using the GoPro Max with an overall F1 score (harmonic mean of precision and recall) of 0.9 on the validation data using the signal of filtered projected lateral distance. It is concluded that GPS-equipped portable devices could be a suitable choice for identifying lane changes, provided there is improvement in the quality of GPS receiver chips with higher data collection frequency such as the ones used in GoPro Max.