MAKING SENSE OF THE NOISE: INTEGRATING MULTIPLE ANALYSES FOR STOP AND TRIP CLASSIFICATION

Abstract

Mobility research is mainly concerned with understanding mobility on a higher level, including environmental factors, e.g., measuring the time out of home or tracking revisited places. This requires preprocessing the raw data obtained from GPS sensors, like clustering significant locations and distinguishing these from periods on the go. We introduce a new stop and trip detection algorithm to transform a list of position records into intervals of dwelling and transit. The system is based on geometrical analyses of the signal noise: Imperfect GPS data tends to scatter around an actual dwell position in a star-like pattern, and this imperfection is what we leverage for our classification. The system contains four independent classification methods, comparing different aspects of the geometrical properties of a given trajectory. If available, accelerometer readings can be used to improve the system’s accuracy further. To evaluate the classifier’s performance, we recorded a large dataset containing gold-standard labels and compared the classification results of our system with the results of Scikit Mobility and Moving Pandas. Our Stop Go Classifier outperforms the traditional distance/time-threshold-based systems. The described system is available as free software.