Accurate bus occupancy estimation for WLAN probing utilising probabilistic models

Abstract

This paper obtains an enhanced estimator of the number of people on buses using probabilistic models and wireless local area network (WLAN)-based device counts. The improved estimator exploits probabilistic models of false positives and false negatives of a previously presented baseline estimator. False positives of the baseline estimator result from devices on the roadside outside the bus that fail to be filtered out by the used threshold approaches. False negatives may be caused by low probe emission frequency, message losses due to collisions of the WLAN transmissions, medium access control (MAC) address randomisation, WLAN channel selection, and occur in particular when passengers stay on the bus only for short duration of time. The extensions of the baseline model by probabilistic models for false positives and false negatives enables to apply maximum likelihood estimation. Distribution parameters for false positives and false negatives are found from measurements. Field tests on buses in a Danish town are used to validate and quantify the gain of the enhanced estimator.