Sensing Quality and Estimation of Public Transport Occupancy During Live Operation

Public transport providers have an interest in obtaining bus occupancy information for optimizing services and doing so at a low cost. For that reason it is of interest to compare new low cost bus occupancy estimation methods with established approaches in terms of estimation quality and characterization of variability. The paper provides approaches to derive confidence intervals for two relevant estimation approaches, WLAN probes and light sensor based, and compares the two approaches based on data from a live deployment. The comparison shows that the light sensor based approach provides high accuracy but errors accumulate over time introducing a bias and uncertainty in the estimate. The WLAN based approach is capable of recovering after errors, but due to the variability introduced by the extrapolation from number of devices to number of people the approach is less accurate.