Road traffic noise calculations require modeling the traffic flow in a road network. The reliability of these calculations can be improved with accurate estimation of the traffic flow, including estimation of its temporal variations. Low-cost noise sensors that run on single-board computers in a noise monitoring network are suitable candidates to simultaneously estimate the local temporal traffic flow from their pass-by measurements, using an on-board traffic flow estimator model. Aside from this model requiring to be computationally efficient, it should also be robust, e.g., invariant to sensor position relative to the source, weather conditions, etc. With noise measurements as an input, different noise features and prediction models are tested for vehicle detection. The accuracy of these models is evaluated using traffic count data obtained from dedicated vehicle-counting infrastructure at the locations of the noise sensors. The analysis is restricted to sparse traffic conditions in this initial study.