Traffic subflow estimation and bootstrap analysis from filtered counts

Traffic flow counts at different places are dependent due to vehicles travelling between the places but also due to other more slowly varying causes. Using a stochastic process approach, we define a kernel regression filter which extracts the high frequency part and effectively filters out the slow variations. From the filtered series, estimates of route selection probabilities and travelling time distributions are derived. The precision of estimates is studied by high level resampling (bootstrap) methods and they are also compared to a classical approach. The methods are applied to a rather extensive set of data.