Data-Driven Bicycle Network Analysis Based on Traditional Counting Methods and GPS Traces from Smartphone

Abstract

This research describes numerical methods to analyze the absolute transport demand of cyclists and then to quantify the road network weaknesses of a city with the aim to identify infrastructure improvements in favor of cyclists. The methods are based on a combination of bicycle counts and map-matched GPS traces and are demonstrated with the city of Bologna, Italy: the dataset is based on approximately 27,500 GPS traces from cyclists, recorded over a period of one month on a volunteer basis using a smartphone application. A first method estimates absolute, citywide bicycle flows, by scaling map-matched bicycle flows of the entire network to manual and instrumental bicycle counts of the main bikeways of the city. As there is a good correlation between the two sources of flow data, the absolute bike-flows on the entire network have been correctly estimated. A second method describes a novel link-deviation index, which quantifies for each network edge the total deviation generated for cyclists in terms of extra distances traveled with respect to the shortest possible route. The deviations are accepted by cyclists either to avoid unpleasant road attributes along the shortest route or to experience more favorable road attributes along the chosen route. The link deviation index indicates the planner which road links are contributing most to the total deviation of all cyclists -in this way, repelling and attracting road attributes for cyclists can be identified. This is why the deviation index is of practical help to prioritize bike infrastructure construction on individual road network links.

Estimation of link‐cost function for cyclists based on stochastic optimisation and GPS traces

“…For the following investigation, two different attribute vectors have been taken into consideration: the attribute vector with a constant exclusive bikeway and where the exclusive bikeway attribute is proportional to the link length , in accordance with other studies [4, 8, 27].…”

“…In the literature, many studies deal with the identification of road attributes that affect the cyclists’ path choice and different models have been calibrated to quantify their respective influence [2–10]. In particular, the studies in [3, 6–10] have compared the cyclist's route choice with the shortest route in order to find attributes that cause the users to deviate from the shortest path: for example, in [10] the shortest path has been used to split the global positioning system (GPS) recorded traces in a set of detour‐classes. Moreover, Rupi et al [8] calibrated a binominal logit‐model able to predict if a cyclist will choose either the shortest route or a non‐overlapping alternative, based on route‐links attributes.…”

“…A major problem of such route choice models is the determination of the choice set among which cyclists can select their routes – it is generally unknown which routes a cyclist considers as alternatives between a specific origin and destination, and in particular how users can detour from the shortest path. Empirical data on detours have been calculated in different ways, see for example Rupi et al [8], Park and Akar [10], Pritchard et al [11] and Griffin and Jiao [12]. For example, Park and Akar [10], in Ohio, found that most bicycle trips (91.1%) include a detour and that these are 13.5% longer on average than their shortest alternatives, with large variations.…”

“…For example, Park and Akar [10], in Ohio, found that most bicycle trips (91.1%) include a detour and that these are 13.5% longer on average than their shortest alternatives, with large variations. Moreover, Rupi et al [8], in Italy, found that detours are 20.7% longer on average with respect to the shortest path.…”

Development of a method for walking step observation based on large-scale GPS data

“…Previous studies have proposed map matching methods to estimate unknown trajectories on roads due to the accuracy of GPS logs, based on the geometric or topological similarity with the road network and the hidden Markov models [31]. Rupi et al [32] showed that cyclists' GPS traces matched to road networks dedicated to bicycle travel correlated highly with the cyclists' flows based on manual surveys. However, the map matching process often requires large-scale computational resources [33].…”