Show simple item record

Pedestrian Mobility Mining with Movement Patterns

dc.contributor.advisorWrobel, Stefan
dc.contributor.authorLiebig, Thomas
dc.description.abstractIn street-based mobility mining, pedestrian volume estimation receives increasing attention, as it provides important applications such as billboard evaluation, attraction ranking and emergency support systems. In practice, empirical measurements are sparse due to budget limitations and constrained mounting options. Therefore, estimation of pedestrian quantity is required to perform pedestrian mobility analysis at unobserved locations. Accurate pedestrian mobility analysis is difficult to achieve due to the non-random path selection of individual pedestrians (resulting from motivated movement behaviour), causing the pedestrian volumes to distribute non-uniformly among the traffic network. Existing approaches (pedestrian simulations and data mining methods) are hard to adjust to sensor measurements or require more expensive input data (e.g. high fidelity floor plans or total number of pedestrians in the site) and are thus unfeasible.
In order to achieve a mobility model that encodes pedestrian volumes accurately, we propose two methods under the regression framework which overcome the limitations of existing methods. Namely, these two methods incorporate not just topological information and episodic sensor readings, but also prior knowledge on movement preferences and movement patterns.
The first one is based on Least Squares Regression (LSR). The advantage of this method is the easy inclusion of route choice heuristics and robustness towards contradicting measurements. The second method is Gaussian Process Regression (GPR). The advantages of this method are the possibilities to include expert knowledge on pedestrian movement and to estimate the uncertainty in predicting the unknown frequencies. Furthermore the kernel matrix of the pedestrian frequencies returned by the method supports sensor placement decisions. Major benefits of the regression approach are (1) seamless integration of expert data and (2) simple reproduction of sensor measurements. Further advantages are (3) invariance of the results against traffic network homeomorphism and (4) the computational complexity depends not on the number of modeled pedestrians but on the traffic network complexity.
We compare our novel approaches to state-of-the-art pedestrian simulation (Generalized Centrifugal Force Model) as well as existing Data Mining methods for traffic volume estimation (Spatial k-Nearest Neighbour) and commonly used graph kernels for the Gaussian Process Regression (Squared Exponential, Regularized Laplacian and Diffusion Kernel) in terms of prediction performance (measured with mean absolute error). Our methods showed significantly lower error rates.
Since pattern knowledge is not easy to obtain, we present algorithms for pattern acquisition and analysis from Episodic Movement Data. The proposed analysis of Episodic Movement Data involve spatio-temporal aggregation of visits and flows, cluster analyses and dependency models.
For pedestrian mobility data collection we further developed and successfully applied the recently evolved Bluetooth tracking technology. The introduced methods are combined to a system for pedestrian mobility analysis which comprises three layers. The Sensor Layer (1) monitors geo-coded sensor recordings on people’s presence and hands this episodic movement data in as input to the next layer. By use of standardized Open Geographic Consortium (OGC) compliant interfaces for data collection, we support seamless integration of various sensor technologies depending on the application requirements. The Query Layer (2) interacts with the user, who could ask for analyses within a given region and a certain time interval. Results are returned to the user in OGC conform Geography Markup Language (GML) format. The user query triggers the (3) Analysis Layer which utilizes the mobility model for pedestrian volume estimation.
The proposed approach is promising for location performance evaluation and attractor identification. Thus, it was successfully applied to numerous industrial applications: Zurich central train station, the zoo of Duisburg (Germany) and a football stadium (Stade des Costières Nîmes, France).
dc.rightsIn Copyright
dc.subject.ddc004 Informatik
dc.titlePedestrian Mobility Mining with Movement Patterns
dc.typeDissertation oder Habilitation
dc.publisher.nameUniversitäts- und Landesbibliothek Bonn
ulbbnediss.affiliation.nameRheinische Friedrich-Wilhelms-Universität Bonn
ulbbnediss.fakultaetMathematisch-Naturwissenschaftliche Fakultät
dc.contributor.coRefereeCremers, Armin B.

Files in this item


This item appears in the following Collection(s)

Show simple item record

The following license files are associated with this item: