Modeling Visitor Movement in Theme Parks A scenario-specific human mobility model Gürkan Solmaz, Mustafa İlhan Akbaş and Damla Turgut Department of Electrical Engineering and Computer Science University of Central Florida - Orlando, FL Oct 23, 2012 G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 1 / 29
1 Introduction G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 2 / 29
1 Introduction 2 Problem definition G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 2 / 29
1 Introduction 2 Problem definition 3 Human mobility model Overview of the model Modeling a theme park Visitor model G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 2 / 29
1 Introduction 2 Problem definition 3 Human mobility model Overview of the model Modeling a theme park Visitor model 4 Simulation study G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 2 / 29
1 Introduction 2 Problem definition 3 Human mobility model Overview of the model Modeling a theme park Visitor model 4 Simulation study 5 Conclusion G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 2 / 29
Introduction Realistic modeling of the movement of people Simulation-based performance evaluation of networks Human mobility: the combination of regularity and spontaneity Human movement patterns depend on the application scenario Need for scenario-specific mobility models G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 3 / 29
Problem definition Problem Theme parks are large areas with one or more "themed" landmarks that consist of attractions Visitors plan to see a subset of these attractions by walking Visitors usually pre-plan their visit Visitors minimize the time it takes to walk Changing decisions spontaneously depending on various factors Objective Realistic modeling and simulation of the visitor movement in theme parks G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 4 / 29
Overview of the model SLAW model Provides an effective strategy in representing social contexts of common gathering places of walking people Fractal points and heavy-tail flights on top of these fractal points Our model Using SLAW as a baseline Application of queuing models to represent the behavior and effects of attractions Preserving the nondeterminism G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 5 / 29
Modeling a theme park The five main phases Fractal points Clusters Attractions and noise points Landmarks Theme park map G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 6 / 29
Fractal points Defined in SLAW People are more attracted to visit popular places Self-similarity and least action characteristics by fractal points A fractal point can be considered as a waypoint G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 7 / 29
Fractal points G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 8 / 29
Clusters The goal: finding the areas where people are more attracted to gather together Determining the parts of the area with highest number of fractal points Using a modified version of DBScan algorithm to specify the number of clusters and noise point ratio G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 9 / 29
Clusters G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 10 / 29
Attractions and noise points The most dense areas found by clustering are marked as "attractions" The weight of an attraction is defined according to the number of fractal points included The central location of an attraction is determined by averaging the coordinate values Nonclustered fractal points are marked as "noise points" The attractions and noise points are defined as the waiting points in a landmark G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 11 / 29
Attractions and noise points Our model represents attractions by queuing models Attraction Queue model Percentage Main rides (RD) M/D/n 17% Medium-size rides M/D/n 56% (M-RD) Restaurants (RT) M/M/1 17% Live shows (LS) M/M/n 10% G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 12 / 29
Landmarks A landmark includes multiple attractions and noise points G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 13 / 29
Theme park map For modeling the theme park, we use a graph theoretical approach Graph consists of vertices and weighted non-directional edges Each vertex in the graph represents a landmark Each edge represents a path between two landmarks Weight of an edge corresponds to the transportation time between the landmarks G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 14 / 29
Theme park map G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 15 / 29
Visitor model The visitors are represented by mobile nodes The states of the mobile nodes: "initial", "inqueue", "moving", "innoisepoint" and "removed" Each visitor Initially decides a hangout time for the particular landmark Selects a subset from the set of all attractions to visit Leaves after the hangout time if its state is not "inqueue" G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 16 / 29
Trip planning The visitors move according to the least action principle If a visitor visits an attraction or a noise point, marks it as visited and does not visit it again A visitor by-passes an attraction if the queue is full Visitors decide their next destinations by the modified version of LATP algorithm G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 17 / 29
Trip planning A visitor tries to minimize the Euclidean distance for travel Every unvisited point has a probability to be the next destination Instead of using identical waiting points, weighted waiting points are used Attractions with larger weight values have more probability When a visitor goes to a noise point, waits in the exact position When an attraction is selected, the visitor goes to a random sitpoint G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 18 / 29
Visitor model Each attraction has a service rate, number of visitors per service and capacity Waiting time of a visitor in an attraction depends on The number of visitors already waiting in the queue Service rate of the attraction Number of visitors per service Waiting time of the visitor in a noise point is generated randomly by truncated Pareto distribution G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 19 / 29
Mobility model demo G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 20 / 29
Simulation parameters Simulation time 10 hours Terrain size 1000x1000 m Number of visitors 2000 Hangout times 2-10 hours Number of attractions 15 Noise point ratio 10% Minimum waiting time 30 sec Visitor speed 1 m/sec Pareto α value 1.5 Number of visitors per 40 service (RD) Number of visitors per 20 service (M-RD, LS) G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 21 / 29
Simulation environment G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 22 / 29
Simulation metrics and results The simulation of our model generates synthetic mobility traces of mobile nodes The results are analyzed by GPS traces of theme park visitors SLAW and RWP mobility model simulations We examine fundamental characteristics of mobility features: Distribution of flight lengths Distribution of waiting times (pause times) Waiting rate of visitors G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 23 / 29
Simulation results Normalized flight length distributions from mobility simulation Our simulation have consistency, among different runs of the simulation Number of flights 250 200 150 100 Experiment 1 Experiment 2 Experiment 3 Experiment 4 Experiment 5 50 0 100 200 300 400 500 600 700 Flight lengths (m) G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 24 / 29
Simulation results Normalized flight length distributions from the mobility simulation, Orlando Disney World GPS traces, SLAW and RWP Mobility Models The mobility model outperforms other two synthetic mobility models Number of flights 60 50 40 30 20 Simulation GPS traces RWP SLAW 10 0 100 200 300 400 500 600 700 800 900 1000 Flight lengths (m) G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 25 / 29
Simulation results The square area for Magic Kingdom is modeled as a landmark with dimensions approximately equal to 850x850 meters using OpenStreetMap Our mobility model performs significantly better than SLAW The flight length distribution of our model is very close to the flight length distribution of GPS traces Number of flights 60 50 40 30 20 10 0 Simulation GPS traces SLAW 100 200 300 400 500 600 700 Flight lengths (m) G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 26 / 29
Simulation results Average number of waiting points per one hour from the simulation, Orlando Disney World GPS traces, SLAW and Random Waypoint Mobility Model Every visitor roughly waits in 10 different locations in an hour on average Average number of waiting points 20 18 16 14 12 10 8 6 4 Simulation GPS Traces RWP SLAW 2 1 1.5 2 2.5 3 3.5 4 4.5 5 Experiments G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 27 / 29
Simulation results Normalized waiting time distributions from the simulation, Orlando Disney World GPS traces and SLAW RWP model is not used since the mobile nodes have constant waiting times The results are closer to the GPS traces, compared to SLAW Number of waiting times 600 500 400 300 200 100 Simulation GPS traces SLAW 0 100 200 300 400 500 600 700 800 900 1000 Waiting times (sec) G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 28 / 29
Conclusion A realistic model of the movement of visitors in a theme park Accuracy of our model is validated through simulations Future steps: Including micro-mobility behaviors in the model Using the mobility model to evaluate applications of wireless sensor networks Event coverage in theme parks using wireless sensor networks with mobile sinks (submitted to IEEE ICC 2013) G. Solmaz, M. İ. Akbaş, D. Turgut (UCF) IEEE LCN 2012 Oct 23, 2012 29 / 29