Elsevier Editorial System(tm) for Pattern Recognition Letters Manuscript Draft Manuscript Number: PRLETTERS-D-13-00229R1 Title: Mobility Analysis of the Aged Pedestrians by Experiment and Simulation Article Type: Special Issue: SIPRCA Keywords: Cellular Automata; pedestrian; crowd dynamics; aging society Corresponding Author: Dr. kenichiro shimura, Corresponding Author's Institution: The University of Tokyo First Author: Kenichiro Shimura Order of Authors: Kenichiro Shimura; Kazumichi Ohtsuka; Giuseppe Vizzari; Katsuhiro Nishinari; Stefania Bandini Confirmation of Authorship Pattern Recognition Letters Authorship Confirmation Please save a copy of this file, complete and upload as the “Confirmation of Authorship” file. As corresponding author I,_Kenichiro Shimura, hereby confirm on behalf of all authors that: 1. This manuscript, or a large part of it, has not been published, was not, and is not being submitted to any other journal. 2. If presented at or submitted to or published at a conference(s), the conference(s) is (are) identified and substantial justification for re-publication is presented below. A copy of conference paper(s) is(are) uploaded with the manuscript. 3. If the manuscript appears as a preprint anywhere on the web, e.g. arXiv, etc., it is identified below. The preprint should include a statement that the paper is under consideration at Pattern Recognition Letters. 4. All text and graphics, except for those marked with sources, are original works of the authors, and all necessary permissions for publication were secured prior to submission of the manuscript. 5. All authors each made a significant contribution to the research reported and have read and approved the submitted manuscript. Signature_____________________________________Date 22/Aug.2013 List any pre-prints: N/A Relevant Conference publication(s) (submitted, accepted, or published): Workshop on Pattern Recognition and Crowd Analysis (PRCA12) in conjunction with ICPR 2012, Tsukuba, Japan, 11 November 2012. Justification for re-publication: This paper is for the special issue associated with the conference stated above. *Revision note Dear Editors Regarding to the Pare # PRLETTERS-D-13-00229 Title: Mobility Analysis of the Aged Pedestrians by Experiment and Simulation Author: Kenichiro Shimura; Kazumichi Ohtsuka; Giuseppe Vizzari; Katsuhiro Nishinari; Stefania Bandini First of all, we thank you very much for processing our paper. The paper is revised based on the reviewers’ comments. Together with the revised manuscript, we attach the answer to each reviewer in the following pages. Best Regard Answers to Reviewer#1 Dear Reviewer#1 Thank you for your careful reading and for giving useful comments toward our manuscript. First of all, the typographical errors and inappropriate expressions are corrected. Also the main text and figures are revised according to the reviewer’s comments. The following are the points of revision. 1, Acceleration and breaking time of the pedestrians. We consider the pedestrian can make immediate start and stop unless he is running. This is a one of the differences between cars and pedestrians. In the experiment, at the starting moment we can observe a starting wave but this effect is due to the restriction of the mobility (e.g. if other pedestrians are in front, the one is not able to move.). of cause particularly looking at the starting wave in detail acceleration issue me become significant, but in this case it is negligible because we are interested in the behaviour in steady state. Thus acceleration is not considered in the model. 2, CA grid and the corresponding size and relationship to the floor tiles Floor tiles are used as reference for participants to maintain the constant speed wile experiment. Beside for the simulation, the grid size is considered as 40cm x 40cm grid is considered. 3, Computational burden and language. We use MATLAB for simulation. The computational time is mater of second for this calculation size. Moreover for reference, simple 2D pedestrian model with 256x256 cells with pedestrian density of 0.5 (32768 pedestrians), it takes about 50 seconds for 1 time step with graphical output using 3.80 GHz Corei7 without parallelization. 4, NaSch traffic model and intriguing points for transformation of normal to slow pedestrian. At first the basic concept of NaSch model is to implement acceleration and deceleration. This means that the cars (or in general namely “particles” in wide sense including pedestrians) can move more than 1 cells in 1 time step. Then the velocity of a particle can be expressed as “number of cells moved in 1 time step”. If applying NaSch model to Normal and Slow pedestrian case, by saying, e.g. Normal moves 5 cells at a step and slow moves 3 cells at a step. in case that the Normal "transformed" to Slow, then those transformed “particles” moves 3 cells lather than 5 cells after transformation. This is good way to implement heterogeneous velocity. Beside in our model, we have implemented the velocity by changing the transition probability. Thus in any case if the velocity of pedestrian changes, then it can be expressed by changing the transition probability of corresponding pedestrians. Answers to Reviewer#2 Dear Reviewer#2 Thank you for in-depth reading on our manuscript. We are greatly appreciated for detailed comments and suggestions. We have made major revision on the main text to add more details on the modelling part. And also the introduction part is enriched to make clear of the point for this paper. 1, The definition of the aging rate The aging rate is defined as ratio of elderly population to that of the younger populations. According to WHO’s definition, the “elderly” is defined as those who are over 65 years old while “younger” is those who are younger than that. 2, regarding to the ransition rules and Eq. (1) We are agree that the expression of those particles do not move for m_{i-1,j}^t and n_{i-1,j}^t are missing. However as it is suggested, Eq. 1 is replaced by graphical expression of the transition rules. The explanations of the rules are also added in the main text. And also some mistake in the expression on transition probability is corrected. i.e. for successful overtake; P_{O}( 1P_{N} ) -> P_{O} P_{N} which is a expression error and it does not affect to the results. 3 Update method. The update method is more appropriately called “particle oriented shuffled update” but with some modification on implementing the transition probabilities to avoid spatial deviation in between the pedestrian in the same group. Details are added to the main text. 4, expanding introduction and typographic errors The introduction is enriched according to the reviewer’s comments and suggested references are included. Explanation of the framework and the points of interests for this paper made clearer. And also grammatical, spelling and typographic errors are corrected. Answers to Reviewer#3 Dear Reviewer#3 Although the journal is focused on pattern recognition this paper is for the special issue including crowd analysis. Counting crowd and analysing the dynamics is greatly difficult. Thus we are trying to make a hybrid image analysis. This idea is to include prediction in to detection algorithm. Thus authors think this paper is in the scope of this special issue. APPENDIX: COMMENTS FORM THE REVIEWERS Reviewer #1: Reviewer #1: In this paper the authors present a Cellular Automaton (CA) model for the analysis of mobility of aged pedestrians. The paper beyond the CA model analysis is granted with a series of experimental studies regarding the pedestrian behavior of elderly people when mixed up with normal speed/younger persons. In every case, the simulation results are found in good agreement with the experimental ones and, consequently, the proposed CA model successfully represents pedestrian dynamics fundamental phenomena such as the pedestrian formation and travel time. This reviewer believes that the manuscript is technically sound and its style is rather clear; as a result its presentation leaves just a few more to be desired. First of all, from my point of view, the authors are kindly requested to enrich their Introduction Section but mostly their Modelling Section with adequate references so as to provide the reader with some fundamentals of Cellular Automata (CAs) computational tool as well as with some CA pedestrian dynamics models. In specific, based on their expertise, they could also refer to their previous works with CAs so as to provide the potential reader with the preliminary characteristics of the CAs model when applied to pedestrian dynamics. For example which is the Moore neighbourhood, how the CA rules apply to the cells and so on. In such a way, to the best of my knowledge, it would be easier for her/him to follow up the proposed model and moreover, to appreciate the presented approach. Moreover, some questions regarding the model presentation and experimental studies are arisen and should be answered. It is clear from the experimental studies presentation that no acceleration and breaking time of the pedestrians are taken into account. Does this mean that they are considered negligible or something else? Some more details regarding the initialization of the CA grid and the corresponding size of the CA cell should be also provided in accordance with the statement that ""normal speed" is defined as two floor tiles per foot". In the view of the foregoing, some info about the detection and tracking algorithm used for the presented experiments would be also valuable. Which is the computational burden of the provided model? Some info regarding the computational complexity of the proposed model as well as the programming language and the computing time could be possibly provided, if available. Furthermore, to the best of my knowledge, in correspondence to the NaSch traffic model, an intriguing addition to the transition rules of the presented CA model would be the following one: if for some reasons, normal speed persons are "transformed" to slow runners due to some personal reasons, i.e. speak with their mobile phone, or something else with some small probability then a new rule could be also applied. Of course, in such a case the characteristics of the under study person, namely his "hat" in the experimental studies, should be changed accordingly. Moreover, regarding the quality of the provided figures, I think that in a few cases, like the provided analysis of figures 6 and 7, could be further enhanced. Finally, as minor comments, a few grammatical, syntax and typographical errors exist and should be appropriately taken good care as follows. For example, across the main text, only right quotation marks are used in every case, like in lines 37 "Social Force model" and 46 "Floor Field Model" of page 5 of the submitted pdf file, in line 77 of page 6 "Walk as normal", etc. Moreover, some sentences need to be slightly rephrased like the ones found in line 182, in page 13 of the submitted pdf: "Thus in any case that the person "fight" for the same cell will be depends on the number of the persons who fight for the call and their corresponding transition parameters" and in line 221, in page 16: "Since the nature of the stochastic CA, number of simulation trials has carried out until 99% convergence of the probability densities." From my point of view, in Figure 3 caption, the "th" should be placed as superscript text just right next to the corresponding number. ##################### Reviewer #2: The authors have focused on the overtaking phenomenon in pedestrian dynamics. They performed experiment with real pedestrians and also developed a cellular automaton model. Their simulation reproduces the result of their experiment well. Up to the reviewer's knowledge, overtaking phenomenon has been seldom focused on in detail so far, although it is very essential phenomenon in the real world. In the paper, the result of three initial conditions, which are very interesting, are clearly compared. However, there are several unclear points for the reviewer; therefore, the review would like to suggest revising them and improving the quality of the paper. [Major Comments] p.2, l.15 The definition of the aging rate (maybe definition of the elderly person and young person) should be described. p.10, Eq. (1) The reviewer thinks that n_{i-1,j}^t in the first line of the Eq. (1) should be replaced by m_{i-1,j}^t. Furthermore, a term which represents the situation that the slow pedestrian at cell (i,j) cannot move forward: m_{i,j}^t * { (1-P_s) (1- n_{i+1,j}^t + m_{i+1,j}^t) + n_{i+1,j}^t + m_{i+1,j}^t} is missing. The authors can revise it; however, the reviewer would like to suggest replacing Eq. (1) by the description of the update rules because the CA model is used to perform simulation and no theoretical analysis is shown in the paper. If some figures are added with the explanation of the update rules, readability of the paper will be greatly improved. p.11, l.180, The combination of parallel update and random sequence update is not easy to understand. The authors should add some detailed explanation. [Minor Comments] p.3 The reviewer would like to recommend expanding introduction by citing some review articles and recent papers to explain the state of art more in detail and show the difference between the authors' study and existing study much more clearly. The following are some examples. [Review articles] D. Helbing, Rev. Mod. Phys. 73, 1067 (2001). T. Nagatani, Rep. Prog. Phys. 65, 1331 (2002). A. Schadschneider, D. Chowdhury, and K. Nishinari, Stochastic Transport in Complex Systems (Elsevier, Amsterdam, 2010). [Recent studies] Colin M. Henein and Tony White. Macroscopic effects of microscopic forces between agents in crowd models. Physica A, 373:694, 2007. Miho Asano, Takamasa Iryo, Masao Kuwahara, Transportation Research Part C, 18, (2010) 842-855 Mohcine Chraibi, Armin Seyfried, and Andreas Schadschneider, Phys. Rev. E, 82, 046111 (2010) Asja Jelic, Cecile Appert-Rolland, Samuel Lemercier, and Julien Pettre, Phys. Rev. E, 85, 036111 (2012) Takahiro Ezaki, Daichi Yanagisawa, and Katsuhiro Nishinari, Phys. Rev. E, 86, 026118 (2012) p.3, l.37, p.5, l.81 Closing double quotation mark should be replaced by opening one. p.3, l.57 walk -> walking p.4, l.63 Experimental -> Experiment p.4, l.70 Japans -> Japanese p.5, l.88 high -> highly? p.6, l.116 illustrate -> illustrates p.6, l.118, p.10, l.149, l.154, p.11, l.175, l.177 overtake -> overtaking p.6, l.123 corrosion -> collision? p.8, Figure 6, Figure 7 sped -> speed p.9, l.138 statistically -> stochastically? p.9, l.139, "NaSch" Abbreviation should be avoided. The following paper should be cited here. K. Nagel, M. Schreckenberg: J. Phys. I France 2, 2221 (1992) p.14, l.226 impassible -> impossible? p.14, l.232 in the situation of slow and normal pedestrians are randomly mixed as scenario 9, -> slow and normal pedestrians are randomly mixed in scenario 9, ? ##################### Reviewer #3: This paper is a study in the field of Simulation and it is not in the scope of this journal. *Highlights (for review) Click here to download Highlights (for review): Highlight.doc In coming decade, the aging rate will dramatically increase in advanced countries. Focus on the mobility issues where the elderly and young pedestrians are mixed. Cellular Automata model is created with aid of experiments. The model shows reasonable consistency with the experimental result. The compatibility of elderly's safety and young's mobility is discussed. *Manuscript [Word or (La)TeX] Click here to download Manuscript [Word or (La)TeX]: paper_Revised.docx Mobility Analysis of the Aged Pedestrians by Experiment and Simulation 1 2 Kenichiro Shimuraa,b,*, Kazumichi Ohtsukab, Giuseppe Vizzaria, Katsuhiro Nishinaria, Stefania Bandinib 3 4 5 6 7 8 9 Click here to view linked References a Department of Informatics Systems and Communication, The University of Milano Bicocca, Viale Sarca 336 - U14, 20126 Milano, ITALY. b Research Center for Advanced Science and Technology, The University of Tokyo,4-6-1, Komaba, Meguro-ku, Tokyo, 153-8904, Japan. Abstract 10 The relative weight of the population shifts from younger to elderly in the most 11 of the region on the planet. Current aging rate in the advanced nations varies 12 from 12% to 13% and is expected to increase up to 21% to 37% in 2050. The 13 increase of aging rate in the society especially in the large city will lead a 14 mobility problem. From a social quality point of view, it is important to achieve 15 the compatibility between safety and mobility respectively for younger and 16 elderly generation. For the purpose of understanding the basic characteristics of 17 the pedestrian dynamics under cohabitation of younger and elderly generation, a 18 Cellular Automata (CA) model is created with the aid of pedestrian experiments. 19 Simulations are carried out to reproduce the experimental results and had shown 20 a good agreement. 21 22 Keywords: aging society, pedestrian analysis, cellular automata 23 2010 MSC: 68Q80,68U20 *Corresponding author Email address: sh i mu ra@ to ka i .t.u -to kyo .a c .j p (Kenichiro Shimura) Preprint submitted to Pattern Recognition Letters 24 1. Introduction 25 The world is aging. The aging rate on the planet is in the increasing trend as 26 a result of decreasing birth rate and increasing life expectancy in advanced 27 nations. The aging rate is defined as the ratio between the elderly populations 28 over 65 years old against of the younger populations and is shifting toward the 29 elderly side. According to WHO report, current aging rate in the advanced 30 nations varies from 12% to 13% and is expected to increase up to 21% to 37% in 31 2050. The rapid change in the aging rate is forcing society change into a more 32 elderly oriented. As well as creating specific welfare services, transportations, 33 impediments removal, the safety issues of elderly generation in public areas 34 would appear to be an important topic. Elderly generations often face to 35 progressive deterioration of physiological and psychological functions which 36 causes slowdown in cognition, reaction and action speed. And health 37 impediments are attributable compare to the younger generations and such health 38 impediments become more frequent as age increases. Such physiological and 39 psychological deterioration affect in the walking speed, endurances and 40 sensitivities. According the review report on behavior and characteristics of older 41 pedestrians by the Department of Transport in U.K. (Dunbar, 2004), the average 42 walking speed of the elderly is about 75% of the younger generations. The 43 increase of the elderly people in the society makes a large change in the social 44 mobility. Considering in the public place, when the elderly people and young 45 people shares the same space, it is foreseen that there would be a major change 46 in the macroscopic pedestrian dynamics. This study focuses on such mobility 2 47 issues in the public space where the pedestrians with different characteristics are 48 mixed. Extensive studies on pedestrian modeling have been made in recent years 49 and they are classified into two main streams such as continuum and discrete 50 model. The “Social Force model” is the one of the successful approaches in the 51 continuum system introduced by Helbing and Molnar (1995). The model 52 describes the pedestrian’s velocity in terms of the collision avoidance mechanism 53 by considering the repulsive force between each pedestrian and is well studied in 54 Helbing et al. (2000) and Helbing (2001). One of the important factors in this 55 approach is the expression of the interaction force where Chraibi and Seyfried 56 (2010) introduces generalized centrifugal-force model to satisfy the collision 57 avoidance by means of excluded volume effect. 58 On the other hand, Cellular Automata (CA) is defined in a discrete lattice 59 and time. CA is defined by update rules such that the state of the current cell 60 changes according to the states of surrounding cells in discrete time step. For 61 pedestrian simulation, the state of a cell is either 0 or 1 to express the existence 62 of a pedestrian in a regular spatial grid. Then the dynamics is defined by 63 interaction between a cell of interest and the neighborhood cells. There are two 64 types of neighborhood selection namely as “Von Neumann neighborhood” and 65 “Moore neighborhood”. The former considers the surrounding four cells 66 orthogonal to the current cell while the latter considers all of surrounding eight 67 cells. The application of CA to hydrodynamics referred to as Lattice Gas 68 Automaton (LGA) is discussed in Wolfram (1994) and then the extended study 69 for pedestrian simulation is implemented by Helbing (2003). Although LGA 3 70 model is based on the random walk, in order to give a characteristic behavior, 71 biased random walk is implemented by Nagatani (2002). Further, based on 72 Nagatani’s model, Jiang (2006) performed a simulation on pedestrian interaction 73 between the large object. Besides the application of hydrodynamics model to 74 pedestrian simulation, Derrida at al. (1993) made a theoretical study on one 75 dimensional Asymmetical Simple Exclusion Process (ASEP) which plays a 76 fundamental role in traffic models. ASEP is a simple binary CA in open 77 boundary such that a particle moves one cell forward if the cell in front is 78 unoccupied which is relating to the Rule 184 in elementary CA studied by 79 Wolfram (1994). In this way, the excluded volume effect is implemented in 80 relatively simple manner. Nagel (1992) introduced a model which allows a 81 particle to move more than one cell at a time often referred to as 82 Nagel-Schreckenberg model (NaSch model). Thus, the homogeneity of the 83 velocity in ASEP is solved to demonstrate the acceleration and deceleration of 84 particles in discrete expression. Although NaSch model is designed to model the 85 traffic on the highway, Kirchner et al. (2004) applied this model to two 86 dimensional pedestrian interactions such as lane change and bottleneck in egress 87 behavior. For more general applications to give characteristic behavior to 88 pedestrian simulation, Floor Field (FF) Model is introduced and applied to 89 analyze variety of pedestrian phenomena (Burstedde et al. 2001; Kirchner and 90 Schadschneider 2002; Kirchner et al. 2003). FF is predefined geometrical 91 information statically or dynamically given to each cell to give the behavioral 92 characteristics. FF is widely applied for various applications such as egress and 4 93 counter flow behaviors. Henein (2007) uses FF for agent based egress model 94 while Yanagisawa (2007) made a theoretical study on CA model for pedestrian 95 behavior at the exit of the room during egress behavior. Eezaki (2012) further 96 extended the study for multiple bottleneck case and shown the existence of 97 symmetry braking on the pedestrian flow. Based on these studies, the general 98 issue is to determine the volume exclusion for each pedestrian for improving the 99 model to fit the realistic situations. While these referred studies consider the 100 excluded volume in the same size as a single pedestrian, more complex 101 interaction occurs when number of pedestrians is increased and the effects of the 102 volume exclusion become significant. For real world applications, personal space 103 and headway distance should also be considered. These extra spaces can be 104 considered as part of the excluded volume. Jelić et al. (2012) performed 105 experiments to show the spatial effect of headway distance greatly affects the 106 pedestrian velocity. In addition of such spatial effect, negotiation process also 107 occurs at pedestrian interaction e.g. give way to another pedestrian to avoid 108 collision. Asano et al. (2010) studied tactical model by implementing pedestrian 109 eyesight and game theory. Among numerous studies on pedestrian modeling, 110 those which deal with heterogeneity of the walking speed and overtaking 111 behavior are rare. Firstly, for different walking speed, other than NaSch model, a 112 method of applying the different update intervals defined by the speed ratio is 113 studied by Weng (2006). But the application of the models is limited to specific 114 variations of speed in the system. When there is wide variation in walking speed, 115 more flexible expression is necessary. Thus we have chosen to implement 5 116 walking speed by values of transition probabilities. On the other hand, the 117 conventional implementation of overtaking behavior in CA is often carried out 118 by as a simple lane change such that, a particle directly moves to the adjacent 119 cells which are perpendicular to the walking direction if the cell in front is 120 occupied. The description of this particle motion in real pedestrian follows that, 121 the walking pedestrian quickly slides into a position where directly left or right 122 of his current position if there is someone in front. But the fact is that, the 123 pedestrian has a momentum thus he moves into a position where diagonally in 124 front. One of the aims for this paper is to clarify the necessary mobility for this 125 movement. Throughout this paper, the study is carried out through experiments, 126 modeling and simulations to supports the consistency of the model. 127 128 2. Experiment 129 In order to obtain the basic characteristics of pedestrian's motion when two 130 different speed walkers are cohabitated, we have performed series of the 131 following experiments. In these experiments, we are especially interested in the 132 overtaking phenomenon and the emergent formation after overtaking. Fig. 1 133 illustrates the experimental arrangements. The experiment is held in Research 134 Center for Advanced Science and Technology (RCAST), The University of 135 Tokyo. First we prepare 25 healthy young Japanese persons (no distinguish 136 between male and female) and make them aligned in 5x5 matrix formation, 137 asking them to start walking unidirectional to the right direction at once with the 138 starting command. The total length of the experimental lane is 17m and an 6 139 experimental run ends when everyone crosses the goal line. Three video cameras 140 are set at the starting, middle and the goal, where the positions referred to as 141 Video 1, 2 and 3 in Fig. 1 respectively. The control of walking speed is by 142 informing them as “Walk as normal” and “Walk with 75% speed of the normal”. 143 Although male and female has different walk speed due to their physical 144 characteristics, for the purpose of obtaining the quantitative data, the pedestrians 145 are informed to walk along with the floor tiles where the "normal speed" is 146 defined as two floor tiles per foot step thus the participants can maintain the 147 constant speed. Moreover the floor tiles are used to obtain the quantitative data 148 for the post-processing of the videos. The normal walkers ware white cap and the 149 slow walkers ware red cap. The initial formations of the pedestrians are shown in 150 Fig. 2. It is trivial that there are no overtaking action and interaction between the 151 slow and normal speed walkers if the slow walkers are arranged behind the 152 normal walkers. Thus we made three initial configurations for of slow walkers at 153 the front part with various formations. Fig. 2(a) and (b) are the low-density 154 arrangement. Fig. 2(c) illustrates the relatively high-density arrangement where 155 the slow walkers act as a bottleneck to the normal walkers behind. During the 156 experiments, the normal walkers overtake the slow walkers so we can observe 157 the overtaking phenomenon. Two runs are taken for each experimental condition. 158 Fig. 3, 4 and 5 illustrate the captured image with corresponding pedestrian 159 formation of the video for experiment 1, 2 and 3 respectively. The frame number 160 superimposed in the picture shows the elapsed video frames from the start. The 161 frame rate is 29.97 frames per seconds (fps). Thus the time step for each 7 162 consecutive frame is 0.033 sec. The pedestrian formation of normal walkers is 163 observed from Fig. 3, 4 and 5. For the experiment 1, the formation after over 164 taking as seen from Fig. 3 is that, three pedestrians with a space in between 165 aligned vertically. Although some jitters in their positions are seen but the 166 alignment remains for alternative columns. For the experiment 2, it is seen from 167 Fig. 4 that three pedestrians are vertically aligned with no space in between for 168 first several columns. And thereafter, the number of the pedestrian in the vertical 169 alignment reduces to two with more space in between. As shown in Fig. 2, the 170 initial space in between the slow walkers is set just enough for three pedestrians 171 to pass through. This initial arrangement causes the vertical alignment of three 172 pedestrians in the front part of the line as seen in experiment 1 and 2. On the 173 other hand, for experiment 3, the space for the normal walkers to overtake the 174 slow ones is as narrow as for one single person to pass through. Since then, the 175 formation after overtaking appears as a single horizontal line. For all the cases, 176 first several columns after overtaking have the same structure as the first front 177 column of the initial formation. This is because of there is less interaction 178 between each of the normal pedestrians in the front regions. Then the initial 179 formation breaks as more pedestrians pass through the slow ones because there 180 are more interactions between normal walkers before overtaking the slow 181 pedestrians. Furthermore we focused on the overtaking phenomenon. Fig. 6 182 illustrates the steps of a normal walker overtakes a slow walker. It is seen from 183 Fig. 6 that the normal walker tries overtaking when a slow walker is just in front 184 of him. The actual overtake action is taken when consecutive spaces is available 8 185 at perpendicularly beside and the diagonally in front. Similarly to this, the 186 normal speed walkers also have possibility to overtake other normal speed 187 walkers as seen from Fig. 7. The necessity of the consecutive space for 188 overtaking can be considered as result of securing enough mobility to avid 189 collision while lane-change. Such behavior is often seen in the real life at the 190 pedestrian crossing, sidewalk, stations, etc. 191 192 Figure 1: The experiment arrangement. 193 194 195 196 Figure 2: The initial formations for high and low density slow walkers at the front respectively for experiment (a), (b) and (c). 197 198 199 200 201 Figure 3: The video capture and the corresponding pedestrian formation of the experiment 1 at 55th and 96th frame after the start which shows the emergent formation. The elapsed time between the two frames is 1.37 sec. 9 202 203 204 Figure 4: The video capture and the corresponding pedestrian formation of the experiment 2. Where the frame 139 shows the emergent formation. 205 206 207 208 Figure 5: The video capture and the corresponding pedestrian formation of the experiment 3. Where the frame 153 shows the emergent formation. 209 210 211 Figure 6: Normal sped pedestrian overtaking the slow speed pedestrian. 212 213 214 Figure 7: Normal sped pedestrian overtaking the normal speed pedestrian. 10 215 3. Modeling 216 A CA based pedestrian model is constructed based on the experimental 217 observations. This approach is characterized by discrete time and space. A 218 pedestrian is considered as particles move in a given lattice so called cellular 219 space. The motion of a particle is defined by certain transition rules which 220 describe the local interaction of cells. The state of the cell of interest is updated 221 according to the states of surrounding cells by some probability so called 222 transition probability. In the model, each cell has occupied and unoccupied states 223 where the maximum occupation number is one. The exclusivity rules such that 224 only one particle can occupy one cell must be strictly applied. For each update 225 step, the whole lattice is scanned and the transition rules are locally applied with 226 the corresponding transition probability. The major effect of introducing the 227 transition probability here is that to represent the different walking speed of the 228 pedestrians as well as for the frequency of overtaking. Then the walking speed of 229 the pedestrians are stochastically expressed, while NaSch traffic model expresses 230 deterministically by controlling the number of cells to be moved in each 231 calculation step. However, the implementation of the walking speed into the 232 model is achieved by the relative speed concept. Considering to the speed of the 233 normal pedestrians is unity then the speed of the slow walkers is expressed as its 234 ratio. We consider a 3x3 Moore neighborhood where a cell is updated according 235 to the states of all eight surrounding cells. Fig. 8 shows the transition rules of the 236 model. The figure represents the possibilities of a particle moving into the center 237 cell referred to as i, j with corresponding probabilities. Where PN and PS 11 238 represent the probability that Normal and Slow speed pedestrian respectively, 239 move one cell forward. Further, we introduce the probability PO for overtaking 240 which represent the decision of the pedestrian either he or she overtakes others 241 and it can be evaluated by counting the ratio of pedestrians proceed overtaking 242 against whole pedestrians. In this case, for the overtaking probability PO, only 243 normal pedestrians try overtaking others when the site in front is occupied. Then 244 the successful overtaking is expressed by probability product of POPN. Together 245 with these transition probabilities, the geometrical condition needs to be satisfied 246 as shown in the figure. 247 248 1 2 PoPN PoPN i i,j j PN (a) 1 2 PoPN (b) (c) PoPN (d) = Vacant Site = Normal Speed = Slow Speed PS Stop 249 250 (e) (f) = or = or Stop (g) Figure 8: The transition rules and the corresponding transition probabilities. 251 252 253 12 254 Fig. 8(a) and (f) shows the basic rule of Normal and Slow pedestrian move 255 forward and Fig. 8(e) and (g) shows those pedestrians don’t move due to the 256 destination cell is occupied. These four rules represent the unidirectional (right to 257 left) motion of normal and slow pedestrians which follows to the single lane 258 ASEP. For the overtaking process, the rules are chosen to conserve the 259 consistency between the experimental results. Fig. 8(b), (c) and (d) show the 260 rules for overtaking behavior where the normal speed pedestrian need to have 261 consecutive space at beside and diagonally in front as referred to Fig. 6 and 7. 262 Fig. 8(b) and (c) is symmetrical arrangement that only one pedestrian tries 263 moving into the center cell. Fig. 8(d) is the situation that two pedestrians fight 264 for center cell. We are not considering any negotiation in this case thus the 265 probability for successful overtake for the one is halved. In this way, the 266 transition rule expresses the following three motions such that moving forward, 267 the volume exclusion and the overtaking. 268 The time evolution of the lattice is calculated by particle oriented shuffled 269 update method. Every cell in the lattice is updated once in every calculation step 270 but the sequence of cell selection is by shuffled order. The Normal pedestrian 271 cells are updated at every iteration because PN = 1, but Slow pedestrian cells are 272 only updated by some chance according to the value of PS. The pseudo code of 273 the update algorithm is as follows; 274 275 276 277 278 WHILE simulation IF random number <= PS THEN TRANSIT Normal and Slow pedestrian, using local values of PN = 1, PS = 1, PO = PO ELSEIF PS < random number <= PN THEN 13 279 280 281 282 283 TRANSIT only Normal pedestrian, using local values of PN = 1, PO = PO ELSE TRANSIT none ENDIF ENDWHILE 284 285 Where the function of “TRANSIT” means that letting all the relevant transition 286 rules are applied with locally specified probabilities. The reason for using 287 shuffled update is to avoid the pseudo dynamics which occurs when sequential 288 update is used. And also the site oriented procedure prevents the time dependent 289 spatial dispersion of the particles, those who have the same transition provability 290 less than unity. In the case if there are two particles with PS in one dimensional 291 lattice, the deviation of the distance between these two particles becomes larger 292 as time evolves due to the stochastic nature. This causes the breakdown on the 293 initial formation of Normal and Slow pedestrian. From the observation of the 294 experimental result, the initial formation of Slow pedestrian is maintained for 295 while after starting. We are specially looking at the interaction between Normal 296 and Slow pedestrians. Thus this implementation can exclusively provide the 297 distance between two groups under different speed. 298 Toward the simulation, the value of PN and PS need to be set. According to 299 the observed statistical data, Dunbar (2004) shows the average walking speed of 300 the youth is 1.51 m/s and the elderly is 1.14 m/s where the speed ratio is 0.75. 301 Thus we choose the transition probability of normal speed walkers PN is set to 1 302 and that of the slow walkers PS is set to 0.75. Considering the size of each cell as 303 40cm x 40 cm, then a calculation step is equivalent to about 0.25 sec for 14 304 matching the speed. For evaluation of overtaking process, the normal pedestrians 305 try overtaking whenever the cell in front of him is occupied. Thus the provability 306 of successful overtaking POPN is set to 1. But the succession of overtake depends 307 on the occupation conditions of surrounding cells. 308 309 4. Results and discussions 310 The simulations are carried out for the same initial pedestrian's formation as 311 those for experiment 1, 2 and 3 where the results are shown in Fig. 9 (a), (b) and 312 (c) respectively. Each figure illustrates the pedestrian formation in the region of 313 interest at certain calculation step as specified in the figure. The CA model stated 314 in Fig. 8 is the stochastic process, thus Fig. 9 shows results of a single simulation 315 trial. Besides, some deviations can be seen for every simulation trials but the 316 main characteristics of the pedestrian formation are conserved. Fig. 9(a) 317 corresponds to the experiment 1 where the experimental result is shown in Fig. 3. 318 The simulation result shows that, the formation of the normal speed pedestrians 319 after overtaking is similar to that observed in the experiment in which, at the 320 alternative column, three pedestrians are aligned vertically with a space in 321 between. Secondly, the simulation result shown in Fig. 9(b) corresponds to the 322 experiment 2 shown in Fig. 4. The final pedestrian formation of normal speed 323 pedestrians are also vertical alignment for alternative columns. First few lines 324 have three pedestrians with no space in between and then it becomes two 325 pedestrian per column with a single space in between. This phase change in the 326 formation is due to the interaction between the slow pedestrians. The 15 327 intermediate formation at step 14 in Fig. 9(b) illustrates this interaction, where 328 the normal speed pedestrians are overtaking the slow ones and moderate jam is 329 observed behind the slow pedestrians. For Fig. 9 (c) referred to as experiment 3 330 and the experimental result shown in Fig. 5. Due to high density of the slow 331 pedestrians at the front, it is seen that the formation of normal pedestrian is much 332 affected by the slow pedestrians. High impact to the mobility of the normal 333 pedestrians creates the single horizontal alignment. This formation can be also 334 seen by experiment 3 where the result is shown in Fig. 5. The overall result 335 shows that the mobility of the normal speed pedestrians is highly affected by the 336 density of the slow pedestrians as well as its initial formation. In reality for 337 example, pedestrian crossing, sidewalk and stations in urban areas, the mobility 338 of pedestrians will play an important role for the sake of safety and crowd 339 management. This result provides further understandings of the pedestrian 340 dynamics when the elderly and younger people are cohabitating. 341 Furthermore, we have performed simulations for scenarios of various initial 342 formations to calculate the average travel time required for all normal speed 343 pedestrians to overtake the slow ones. Fig. 10 illustrates the calculated results. 344 Since the nature of the stochastic CA, number of simulation trials has repeatedly 345 carried out for the same conditions to obtain the statistically meaningful result. 346 The travel time is counted for each trial and the frequencies of results are 347 statistically resolved by means of probability densities. The calculation is 348 repeated until the probability density leaches to 99% saturation. One of the trivial 349 results is that, the travel time is short when the slow pedestrians are located 16 350 behind the normal pedestrians since there are no overtaking processes. However 351 this kind of arrangement is almost impossible for real world application. More 352 feasible possibility is that, separation of the walking lane for slow and normal 353 pedestrians as shown in scenario 14 in Fig. 10. The travel time of this scenario is 354 minimum among the considered scenarios since there is no interaction between 355 slow and normal speed walkers. The example of such social structure in the real 356 world is a highway that has slow and overtaking lane for cars. Contrary to this, in 357 the situation of slow and normal pedestrians are alternately placed like a 358 checkerboard pattern as scenario 9, which is closer to the real world situation 359 without pedestrian control. It takes more than double amount of time than that 360 for scenario 14. However, the important point to minimize the travel time and 361 maximizing the mobility is to reduce the number of interaction between slow and 362 normal speed pedestrians. 363 364 17 365 366 367 Figure 9: The results simulated for the same initial formation as experiments 1,2 and 3 respectively. 368 369 370 Figure 10: Averaged travel time to overtake slow pedestrians for various initial formations. 18 371 372 5. Conclusion 373 In this paper the pedestrian experiment has carried out for understanding the 374 characteristics of the pedestrians’ mobility when the elderly and the younger are 375 mixed. Further, CA model is created based on the experimental results by means 376 of the interaction between pedestrians with groups of different speed. It is found 377 that the emergent spatial formation is characterized by the initial formation, 378 walking speed and overtaking behavior. During the overtaking process, the 379 slower speed pedestrians are considered as moving obstacles to others acting as a 380 transient bottleneck. Current pedestrian model is in the simplest form but it 381 successfully shows the fundamental phenomena such as the spatial formation 382 and travel time. Simulation results suggest that separation of the walking lane for 383 slow and normal speed pedestrians is one of the solutions for the compatibility 384 issue such as elderly's safety and the non-elderly's mobility since it minimizes 385 their interactions. 386 387 Acknowledgement 388 This work is partially supported by Fondazione CARIPLO within the ALIAS 389 Project. (www.alias.disco.unimib.it). 390 391 392 393 19 394 References 395 Asano, M., Iryo, T., Kuwahara, M., 2010. Microscopic pedestrian simulation 396 model combined with a tactical model for route choice behaviour. 397 Transportation Research Part C 18, 842. 398 Burstedde, C., Klauck, K., Schadschneider, A., Zittartz, J., 2001. Simulation of 399 pedestrian dynamics using a two dimensional cellular automaton. Physica A 400 295, 507. 401 402 Chraibi, M., A. Seyfried, 2010. Generalized centrifugal-force model for pedestrian dynamics. Phys. Rev. E 82, 046111 403 Derrida, B., Evans, M. R., Hakim, V., Pasquier, V., 1993. Exact solution of a 1D 404 asymmetric exclusion model using a matrix formulation. J. Phys A: Math. 405 Gen. 26, 1493. 406 Dunbar, G., Holland, C., Maylor, E., 2004. Older Pedestrians: A Critical Review 407 of the Literature, Road Safety Research Report No. 37. Department of 408 Transport, London 37. 409 410 411 412 413 Ezaki, T., D., Yanagisawa, K., Nishinari, 2012. Pedestrian flow through multiple bottlenecks. Phys. Rev. E 86, 026118 Helbing, D., 2001. Traffic and related self-driven many-particle systems. Rev. Mod. Phys. 73, 1067. Helbing, D., Farkas, I., Vicsek, T., 2000. Simulating dynamical features of 20 414 415 416 417 418 escape panic. Nature, 407, 487. Helbing, D., Isobe, M., Nagatani, T., Takimoto, K., 2003. Lattice gas simulation of experimentally studied evacuation dynamics. Phys. Rev. E 67, 067101. Helbing, D., Molnar, P., 1995. Social force model for pedestrian dynamics. Phys. Rev. E 51, 4282. 419 Henein, M. C., White, T., 2007. Macroscopic effects of microscopic forces 420 between agents in crowd models Original Research Article. Physica A 373, 421 694. 422 423 424 425 Jelić, A., Appert-Rolland, C., Lemercier, S., Pettré, J., 2012. Properties of pedestrians walking in line: Fundamental diagrams. Phys. Rev. E 85, 036111. Jiang, R., Wu, Q., 2006. The moving behavior of a large object in the crowds in a narrow channel. Physica A 364, 457. 426 Kirchner, A., Kluepfel, H., Nishinari, K., Schadschneider, A., Schreckenberg, M., 427 2004. Discretization effects and the influence of walking speed in cellular 428 automata models for pedestrian dynamics. J. Stat. Mech.: Theory Exp. 364, 429 P100011. 430 Kirchner, A., Klüpfel, H., Nishinari, K., Schadschneider, A., Schreckenberg, M., 431 2003. Simulation of competitive egress behavior: comparison with aircraft 432 evacuation data. Physica A 324, 689. 433 Kirchner, A., Klüpfel H., Nishinari, K., Schadschneider, A., Schreckenberg, M., 21 434 2004. Discretization effects and the influence of walking speed in cellular 435 automata models for pedestrian dynamics. J. Stat. Mech., P10011. 436 Kirchner, A., Schadschneider, A., 2002. Simulation of evacuation processes 437 using a bionics-inspired cellular automaton model for pedestrian dynamics. 438 Physica A 312, 260. 439 Nagatani, T., 2002, The physics of traffic jams. Rep. Prog. Phys. 65, 1331 440 441 442 Nagel, K., Schreckenberg, M., 1992, A cellular automaton model for freeway traffic. J. Phys. I France 2, 2221. 443 Weng, W., Chen, T., Yuan, H., Fan, W., 2006. Cellular automaton simulation of 444 pedestrian counter flow with different walk velocities. Phys. Rev. E 445 74,036102. 446 447 448 449 Wolfram, S., 1994. Cellular Automata and Complexity: Collected Papers. (Addison-Wesley Publishing Company). Yanagisawa, D., Nishinari, K., 2007. Mean-field theory for pedestrian outflow through an exit. Phys. Rev. E 76, 061117. 450 451 452 22 Figure Click here to download high resolution image Figure Click here to download high resolution image Figure Click here to download high resolution image Figure Click here to download high resolution image Figure Click here to download high resolution image Figure Click here to download high resolution image Figure Click here to download high resolution image Figure Click here to download high resolution image Figure Click here to download Figure: Fig9.eps (a) 1 Intermediate formation, Step 0. 58 1 Intermediate formation, Step 22. 58 44 Intermediate formation, Step 76. 101 1 Intermediate formation, Step 0. 58 1 Intermediate formation, Step 14. 58 41 Intermediate formation, Step 75. 98 1 Intermediate formation, Step 0. 58 1 Intermediate formation, Step 21. 58 61 Intermediate formation, Step 100. 118 (b) (c) Normal speed pedestrian Slow speed pedestrian Figure Click here to download Figure: Fig10.eps Normalised Travel Time Averaged Required Travel Time to Overtake Slow Walkers Initial Formations 1 1 2 3 4 6 7 8 9 5 0.5 11 0 1 2 3 4 5 6 7 8 9 10 11 12 13 14 Scenario Number 12 13 14 Scenario Number is stated underneath of each formation 10 Slow Normal
© Copyright 2024