Suggestion for a Visual Dynamics Analysis Model Using a Natural Movement Model

Transcription

1 Suggestion for a Visual Dynamics Analysis Model Using a Natural Movement Model Seung-Jae Lee 1 and Kyung-Hoon Lee* 2 1 Assistant Professor, Division of Architecture, Mokwon University, Korea 2 Professor, Department of Architecture, Korea University, Korea Abstract The purpose of this study is to suggest a visual dynamics analysis (VDA) model that comprises a visibility analysis model integrated with a natural movement model. The theory of 'natural movement' from the background of Gibson's ecological theory of perception and Hillier's research could help to understand the relations between visual perception and movement. In a previous study, the natural movement model called the EVA (Exosomatic Visual Architecture) model was proposed by Turner and Penn with the concept of visual dynamics. However, the EVA model could not give any information about agent visual experience, but only a pattern of vision-driven agent movement. This is because spatial visual information already exists outside of the agent. Most of all, the model limits any visual dynamics analysis because the agents' movement patterns are irregular at a micro level. Therefore, in this study, a new natural movement model rule that shows good movement patterns at the micro scale is discussed, and visual indexes considering human visual characteristics and dynamic analysis are proposed. These are integrated with the VDA model and programmed with NetLogo. The analysis of test spaces by the VDA model and the meaning of the model are discussed. Keywords: visual dynamics; visibility analysis; natural movement; pedestrian simulation; agent-based model 1. Introduction The main factors in perceiving space are 'vision' and 'movement'. Visual perception through movement generates spatial experiences. As a result, they have been important subjects in architectural and urban studies. The purpose of this study is to suggest a visual dynamics model composed of a visibility analysis model integrated with a natural movement model. Turner (2003) used the term 'visual dynamics.' He said that visibility analysis should assess not the visibility alone, but the visual ecological process which occurs between occupant and space. He discussed visual dynamics with respect to the EVA (Exosomatic Visual Architecture) model, which was proposed by Turner and Penn (2002). They encoded natural movement as an agent-based system in the model. Through this simulation model, movement patterns generated by vision-driven agents can be investigated. However, their model could not give dynamic visibility information such as an agent's visual access or spatial visual exposure by agents during or after *Contact Author: Kyung-Hoon Lee, Professor, Department of Architecture, Korea University, 5 Ga-1, Anam-dong, Seongbuk-gu, Seoul, , Korea Tel: Fax: kh92lee@korea.ac.kr ( Received October 6, 2013 ; accepted March 7, 2014 ) movement. Because the term 'exosomatic' means that visual information at every space (point) should be measured prior to a simulation, agents could not have interactive vision or visual memory, but only refer to exosomatic visual information for their movement. We might have only to assign an agent's visibility module, which could be measured in real time, to their natural movement model in order to obtain information of the agents' visual experience. However, the movement pattern at a micro scale in the EVA model is somewhat irregular and differs from our general behavior. It resembles Brownian motion. The reason is that a pattern comes from the agent-decision process in which agents pick a random point, or, more correctly, a visibility graph vertex within their visual field every few steps. Even though the result of the EVA model is well correlated with the global movement patterns such as a gate counts survey, a proper visual analysis could not be expected from the simulation showing irregular movement behavior. This is because seeing is a continuous behavior and depends on an agent's heading. Therefore, we considered another agent's rules to obtain a more plausible natural movement pattern at a micro scale, and a visual dynamics analysis model which could measure what agents had seen through natural movement. Prior to this analysis, related studies and their limitations are discussed. Some visibility indexes for visual dynamics analysis are introduced because Journal of Asian Architecture and Building Engineering/May 2014/

2 we need to consider human visual perception and its dynamics. Our natural movement model has a concept similar to that of the EVA model, but with a different way in which perceivable objects such as walls, columns or other agents might afford their movement. The reason is that considering perception, objects are encountered prior to walkable surfaces, such as floors. In a previous study (Lee et al., 2013) using an agentbased model and applying this rule, the pattern of natural movement was smoother and more continuous at the micro scale than the EVA model, and showed good correlation with the gate-counting data. This model will also be introduced shortly in this paper. We then integrate a real-time visibility process into the model for visual dynamics analysis. Using the analysis model, some environments are tested and analyzed. Finally, we discuss the meaning of the model and its potential applications. 2. Related Works 2.1 Seeing and Movement We continuously look around during movement and perceive an environment. In Gibson's (1979) ecological perception theory, he indicated a complementary relationship between visual perception and movement. Humans obtain spatial information by seeing, which affects their movement. The movement proceeds to recursively give them more visual information. For this reason, we could not understand how humans could have a spatial experience without considering vision and movement together. There have been many studies on visual perception and movement. The studies of visual perception were mostly based on the perspective of visual cognition, such as aesthetics or subjective feelings or memories of space. In the field of movement study, the main focuses were on modeling for the prediction of the pedestrian volume or of patterns through comparisons with field surveys. However, both fields have been dealt with separately, except for some studies that were interested in the visual sequence. In his book 'Townscape,' Cullen (1961) showed us continuous urban scenes through movement and proposed that a wider urban experience could be generated by various sequences. Thiel (1996) developed a 'Sequence Notation Methodology' as a visual experiential score. Even though it is hard to symbolize and understand the score, we found that his method could make it possible to understand a participant's visual experience integrated with his movement through time and his path in space. These studies demonstrated efforts to integrate visual perception and movement. Due to the advance of computing technology, it has become possible to measure visibility quantitatively and to simulate movement using an agent-based model. The visibility could be measured by radiating optical arrays from a point or checking intersections which cause a visual hindrance between point pairs in a convex space. The movement, however, is generated by a system that is so complex and dynamic that it is hard to investigate the mechanism and make a model. Using an agent-based model, many pedestrian simulation models have been introduced. However, most of these models could be navigational models which should have OD (Origin-Destination) pairs without consideration of agent' vision. If vision is complementary with movement as in Gibson's discussion, the pedestrian models without agent vision seem to have overlooked the relationship between them. On the other hand, visibility analysis models without agents could be used to grasp spatial visual features only, not what visual experiences we could have. 2.2 VGA Model The VGA (Visibility Graph Analysis) by Turner et al. (2001) came from an idea which integrates the isovist model as a visibility analysis model with space syntax theory. Benedikt (1979) defined isovist as 'the set of all points visible from a given vantage point in space' (page 47) to measure visible relations between all points in an environment. Although Benedikt also proposed a Minkowskian isovist model considering visual change by movement through time, a common visual analysis of an isovist area field was generally static. The space syntax theory by Hillier and Hanson (1984) was based on network theory. They thought that the connection between spaces might be related to human behavior which would embrace 'social' interpretation such as movement. As Hillier et al. (1993) discussed with 'natural movement,' we found that a network structure of space is relatively correlated with movement. Turner et al. discussed the limitations of isovist methodology, which only measures the local properties of space and has no consideration of the visual relationship of space, and then proposed a visibility graph analysis. Though he said 'relationship,' it could be regarded as 'movement.' Therefore, we think that VGA is a type of visibility analysis considering movement. However, VGA only deals with spatial visibility and its relationship to movement. There are no agents in an environment. 2.3 EVA Model As a target model of this study, which considers the relationship between an agent's movement and their vision, we focus on the EVA model as a natural movement model by Turner and Penn (2002). They pointed out the limitations of most pedestrian models that should have only OD (Origin-Destination) pairs and no agent vision. Hillier et al. (1993) discussed 'natural movement,' which would be a basic movement generated by the urban grid configuration in which the correlations between the global variables of the configuration and the movement patterns could be found. Turner and Penn extended Hillier's theory with Gibson's affordance theory and transferred the generator of natural movement from the relationship 382 JAABE vol.13 no.2 May 2014 Seung-Jae Lee

3 of configuration (Hillier et al.) to a walkable surface (Gibson). They thought of walkable surfaces as a provider of the possibility to move, and proposed a rule in which agents would select a random point as a temporary destination in their visible floor every few steps and then walk towards it (page 480). The iteration of the process generated a natural movement. Since the visual information should be calculated prior to agent decisions concerning movement and was external to the agents, they called the model the Exosomatic Visual Architecture (EVA) model. They appeared to be trying to integrate a movement model with a visual model. Turner (2003) later discussed this model with the concept of visual dynamics. However, as mentioned above, we could not obtain a visibility analysis, and only a natural movement pattern could be obtained from the EVA model simulation. This is because agent vision is exosomatic. Exosomatic visual architecture is helpful to reduce the simulation running time, but there is a contradiction with his initial concept since vision is external to the agents. It is impossible for agents to change their visual field for ecological adaption to dynamic environments. Most of all, the system gives us no data to examine what the agents have seen because agent visibility cannot be saved in real time. If that were the case, it would have only to import a module to calculate an agent's vision in real time into the model. However, this method has another serious problem. Since the agents have a decision rule to select a temporary destination target in their visual field, they move in zigzag patterns, or appear to be roving randomly and often bump into a wall. We could observe this irregular movement pattern at a micro scale with Depthmap software, where the EVA model is programmed. It is such an unusual pattern that the mere supplementation of such a visibility module would not help to obtain a practical analysis. So, this study needed another movement rule for an agent-based natural movement model with an interactive visibility module, and some indexes for visual dynamics analysis were proposed. 3. VDA Model 3.1 Visual Indexes General visibility analysis models such as the VGA model were developed from the isovist model by Benedikt. These models seem to have two critical but paradoxical suppositions that human visual ability is omnipotent and seeing behavior equally and simultaneously occurs throughout all spaces. This means that the distance to which humans can have visual access is limitless, and the difference of visual perception according to distance is ignored. All spaces should be occupied at the same time by agents who can see in all directions. However, our behavior cannot be like that. We keep 'the eyes-in-the-head- on-thebody-resting-on-the-ground' forward, as in Gibson's expression (1979: page.205), so the visual field ought to be partial. The dynamics of movement would generate an inhomogeneity of occupancy. For this reason, a previous study (Lee, 2009(b)) considered a partial isovist, which consists of three variables: the agent's orientation (α), the visual field angle (θ), and the maximum visual distance (l f ). A partial isovist of an agent (A) would be expressed as I (α,θ,lf )(A). Another consideration was the visual level (VL), which is affected by distance. We proposed and defined the visual level at a point (S) with distance (r) as 'the ratio between the maximum visual distance and the distance from the maximum visual distance.' This could be expressed as: (S) = ( ) = ( ) If the distance at a position is more than the maximum visual distance, the visual level (VL) will be 0. The visual level within a partial isovist will be between 1 and 0. The closer a position is to an agent, the greater the visual level at the position is. Moreover, we considered visual access (VA) and visual exposure (VE). We thought visual access could be measured by how much space an agent could see, whereas visual exposure depends on how much visual access the agent has to a space. The distinction between the values in conventional static analysis with a full isovist area has no meaning because they would be the same at a certain point. However, considering a partial isovist in a dynamics model, the values could be different between VA and VE. Using the visual level (VL), the visual access level (VA VL ) of an agent (A) would be formulated as a sum of the visual level at all points within a partial isovist: Fig.1. Difference of VA VL by Shapes with the Same Isovist Area (VA VL (α) < VA VL (b)) (A) = () (). Comparing this with Benedikt's isovist area, the visual access value (VA VL ) considering the visual level (VL) could represent the shape of an isovist. For example, with the same isovist area, VA VL in a square room will be greater than VA VL in a narrow corridor, which means that a square room is an easier space to have visual access to than a narrow corridor [Fig.1.]. The visual exposure level (VE VL ) at a point (S) would be a sum of the visual level of the visual access by all agents: (S) = () ()}. JAABE vol.13 no.2 May 2014 Seung-Jae Lee 383

4 a) Full Isovist Field or Connectivity of VGA Model b) Mean Depth of VGA Model c) VDA Model Fig.2. Conceptual Diagram of a Static Visibility Graph Analysis Model (a, b) and a Visual Dynamics Analysis Model (c) In a dynamics model, we are practically interested in the visibility variation of an agent who moves in an environment or the visibility accumulation of a space which agents have seen during a natural movement simulation. When focusing on an agent, the fluctuation of his (her) VA VL can be plotted on a time graph. This would be called the DVA (Dynamic Visual Access). On a space, the DVE (Dynamic Visual Exposure) at space (S) would be a sum of all VE VL (S) values during the simulation time: DVE(S) = () ()}. So, if we wish to have a static analysis of the isovist area field through the DVE concept, we only have to position an agent at all points and calculate the DVE without movement of the agents who have limitless visibility: () () ()}. 3.2 Natural Movement Afforded by Wall Vector We discussed the problem of the irregular movement pattern of Turner's EVA model at a local scale, and rethought his concept of the EVA model. He regarded a walkable surface as a provider of the possibility of natural movement. A walkable surface could be interpreted as a floor in a built environment. A walkable floor is bounded by visible objects such as walls. Walls should be perceived prior to a walkable floor. Therefore, perceivable walls could be a provider of natural movement according to his concept. We generally perceive walls or fences rather than floors when walking. A previous study (Lee, 2011; 2013) included an experiment on natural movement. Fifty subjects without information of a whole space or the destination were asked to move naturally in some virtual models. The remarkable result of the experiment was that the subjects' movements seemed to be more guided by objects like columns or walls. However, when simulated by the EVA model in Depthmap under the same conditions, the agents' movements tended to be driven by the open space. In an ideal open space with only a floor and without any objects, we would be bewildered by the absence of references to guide our movements. Turner (2006) also discussed the difference between observed people paths and the movement modeling patterns of agents driven by direct-perception in a large open space. People showed the tendency to move along the major axes according to the boundaries, whereas agents were aggregated in the open space. As a result, he formed another hypothesis that agents would be guided by occluding radials that mark a boundary between visible and occluded objects. The simulation result showed that the occlusion-driven agents picked out the sorts of observed pedestrian paths much better than the agents driven by direct perception. However, an experimental study (Lee, 2011) showed that the simulation results of his new approach did not correspond well with the observed subjects' movement data in some cases. It is notable that occluding radials are shaped by perceivable objects such as walls, but we cannot directly perceive occluding radials. Moreover, humans tend to pay attention to farther walls, which give more information regarding the possibility to move. For this reason, we proposed another agent rule, that a configuration of perceivable walls would guide an agent's natural movement. The 'configuration' could be modeled as vectors with direction and distance from an agent to the walls. Therefore, in our model, first, 384 JAABE vol.13 no.2 May 2014 Seung-Jae Lee

5 the sum of the vector field to perceivable (visible) walls would guide the agent's movement. Other agents would also be perceived as visual hindrances. This rule would naturally make agents avoid collision with walls or other agents, even though the next rule would need a collision avoidance rule. Second, there would be a noise variable such as looking around to avoid route fixation because of the first rule. The noise variable could make it possible for agents to explore unexpected routes. Third, if an agent's partial isovist is so small or has no occluding radial, it means that there is no more space to explore, so the agent will turn back. Since a wall in a built environment is generally continuous, it could make agent movement more stable and continuous. This model was introduced in a previous study (Lee, 2013) where the simulation result in the Tate Gallery produced good movement patterns both locally and globally. The data of the total 67 places in the actually measured data were used for analysis, and the correlation coefficient of the new natural movement model by Lee (2013) was r=0.7678, which was similar to the correlation coefficient of the EVA model by Turner (2002) (r=0.72). The EVA model also shows good correlation with the observed visitor volume at a global scale. However, the irregular movement of the EVA model at a micro scale might induce unsuitable visual dynamics analysis. The plausibility of a microscopic movement pattern would be important to examine what agents have seen. However, in the study (Lee, 2013), the agents had vision in order to move, but there was no consideration of visual analysis through the agent's movement. So, we put a visual analysis module with the indexes mentioned above into the new natural movement model using a wall vector field, and called this model the visual dynamics analysis model (VDA model). The meaning of visual dynamics analysis is discussed through some applications in this paper. by the moving agents (DVE), or a graph showing changes in the status of the visual access of a moving agent through time (DVA). The model was applied in this paper to a T-shaped space and Mies's work. 4.1 T-shape Space The T-shaped space was introduced in many visibility analysis models. Previous visual analysis models were static and deterministic compared with the VDA model. The VDA model is compared with the VGA model as a static visual analysis model and the EVA model as a natural movement model [Fig.4., Fig.5.]. The test environment consists of cells. Fig.4. is the result of the simulation with an agent with a visual field angle θ = 120, maximum visual distance l f = 60cells, noise n 30 noise is a variable of looking-around behavior which might generate unexpected route choices and speed s=1 cell/t. The agent moved for a specific simulation time t = 2,000. The agent had a counterclockwise rotation movement around the T shape in this simulation (but it could be clockwise or other movements in other cases). As a result, the distribution pattern of DVE appears asymmetrical like a pinwheel. According to the direction of the agent's movement, its DVE pattern could also be reversible. The DVA of the agent could be examined with the DVA-time graph [Fig.4.-c)]. The graph shows that the visual access value fluctuated so much that the agent might experience open and closed spaces repetitively. a) Movement Pattern b) DVE Distribution c) DVA-Time Graph of an Agent during 270 steps Fig.3. Concept of the VDA Model 4. Applications As a discrete model, the VDA model was programmed in NetLogo. The visibility information of an agent is updated at every time step. The information also guides an agent's movement. After the simulation was complete, we could obtain the agents' natural movement patterns and a distribution of spatial visual exposure accumulated Fig.4. DVA Simulation Results of T-shaped Space In contrast, the visual connectivity, which is similar to the isovist area field, or the visual integration of the visibility graph analysis (VGA) model as a static model, shows a symmetric pattern vertically [Fig.5.-a), b)]. It is an ideal one which is not shown in the real world since human occupation and movement are generally asymmetric. The static models are affected by JAABE vol.13 no.2 May 2014 Seung-Jae Lee 385

6 a) Connectivity of VGA (=Full Isovist Area Field) b) Visual Integration of VGA c) Gate Counts of EVA d) Movement Patterns of EVA Fig.5. VGA and EVA Results of T-shaped Space a spatial geometry, not by the agents. So, it appears like a geometrical analysis, as a visual topology using the visual-spatial relationship, whereas the VDA model is focused on the agent's experiences, which are moving and seeing related to his (her) environment. In this case, the maximum visibility points are not at the corner or at the edge of the rectangular boundary, but at the corner of the T-shape. However, the more agents and the more simulation time there are in the VDA model, the closer the dynamic visibility pattern would be to a static analysis. Fig.5.-c) is the result of the gate counts at every cell using the EVA model with agent released rate (agent per time step) 0.01, analysis length 5,000 timesteps, and agent field of view 9 bins ( ). Between them [Fig.4.-a), Fig.5.-c)], a similarity in the whole occupation pattern could be found. Although their correlation coefficient is (p<0.01), the numerical comparison at every cell seems meaningless. However, it is important that the agent movement patterns of the EVA model at the micro scale [Fig.5.-d)] are irregular. The irregular movement patterns would converge into the whole occupation pattern like [Fig.5.-c)], but human movement is not generally like that. The visual experience analysis using the EVA model could not guarantee validity. So, it would be suitable to use the new natural movement model by Lee (2013) for the VDA model, in which the wall vector field guides the agent movement. It also supports a multi-agent model in which agents could avoid collisions with each other. 4.2 Brick Villa by Mies Mies van der Rohe was a modern architect. His works show many features of modern architecture, such as 'transparency' and 'movement' (Giedion, 1967). However, the application of previous static visual analysis models in his works has limitations in that there is no movement concept, and the operational definition for the visual range by its transparency is ambiguous. So, we applied the visual dynamics analysis model to his brick villa, which was not constructed, but has characteristic designs of the wall arrangements. First, when applying the isovist area analysis model to an environment with transparent walls like glass windows, the analysis requires us to define the visual range. However, the results are different according to the range, as shown in [Fig.6.], because there is no visual limit distance. If the range of the analysis is out of a building, the visibility value inside would be affected by the outer size. However, the outer size could also vary with the outer space range defined by an analyzer. A restriction of the visibility range to the inner space would leave a problem with no consideration of transparency. So, a visual index considering visual distance would make it possible to overcome this problem. The result of the VDA model is shown in [Fig.7.]. In the simulation, an agent with a visual field angle θ = 90, maximum visual distance l =30m, noise n 30, and speed s=1cell/t had moved for the simulation time t=5,000, and we could obtain the DVE distribution. The result shows that the space around the inner center wall has more visual exposure than other spaces. Also notable is the fact that the outward long walls were more visually exposed than any others. a) Isovist Area Field with Outer Space b) Isovist Area Field within Inner Space Fig.6. Difference of the Isovist Area Field by the Analytical Range 386 JAABE vol.13 no.2 May 2014 Seung-Jae Lee

7 a) Natural Movement Pattern a) Visual Integration of VGA b) DVE Distribution Fig.7. Simulation Result of the Brick Villa (original plan) The DVE is compared with the visual integration of the VGA model [Fig.8.]. Their correlation coefficient is good as r=0.766 (p<0.01). It shows that the visual integration is related to movement, because the DVE could be obtained by the agent's movement. However, the VGA model also has the same problem in that the result might be different depending on the analysis range, or there is no consideration of the agent's visual direction. Fig.9. shows the gate counts at every cell using the EVA model with agent released rate (agent per time step) 0.01, analysis length 5,000 timesteps, and agent field of view 9 bins ( ). As disscussed in the T-shaped model, the whole movement pattern is similar to [Fig.7.-a)]. However, it could not give any information on the agent visual experience. The visibility analysis and the natural movement simulation are integrated into the VDA model. The visual information and the movement could be examined simultaneously by the model, and there is no need to define the visual range. To examine what effect a change of the inner center wall would have on the visibility features, we removed the partial wall to form a longer visual axis inside. This would alter the agent's movement. The simulation results with the same conditions above except for the wall configuration are shown in [Fig.10.]. It shows that the DVE of the outward walls was weakened compared with the configuration in [Fig.7.-b)]. b) Scattergram between VGA and DVE Fig.8. Analysis Results of VGA Fig.9. Gate Counts of EVA These outward walls seem to be exaggerated in the whole original drawing. We cannot know whether Mies intended to focus on that or not, but if the walls were intended, this analysis shows that someone moving in the brick villa would have more visual access to the outward walls, and the inner center wall might play an important role in an agent's visual experience. Previous static visual analysis without consideration of movement could not find this. 5. Conclusion Turner showed us a visual dynamics concept for vision-driven agent movement pattern, but in his model, we could not investigate the status of the JAABE vol.13 no.2 May 2014 Seung-Jae Lee 387

8 a) Natural Movement Pattern visual access of an agent or the visual exposure in an environment through the agents' natural movement. So, by integrating a visibility analysis model into the new natural movement model, this study extended the previous static visual analysis model to another dynamic model. The result of the VDA model in which movement was generated by the visual interaction between agents and their environment would help us predict the actual visual experiences. The EVA navigational model has some problems in integrating the visual analysis model for an irregular movement pattern at a micro scale. Hence, we proposed another natural movement rule in that the agent's movements are generated by a vector field of perceivable walls. The visual dynamics analysis model where visual indexes considering human visual behavior could be measured in real time was programmed by NetLogo. Some applications of the model showed the possibility for expanding the visibility analysis into a dynamic one, so that the analysis would give us plausibility and diversity of interpretation. One of the implications of the visual dynamics analysis in this study is that the model could obtain more practical visual analysis reflecting the users' behavioral characteristics. Second, it could help to understand the correlation between visibility and movement. Third, it could examine the affordance of an environment through movement simulation by visiondriven agents. Finally, it could carry out quantitative and qualitative visual analyses at the same time. However, we should develop a natural movement model to obtain a more valid visual dynamics analysis. The mechanism of movement is so complex that it is hard to make a model. This needs follow-up studies in order to support its plausibility through gathering and comparing real-world data, and fitting the variables of the movement and visibility model. It might also be required to develop the representation of the pedestrian simulation to a higher level, and to improve the program technically. This study has significance regarding the expansion of static visual analysis to visual dynamics analysis, and in the integration of models into one system between visibility and movement models. Acknowledgement This research was supported by the Basic Science Research Program through the National Research Foundation of Korea (NRF) funded by the Ministry of Science, ICT & Future Planning (2013R1A1A ). b) DVE Distribution Fig.10. Simulation Result of the Brick Villa (altered plan) References 1) Benedikt, M. L. (1979) To take hold of space: Isovists and Isovist Fields. Environment and planning B 6. 2) Cullen, G. (1961) The Concise Townscape, London: Architectural Press. 3) Gibson, J. J. (1979) The Ecological Approach to Visual Perception, Houghton Mifflin. 4) Giedion, S. (1967) Space, time and architecture: the growth of a new tradition. Harvard University Press. 5) Hillier B., Penn A., Hanson J., Grajewski T., Xu J. (1993) Natural movement: or, configuration and attraction in urban pedestrian movement. Environment and planning B 20, pp ) Lee, Seungjae (2009) Visual Access and Exposure Model for the Agent-based Visual Analysis: Suggestion and Examination of a New Index. Journal of the Architectural Institute of Korea 252, pp ) Lee, Seungjae (2011) A Study on the Spatial Factors Affecting Natural Movement: through the Construction of Experimental Model for Natural Movement. Journal of the Architectural Institute of Korea 269, pp ) Lee, Seungjae (2013) Study on a Pedestrian Simulation Model of Natural Movement. Journal of Asian Architecture and Building engineering 12, pp ) Thiel, P. (1996) People, Paths, and Purposes: Notations for a Participatory Envirotecture. University of Washington Press. 10) Turner, A., Doxa, M., O'Sullivan, D. and Penn, A. (2001) From isovists to visibility graphs: a methodology for the analysis of architectural space. Environment and Planning B 28(1), pp ) Turner, A., Penn, A. (2002) Encoding natural movement as an agent-based system: an investigation into human pedestrian behaviour in the built environment. Environment and planning B 29, pp ) Turner A. (2003) Analysing the visual dynamics of spatial morphology. Environment and Planning B 30, pp ) Turner, A. (2006) Isovist, Occlusions and the Exosomatic Visual Architecture. Bartlett School of Graduate Studies, UCL, London, pp ) Turner, A. and Penn, A. (2007) Evolving direct perception models of human behavior in building systems. In Pedestrian and Evacuation Dynamics 2005, edited by Waldau, N. et al. 388 JAABE vol.13 no.2 May 2014 Seung-Jae Lee