Pedestrian Behaviour Modelling An Application to Retail Movements using Genetic Algorithm Contents Requirements of pedestrian behaviour models Framework of a new model Test of shortest-path model Urban planning Spatial marketing Location-based services Kay Kitazawa, Michael Batty Genetic Algorithm simulation Centre for Advanced Spatial Analysis 2 Results 3 Urban planning Pedestrian-oriented urban planning Spatial marketing Not-compact city Deserted town centres Lively town centres Safety less crime, fewer traffic accidents Convenience accessibility to transport, shops, services Amenity comfortable walking environment Tenant strategy (leasing, fee) Improvement of -floor plans -signage system 4 Compact city Pedestrian-oriented planning 5 Actual movements Necessary information Influential factors Needs for Pedestrian behavior model 6 Actual movements Influential factors Needs for Pedestrian behaviour model
3D-GIS Location-based services Requirements of pedestrian behavior models Current spatial movement models trajectory Location planning There are several needs to develop pedestrian behaviour models Provide appropriate information according to user s location / needs Crowd dynamics Key issues How to avoid traffic jam? Transport model Understand and explain real pedestrian s movement Patterns of users routes/activities Necessary Information - contexts Network analysis and OD/route estimation Represent dynamic interaction process between pedestrians and their environment Needs for Pedestrian behavior model Where are my pals? Probability of choice Micro scale behaviour (e.g. obstacle avoidance) Routes for wheel chair user? 7 Requirements of pedestrian behavior models Positioning technology Stochastic model ( esp. Information which people obtain ) 8 Crowd dynamics Probability of state-to-state transition 9 Transport model Current position (xi, yi) Velocity (ui, vi) Radius ri Normal walking speed Vi Destination (pxi, pyi) (qxi, qyi) speed ratio ki Personal space ratio ci Information space (di, dit ) Stochastic model Area: S, S2 Sn Trips between Si to Sj : yij Distance between Si to Sj : dij B Origin Network analysis total Only the last state determines what will happen next Home ( weights associated with each link can be A distance, costs, condition of the road, etc) Destination Marcov chain model Influence of other areas? Which area generates more trips than others? Why? Probability of visiting from one place to another The observed number of people at their first destination Probability of being the last destination Home (OD) Gravity model Place (node) αi potential as origin βj potential as destination Number of people who visit each place via another ( Trip n : n> ) 3 home Estimation of the next steps of other pedestrians Trip 0 Most evacuation models adopt this concept Collision avoidance bahaviour 0 (Kai Bolay) Crowd dynamics Ltd Trip Trip 2 2 2
Requirements of pedestrian behavior models advantage disadvantage Crowd dynamics Well represent micro-scale physical response Not take it into account: where they are going to and why Dynamic pre-fixed route = static model geographical attributes Transport model Suitable for description of Several things can t be represented: selection behavior interaction between others/environment cognitive process of pedestrian Research Aim and Objectives To develop a new pedestrian behavior model be capable of explaining real pedestrian s movement Every factors should be determined based on observed data It can deal with more complex behavior (e.g. shopping ) represents dynamic interaction between pedestrians and their environment Framework of the model Built environment agents Geographic attributes Knowledge Needs Interaction between environment collision avoidance walking speed basic walking tendencies (e.g.avoid rapid turn over) Stimuli-Response Calculation of the optimum route shortest path cognitive process spatial knowledge Stochastic model Useful for being briefed on how people move around Capable of representing changeability of movements Inadequate to small scale movement Not explain why they choose certain place To deal with not only pre-determined route-choice but also people s cognitive process or other changeable events can be used as a simulation model To visualize, To make the model easy to understand, more transferable Integrated Simulation Model of Pedestrian Movements Pedestrian agents Matching between people s preference/needs and attributes of places Understand and explain real pedestrian s movement Represent dynamic interaction process between pedestrians and their environment be validated through comparison between actual trajectories Multi-agent-based model Which place to be chosen as a destination? 3 New pedestrian behaviour models are needed 4 It should be different from playing with beautiful animation 5 Framework of the model 3 levels of pedestrian s behavior congestion Obstacle avoidance Methodology Survey of pedestrian movement in public spaces Test Simulation ~m / min Stimuli-Response How they walk around? feedback Stimuli-response Measurement systems / sensors Trajectory walking patterns Retail movement in a large shopping centre» Visitors have the same objective = Shopping» Survey area has distinct boundary» Shoppers walk around model O D Spatial knowledge Environmental info collecting information New info Records of Optimization criteria Surveys of route-choice behaviour Route A Route B DESTINATION Route C Shortest path model Which route to take? research 6 DB Strategy of each shops marketing choose destinations User s Cost Attributes Distance matching Preference Needs Restriction 7 Geo-demographic Database Develop DB of attributes of the place Analysis on relationship between the shop s attributes and those of individuals 8 Planned destination Comparison START Real route 3
Type Category Category 2 Proposed critical factor YES Satisfaction Typology of shoppers Shop-till-you-drop consumer Shop explorer NO information POTENTIAL Visibility of potential purchases Repeat guest ( Regular customer) Fixed route middle Shopping opportunity (Time) Visibility of potential purchases People who doesn t like to shop YES Spatial knowledge NO Surveys of route choice behaviour Tracking retail movement 8 samples (female, 20 year-old) 2 hours shopping * 3 times Analysis on influential factors on shopper s route choice Retail movement Test simulation using GA Time resolution=30 seconds Time 0 2 3 4 5 A B C D E chromosome Route Knowledge about the place Time constraints Preferences Floor plans/ networks of the shopping centre shop network trajectory Behaviour pattern 9 Complex Time: long Try to see whole area Shortest path & Other factors Shortest path Time: long Deviate from prefixed route by visual stimulus Shortest path Time: short not go shopping 20 Shop-till-you-drop consumer? People who doesn t like to shop? Sample trajectory 2 326 nodes (shops, centre points of corridor-every 0m) 364 links (corridor) Test simulation using GA Results Results Observed route d3 dn Evaluation criteria N maxv = a i x i i= α Parameter X Evaluation function for criterion i calibration Observed route Observed route (real route) / n D = d d2 Travel distances (the shortest-path model) Does it include the ID of nodes which were scheduled to visit? Prefixed Start point and Goal point 22 Physical restriction walking speed (average 60 metres per minute) rotation angle (less than 50 degree) limited vertical movements Test simulation Evaluated value 7.62 Observed route 7.69 23 without restriction on distance Set weighted parameters values with severe restriction on distance Given the real route Simulation Simulation 2 as one of initial chromosomes Evaluated value 00 99.5 08.8 Distance between 2 routes 68.8m 52.4m.25m 24 4
Conclusion Future research Current positioning technologies 25 Shortest path model capable of predicting outlines of the routes other influential factors Evaluation criteria and parameter values tested 26 Improving the simulation system Combining network and potential distribution Network analysis 評価ポテンシャル場 Potential value 店舗ノード node 店舗フロア Shop polygon 店舗ノードとの距離に応じて評価 factors in route selection width of corridor, visibility, connection to other network research Develop DB of attributes of the place Analysis on relationship between the place s attributes and those of pedestrian What kind of people go to WHICH place (shop/restaurant) HOW often? WHY? Implement simulation 27 GPS-based technology Cell-based technology Image processing Autonomous-positioning Laser scanning Ultra-sonic wave Traffic counter Laser scanning GPS RFID tag Thermal infrared gyro compass barometer magnetic Ultra-sonic wave sensor Autonomous positioning system Thank you! Kay Kitazawa k.kitazawa@ucl.ac.uk http://www.casa.ucl.ac.uk/kay 28 5