CN103279800B - Based on game theoretic large-scale crowd evacuation optimum escape route generating method - Google Patents

Abstract

The present invention is in order to when there is disaster, during simulation large-scale crowd evacuation, calculate the Evacuation Strategies of optimum, and during this basis is to whole evacuation occur crowding phenomenon, each evacuates the crowd density situation in path, critical path artificial abortion's situation is analyzed statistics, it is achieved the generation to the optimal strategy of crowd evacuation.The present invention provides a kind of optimization method based on game theoretic large-scale crowd evacuation path, according to disaster radiation areas scope, at a distance of Safe Escape Distance, the safety measure number that best-effort path includes, degree of crowding factor carries out quantitative risk analysis, and quantitative risk analysis uses Monte Carlo model, using the Risk Results of assessment as the input parameter of the revenue function of path game, determine to evacuate the choice in path by game theory, form optimum dispersal plan.

Description

Based on game theoretic large-scale crowd evacuation optimum escape route generating method

Technical field

The inventive method is based on game theoretic large-scale crowd evacuation escape optimization method under complex scene, crowd More concentrate, in the case of jam is extremely serious, it is possible to obtain gratifying degree of optimization；This method can extensively be applied Crowd's best-effort path and crowding phenomenon problem during under various scenes, the optimal problem of crowd's Evacuation Strategies, statistical analysis are evacuated, Thering is provided foundation for architectural design scheme assessment, that reduces Dense crowd produces probability, the building knot in the different clustering place of combination Structure feature and composition of residents character, it is provided that there is optimum crowd evacuation path navigation targetedly, improve people from public place, city Group dredges efficiency, other this method breaks away from region, the time, passenger flow quantitative limitation, provide very for the department such as fire-fighting, public security, medical treatment Real quickly emergency preplan content, improves emergency response and collaboration capabilities.

Background technology

Crowd evacuation analogue technique is increasingly mature in recent years, the evacuation model of comparative maturity such as: social force model, fluid are dynamic Mechanical model, cellular Automation Model and personnel's motion LATTICE GAS MODEL WITH；Hydrodinamical model can simulate evacuation process, for The natural quality simulation of people is true not, and its Evacuation Strategies reference value generated is by big heavy discount；Social force model, cellular are certainly Motivation Model and personnel's motion LATTICE GAS MODEL WITH can simulate behavior and the evacuation process of people more really, local optimum determine Global optimum, so the overall Evacuation Strategies produced is extremely difficult to optimum；Therefore true mould is carried out according to the natural quality of personnel The method of the Evacuation Strategies intended and can generate overall situation and partial situation's optimum is the key point solving crowd evacuation problem.

Summary of the invention

The invention aims to when there is disaster, during simulation large-scale crowd evacuation, calculating optimum Evacuation Strategies, and during this basis is to whole evacuation occur crowding phenomenon, each evacuate path crowd density state Gesture, critical path artificial abortion's situation is analyzed statistics, it is achieved the generation to the optimal strategy of crowd evacuation.

The present invention provides a kind of optimization method based on game theoretic large-scale crowd evacuation path, according to disaster radiation area Territory scope, at a distance of Safe Escape Distance, the safety measure number that best-effort path includes, degree of crowding factor carries out Quantitative risk Analyzing, quantitative risk analysis uses Monte Carlo model, using the Risk Results of assessment as revenue function defeated of path game Enter parameter, determine to evacuate the choice in path by game theory；During evacuating, each escape personnel is as an Agent Processing, each Agent has the natural quality of oneself, such as: speed, quality etc. at heart；When Agent meets (adjacent Agent speed Degree difference is an outlet less than 0.1m/s and destination) time, qualified Agent is merged, forms an Agent Bunch, improve running efficiency of system by processing Agent bunch as an Agent individual subject.

Cause danger start evacuate during, the first step be select evacuate path, with room for ultimate unit apply Monte Carlo model, is estimated following factor: each room calculates its shortest path arriving all extra exits (Exit) Footpath ranking P_i, i ∈ N | N is room number, further according to P_iCalculate degree of danger (dangerous radiation scope, the fatal journey at a distance of burning things which may cause a fire disaster Degree), according to the average speed S of Agent all in room_aComplete each shortest path required time t_ai、t_aiExit after time_i's Congested conditions, Exit_iAdjacent emergency exit number and distance, evacuate safety measure situation on path；Commenting of Monte Carlo model Valuation is as the input parameter of game theoretic revenue function, and additionally input parameter also includes the crowd density situation evacuating path (number/unit area), obtains Bayesian Nash equilibrium point, determines the path of evacuation according to the numerical value of equilibrium point.

Additionally use game theory can solve the collision problem congested conditions from macroscopic perspective, i.e. set current evacuation Adjacent emergency exit (NeiExit) number of outlet (CurExit) is N, arrives i-th NeiExit_iTime is tn_i, i ∈ [1 ... N], the degree of crowding GCrowdLevel (NeiExit of i-th emergency exit_i), it is thus achieved that to NeiExit_i, i ∈ [1, ... N] emergency exit risk class list GHazardLevel (NeiExit_i), tn_iThe degree of crowding of moment CurExit GCrowdLevel (CurExit), is obtained the input parameter of revenue function by factors above by Monte Carlo model, and application is rich Play chess opinion and obtain Bayesian Nash equilibrium point, determine that the solution solving congested conditions (is that original place waits or selects other roads Escape in footpath).

Equipment needed thereby: use 55 array monitoring camera-SLP-ZS39160D, VGA6464 computer VGA Inverse problem Device, PC, Haikang DVR DS-8016HS-S mono-, and for connecting the circuit of PC and embedded device；Photographic head Layout: photographic head working range is long 30 meters wide 6 meters of long regions；Photographic head arrangement principle is to cover whole corridor area, from just Door starts to lay, and the first floor for building is arranged, other floors are arranged by this principle, one layer of totally 11 photographic head, 5 floor Need 55 photographic head altogether.

Method step is as follows:

(010) part, video surveillance specifically comprises the following steps that

Step C011: according to the video extraction background of photographic head shooting, obtains pattern mask；

According to threshold method, step C012: set time threshold x, judges whether background undergos mutation；If background is undergone mutation Then arranging the image after change is background, utilizes the adaptivity of background model to be caused when initialization background by moving target Impact eliminates, and then utilizes adaptive RTS threshold adjustment algorithm, follows the tracks of background detail；

Step C013: pedestrian target detection uses frame difference method (Frame difference) to determine pedestrian target；

Step C014: use maximum variance between clusters to form the overall profile of pedestrian；

Step C015: use morphology processing image application opening operation, its effect is to eliminate small objects, more Very thin junction point separating objects, the border of smooth larger object, keep the area of object not change simultaneously；

Step C016: use the pedestrian that the segmentation of many people coordinate map segments method is connected, through coordinate map segments, in image Pedestrian area divided out, with minimum enclosed rectangle frame surround, formed rectangular target block；In these blocks, have and comprise list Individual pedestrian's, also have and comprise multiple pedestrian's contacted with each other；Multiple pedestrians contact, and define the merging in image (merge) problem, utilizes in block that target pixel points number is to estimate the number included in it, after determining many people block, The sub-block number of segmentation according to its area and width, will be determined；

Step C017: merging, separating treatment, merging, separation problem are to use the vertical style of shooting of video camera to carry out video A key issue in demographics；Can preferably process merging, separation problem, be the key of accurate statistics number, first First, when merging, separation situation occur, rely on many people block to split, the object block containing many people is divided into and comprises only a row The sub-block of people, and between sub-block, there is not overlap；Then, during following the tracks of, according to sub-block overlap coefficient as matching characteristic, Distinguish the sub-block combined；

Step C018: pedestrian counting method, this method is divided into two statisticals: one is the row in statistics current scene People's sum, another is to add up the number by counting line；First two statistical contents, will depend on drawing of statistical regions Point, use double-counting line to divide statistical regions, on the basis of two-wire three district, set up release areas；

(020) part, initializes escape configuration, specifically comprises the following steps that

Step C021: start, loads three-dimensional scenic, and according to the quantity such as number auto-initiation of photographic head detection Agent；

Step C022: arrange disaster type, disaster occurs position；

(030) part, simulation process process, specifically comprise the following steps that

Step C031: calculate the crowd density situation in each path, i.e. during crowd evacuation on each unit area The number flowed through；

Step C032: use dijkstra's algorithm to calculate each room beeline at a distance of nearest extra exit；

Step C033: calculate each room crowd time needed for the shortest path that C032 calculates, speed is room The average speed of interior personnel；

Step C034: application Monte Carlo model assesses the risk class of each extra exit, and assessment factor includes: in time The rate of change p of change number_c, safety measure rank m_s=for path/safety measure number；Escape way interval rank d_s =escape way distance/escape way number；d_fIt is at a distance of disaster distance；Q_rIt it is the value-at-risk of assessment；Q_r=m_s*d_s/d_f* p_r；

Step C035: application game theory determines best-effort path, and its revenue function reference factor includes: each room shortest path Footpath, the risk class of each extra exit, best-effort path calamity source grade, crowd density situation is distributed, according to revenue function value, Take Bayesian Nash equilibrium point, determine the escape route of each Agent；

Step C036: when crowding phenomenon occurs in extra exit, is pacified by other with the crowded crowd of game theory calculating section The revenue function that full outlet is evacuated or waited in crowded outlet, determine to select which kind of strategy escape, i.e. original place to wait with this or Other paths are selected to evacuate；

Step C037: evacuate escape simulation and terminate, acquisition process data；

(040) part, analyzes and processes process, specifically comprises the following steps that

Step C041: according to evacuating process data, calculate each escape opening crowd evacuation situation, i.e. number of evacuation；

Step C042: according to evacuating process data, calculate unsuccessfully Agent escape route of escaping；

Step C043: according to evacuating process data, unsuccessfully Agent is escaped in analysis can success escapeway line；

Step C044: output analysis result；

Beneficial effect

The present invention proposes a kind of optimization method based on game theoretic large-scale crowd evacuation, and the motility of the method is relatively Height, can accomplish the integration of local optimum and global optimization, reach optimum evacuation effect, can meet the whole process evacuated To precision and the requirement of efficiency；This method is the degree utilizing Monte Carlo model to analyze risk, rich according to value-at-risk application Play chess opinion and carry out global optimization, when the situation of getting congestion is that this method equally solves collision problem, reach the local evacuated Optimize；Additionally Evacuation uses Agent model can simulate the process of evacuation really so that generation strategy has practical valency Value；The present invention not only solves the collision problem of global optimization and local optimum, proposes have high efficiency evacuation simultaneously and optimizes calculation Method.

Accompanying drawing explanation

Fig. 1 device connection diagram (monitoring device, matrix switcher and control main frame)；

Fig. 2 photographic head working region；

Fig. 3 evacuation floor watch-dog layout；

Fig. 4 frame difference method threshold value comparative selection；

Fig. 5 merges, separates situation signal；

Fig. 6 merges, separates sub-block differentiation；

Fig. 7 walks in the same direction, the experimental result separate, merged；

Fig. 8 statistical regions divides；

Fig. 9 flow process of demographics judgement herein；

Detailed description of the invention:

First the video extraction background obtained according to photographic head, obtains mask, applies formula below；

The detection background change of cycle applications step C012, uses frame difference method to obtain pedestrian target simultaneously, applies formula below, Obtain effect as shown in Figure 4；

$R_k(x,y)=\left\{\begin{array}{cc}255,& \begin{array}{cc}if& |f_k(x,y)-f_{k-1}(x,y)|>T\end{array}\\ 0,& else\end{array}\right.$

Determine according to Fig. 4 applying step C014 to step C016 and to be partitioned into how many sub-blocks, perform step C017 and enter Row merges, lock out operation, and effect as shown in Fig. 5,6,7 finally performs step C018 of Part I, counts the number of people, and determines Personnel's direction of motion, calculates number and the room number in each camera head monitor region；

Running system simulation evacuation to process, system initializes Agent, loading scenario according to regional number, arranges calamity There is position in evil；First with room as ultimate unit, calculate the escape shortest path in each room, then escape out according to each Mouthful passenger distributing volume and the number amount of reaching, calculate formation time that each escape opening blocks up and block up the elimination time, gathering around in generation The all rooms or all regions that reach this emergency exit in stifled time range are considered as conflict, and application model Carlow model is as receipts The input parameter of benefit function, utilizes game theory to calculate conflict room or the escape route in region, and this step is as global optimum Escape route processes, and starts crowd evacuation, and each Agent is individual to be evacuated according to the attribute of self during escape, When there is jam situation, using game theory to solve conflict and block up collision problem, under jam situation, the participant of game is road Footpath, i.e. be select original place wait block up elimination or alternative select other paths；Evacuation terminates, and exports analysis result.

Claims (1)

1. generating system based on game theoretic large-scale crowd evacuation optimum escape route, application uses traditional Agent motion Model, blocks up conflict and Path Selection in conjunction with Monte Carlo model evaluation and test risk, the application game theory each Agent of solution, logical Cross the simulation to escape and evacuation process and analysis, produce the optimum escape scheme of the overall situation；

Equipment needed thereby: use 55 array monitoring camera-SLP-ZS39160D, VGA6464 computer VGA matrix switcher, PC, Haikang DVR DS-8016HS-S mono-, and for connecting the circuit of PC and embedded device；Photographic head cloth Office: photographic head working range is long 30 meters wide 6 meters of long regions；Photographic head arrangement principle is to cover whole corridor area, from main entrance Starting to lay, the first floor for building is arranged, other floors are arranged by this principle, one layer of totally 11 photographic head, and 5 floors are altogether Need 55 photographic head；

Method step is as follows:

(010) part, video surveillance specifically comprises the following steps that

Step C011: according to the video extraction background of photographic head shooting, obtains pattern mask；

According to threshold method, step C012: set time threshold x, judges whether background undergos mutation；If background is undergone mutation, set Putting the image after change is background, utilizes the impact that moving target is caused by the adaptivity of background model when initialization background Eliminate, then utilize adaptive RTS threshold adjustment algorithm, follow the tracks of background detail；

Step C013: pedestrian target detection uses frame difference method (Frame difference) to determine pedestrian target；

Step C014: use maximum variance between clusters to form the overall profile of pedestrian；

Step C015: use morphology processing image application opening operation, its effect is to eliminate small objects, the most very thin Junction point separating objects, the border of smooth larger object, keep the area of object not change simultaneously；

Step C016: use the pedestrian that the segmentation of many people coordinate map segments method is connected, through coordinate map segments, the row in image People region is divided out, surrounds with minimum enclosed rectangle frame, forms rectangular target block；In these blocks, have and comprise single row People's, also have and comprise multiple pedestrian's contacted with each other；Multiple pedestrians contact, and define the merging in image (merge) and ask Topic, utilize in block target pixel points number to estimate the number included in it, it is necessary to according to it after determining many people block Area and width, determine the sub-block number of segmentation；

Step C017: merging, separating treatment, merging, separation problem are to use the vertical style of shooting of video camera to carry out video number A key issue in statistics；Can preferably process merging, separation problem, be the key of accurate statistics number, first, When merging, separation situation occur, rely on many people block to split, the object block containing many people is divided into the son comprising only a pedestrian Overlap is there is not between block, and sub-block；Then, during following the tracks of, according to sub-block overlap coefficient as matching characteristic, distinguish The sub-block combined；

Step C018: pedestrian counting method, this method is divided into two statisticals: one is that the pedestrian in statistics current scene is total Number, another is to add up the number by counting line；First two statistical contents, will depend on the division of statistical regions, adopt Divide statistical regions with double-counting line, on the basis of two-wire three district, set up release areas；

(020) part, initializes escape configuration, specifically comprises the following steps that

Step C021: start, loads three-dimensional scenic, and according to quantity Agent such as number auto-initiation of photographic head detection；

Step C022: arrange disaster type, disaster occurs position；

(030) part, simulation process process, specifically comprise the following steps that

Step C031: calculate the crowd density situation in each path, i.e. flow through on each unit area during crowd evacuation Number；

Step C032: use dijkstra's algorithm to calculate each room beeline at a distance of nearest extra exit；

Step C033: calculate each room crowd time needed for the shortest path that C032 calculates, speed is people in room The average speed of member；

Step C034: application Monte Carlo model assesses the risk class of each extra exit, and assessment factor includes: change over The rate of change p of number_c, safety measure rank m_s=for path/safety measure number；Escape way interval rank d_s=peace Full tunnel distance/escape way number；d_fIt it is the distance at a distance of disaster center；Q_rIt it is the value-at-risk of assessment；Q_r=m_s*d_s/ d_f*p_c；

Step C035: application game theory determines best-effort path, and its revenue function reference factor includes: each room shortest path, respectively The risk class of extra exit, best-effort path calamity source grade, crowd density situation is distributed, according to revenue function value, takes shellfish This Nash Equilibrium point of leaf, determines the escape route of each Agent；

Step C036: when crowding phenomenon occurs in extra exit, is gone out safely by other with the crowded crowd of game theory calculating section The revenue function that mouth is evacuated or waited in crowded outlet, determines to select which kind of strategy to escape with this, i.e. original place waits or selects Other paths are evacuated；

Step C037: evacuate escape simulation and terminate, acquisition process data；

(040) part, analyzes and processes process, specifically comprises the following steps that

Step C041: according to evacuating process data, calculate each escape opening crowd evacuation situation, i.e. number of evacuation；

Step C042: according to evacuating process data, calculate unsuccessfully Agent escape route of escaping；

Step C043: according to evacuating process data, unsuccessfully Agent is escaped in analysis can success escape route；

Step C044: output analysis result.