CN111080027A - Dynamic escape guiding method and system - Google Patents
Dynamic escape guiding method and system Download PDFInfo
- Publication number
- CN111080027A CN111080027A CN201911365696.7A CN201911365696A CN111080027A CN 111080027 A CN111080027 A CN 111080027A CN 201911365696 A CN201911365696 A CN 201911365696A CN 111080027 A CN111080027 A CN 111080027A
- Authority
- CN
- China
- Prior art keywords
- path
- escape
- dynamic
- picture
- real time
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 43
- 238000012549 training Methods 0.000 claims abstract description 57
- 238000004422 calculation algorithm Methods 0.000 claims abstract description 40
- 239000011159 matrix material Substances 0.000 claims description 19
- 238000013528 artificial neural network Methods 0.000 claims description 9
- 239000000779 smoke Substances 0.000 claims description 8
- 238000012544 monitoring process Methods 0.000 claims description 7
- 238000013508 migration Methods 0.000 claims description 6
- 230000005012 migration Effects 0.000 claims description 6
- 241000287107 Passer Species 0.000 claims description 2
- 230000008569 process Effects 0.000 description 9
- 238000013527 convolutional neural network Methods 0.000 description 6
- 238000013461 design Methods 0.000 description 6
- 238000005516 engineering process Methods 0.000 description 5
- 241000257303 Hymenoptera Species 0.000 description 4
- 238000013135 deep learning Methods 0.000 description 4
- 238000004364 calculation method Methods 0.000 description 3
- 230000001965 increasing effect Effects 0.000 description 3
- 238000012545 processing Methods 0.000 description 3
- 230000005540 biological transmission Effects 0.000 description 2
- 238000004891 communication Methods 0.000 description 2
- 230000007547 defect Effects 0.000 description 2
- 238000010586 diagram Methods 0.000 description 2
- 230000006872 improvement Effects 0.000 description 2
- 238000004088 simulation Methods 0.000 description 2
- 206010063385 Intellectualisation Diseases 0.000 description 1
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008859 change Effects 0.000 description 1
- 230000001419 dependent effect Effects 0.000 description 1
- 230000000694 effects Effects 0.000 description 1
- 230000002708 enhancing effect Effects 0.000 description 1
- 230000006870 function Effects 0.000 description 1
- 230000002068 genetic effect Effects 0.000 description 1
- 238000009434 installation Methods 0.000 description 1
- 238000012986 modification Methods 0.000 description 1
- 230000004048 modification Effects 0.000 description 1
- 239000000178 monomer Substances 0.000 description 1
- 238000003062 neural network model Methods 0.000 description 1
- 238000005457 optimization Methods 0.000 description 1
- 239000002245 particle Substances 0.000 description 1
- 238000011176 pooling Methods 0.000 description 1
- 230000003068 static effect Effects 0.000 description 1
- 238000013526 transfer learning Methods 0.000 description 1
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06Q—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES; SYSTEMS OR METHODS SPECIALLY ADAPTED FOR ADMINISTRATIVE, COMMERCIAL, FINANCIAL, MANAGERIAL OR SUPERVISORY PURPOSES, NOT OTHERWISE PROVIDED FOR
- G06Q10/00—Administration; Management
- G06Q10/04—Forecasting or optimisation specially adapted for administrative or management purposes, e.g. linear programming or "cutting stock problem"
- G06Q10/047—Optimisation of routes or paths, e.g. travelling salesman problem
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/08—Learning methods
Landscapes
- Engineering & Computer Science (AREA)
- Business, Economics & Management (AREA)
- Physics & Mathematics (AREA)
- Human Resources & Organizations (AREA)
- Theoretical Computer Science (AREA)
- Economics (AREA)
- General Physics & Mathematics (AREA)
- Strategic Management (AREA)
- Entrepreneurship & Innovation (AREA)
- Biomedical Technology (AREA)
- Quality & Reliability (AREA)
- Tourism & Hospitality (AREA)
- Marketing (AREA)
- General Business, Economics & Management (AREA)
- Game Theory and Decision Science (AREA)
- Development Economics (AREA)
- Health & Medical Sciences (AREA)
- Life Sciences & Earth Sciences (AREA)
- Artificial Intelligence (AREA)
- Operations Research (AREA)
- Biophysics (AREA)
- Computational Linguistics (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Computing Systems (AREA)
- General Engineering & Computer Science (AREA)
- Mathematical Physics (AREA)
- Software Systems (AREA)
- Alarm Systems (AREA)
Abstract
The invention belongs to the technical field of safe evacuation, and relates to a dynamic escape guidance method and a dynamic escape guidance system. The method comprises an off-line training stage and an on-line dynamic escape guiding stage; in the off-line training stage, a path safety state identification model is obtained based on image identification; the on-line dynamic escape guiding stage acquires pictures of a traffic area in real time, inputs the path safety state identification model, judges whether a path is safe, and changes the weight of a corresponding unsafe path into infinity in an undirected graph abstracted from a scene where the path is located if the path is unsafe; and (3) carrying out shortest path planning for avoiding unsafe paths on the modified undirected graph in real time based on a shortest path algorithm, and dynamically carrying out escape guidance in real time according to the planned shortest paths. The invention can realize the real-time dynamic solution of the shortest escape path and carry out escape guidance according to the shortest escape path, thereby solving the technical problem that the prior art is difficult to carry out the escape guidance in real time.
Description
Technical Field
The invention belongs to the technical field of safe evacuation, and particularly relates to a dynamic escape guidance method and a dynamic escape guidance system, which are suitable for real-time and dynamic escape guidance in a large-scale transportation junction and can also be suitable for real-time and dynamic escape guidance in a general scene.
Background
In recent years, large transportation junction sites such as a plurality of underground complexes are built in each large city, and the sites have the characteristics of large internal space, complex structure, high personnel density and the like. With the rapid increase of the number of the traffic junctions, the gradual expansion of the scale and the increasingly complex internal structure, the defects of long evacuation path and complex evacuation path are directly caused, and further the evacuation action of people is disordered, the evacuation time is long, and the evacuation efficiency is low. In the face of emergencies such as fire, terrorist attack, trampling, earthquake and the like, people put forward new higher requirements on the escape guidance system.
In order to solve the problems, some students propose to utilize an ant colony algorithm to directly perform escape path simulation calculation by taking the number of people as a dependent variable, so that different escape paths are directly planned according to different numbers of people. The method has a good effect of planning in advance for places with fixed number of people in a small range. However, the ant colony algorithm has a slow convergence speed, is easy to fall into local optimum, generally needs a long search time, and is prone to have a stagnation phenomenon, that is, after the search is performed to a certain degree, solutions found by all individuals are completely consistent, a solution space cannot be further searched, and a better solution cannot be found, so that the method has the following problems for a real-time escape route of a large-scale traffic junction place:
(1) because the number of people in the large-scale traffic hub site is dynamically changed in real time and cannot be counted in real time, the difference between the number of people in the peak period and the number of people in the low peak period is huge, the initial number of ants in the ant colony algorithm training process is fixed, and all ants need to be traversed, the number of the ants directly determines the algorithm solving efficiency, the more the number is, the slower the solving is, the worse the real-time performance is, and the real-time path planning based on the number of people by adopting the ant colony algorithm is difficult to adapt to the actual scene requirements of the large-scale traffic hub;
(2) because the large-scale traffic hub places have numerous channels and the number of people is uncontrollable, when a danger comes, especially when an accident occurs on a escaping route, partial channels and spaces are probably unusable due to certain problems, such as fire blockage, collapsed object blockage, crowd messy congestion or treading, and the like. The essence of the ant colony algorithm is to simulate ant habits, take the path selected by the most ants as the optimal path, and easily guide crowds to gather on the same escape path to cause congestion and risk.
(3) In addition, because the global optimal path planning based on the ant colony algorithm belongs to the prior planning, most of the adopted emergency evacuation sign lamps exist in a static and invariable monomer form according to the pre-planned path, and during emergency evacuation, an escape blind area often appears, even a misleading phenomenon is caused.
All the reasons make the escape route of a large-scale transportation junction place difficult to plan and adjust dynamically in real time.
Therefore, a method and a system for real-time and dynamic evacuation and escape guidance are needed to realize the purpose of safely, accurately and rapidly guiding people to escape.
Disclosure of Invention
Aiming at the defects or the improvement requirements in the prior art, the invention provides a dynamic escape guidance method and a dynamic escape guidance system, and aims to monitor and recognize the available states of spaces and channels in a traffic junction in real time based on an image recognition technology of deep learning, dynamically correct the weight of the corresponding path in a scene undirected graph according to the available states, then directly utilize a shortest path algorithm to quickly solve the shortest escape path, realize the real-time dynamic solution of the shortest escape path, and carry out escape guidance according to the shortest escape path, thereby solving the technical problem that the prior art is difficult to dynamically carry out escape guidance in real time.
To achieve the above object, according to one aspect of the present invention, there is provided a dynamic escape guidance method, including an offline training phase and an online dynamic escape guidance phase:
in the off-line training stage, a neural network is trained by adopting at least one combination of a flame picture and a fireless picture, an manned picture and an unmanned picture, a person falling, a congestion picture and a person normal traffic picture, and a collapse picture and a collapse and congestion picture and a normal scene picture to obtain a path safety state identification model;
the online dynamic escape guiding stage comprises the following steps:
step 1: acquiring pictures of the passing area in real time, inputting the path safety state identification model, judging whether a path is safe or not, and if a certain path is unsafe, turning to the step 2;
step 2: in an undirected graph abstracted from the scene, the weight of the corresponding unsafe path is changed to be infinite in real time;
and step 3: and (3) carrying out shortest path planning for avoiding unsafe paths on the undirected graph modified in the step (2) in real time based on a shortest path algorithm, and dynamically carrying out escape guidance in real time according to the planned shortest paths.
Furthermore, the neural network performs migration training by adopting the existing underlying network, so that a path safety state recognition model is rapidly obtained.
Furthermore, the learning rate of the underlying network is adjusted to be small, and the learning rate of the upper network is adjusted to be large.
Further, in step 1, the identification mode that the congestion of the people causes unsafe route adopts at least one of the following modes:
(1) if the person picture and the unmanned picture are adopted for training in the off-line training stage, recording the number of the passers identified by the path safety state identification model in unit time in a certain passing area in step 1, comparing the number with the maximum passing capacity designed in the corresponding passing area, and determining that the path of the passing area is unsafe if the maximum passing capacity is exceeded;
(2) if the person falling picture and the person normal passing picture are adopted for training in the offline training stage, in the step 1, the corresponding path is regarded as unsafe when a person falls down;
(3) if the person congestion picture and the person normal traffic picture are adopted for training in the offline training stage, in the step 1, whether the person congestion exists or not is directly identified.
Further, the undirected graph in step 2 is abstractly obtained as follows: abstracting all intersections needing path selection into points, connecting adjacent points according to paths capable of being passed under normal conditions to form an undirected graph, and setting the weight of the connecting line as the distance between two adjacent intersections, so that the undirected graph comprises node numbers of all intersections from a starting point to a terminal point, the distance between the intersections and the passable paths between the intersections.
Further, the escape starting point position in the undirected graph is represented by a decimal matrix s, the escape end point position is represented by a decimal matrix t, the distance between the starting point and the end point is represented by a decimal matrix weight, and the shortest path algorithm is a dijkstra algorithm; and (3) updating the decimal matrix weights in real time, and quickly calculating the shortest safe path bypassing the unsafe path by using a dijkstra algorithm.
Further, step 1 also includes collecting flame and smoke information from temperature and/or smoke sensors in the scene, so as to judge the path safety state more accurately and comprehensively.
In order to achieve the purpose, the invention also provides a dynamic escape guidance method, when no dangerous case occurs, taking each intersection as a starting point, planning a plurality of globally optimal escape routes leading to the same or different exits for each intersection in advance;
during emergency evacuation, a plurality of pre-planned escape routes are all opened or multiple paths are opened to disperse escape personnel, meanwhile, the dynamic escape guiding method is adopted, the safety of the routes between each intersection node on each escape route is monitored in real time, and on the basis of the pre-planned global optimal escape route, secondary planning and guiding of the escape routes are dynamically carried out in real time according to road condition changes, so that unsafe routes are bypassed.
In order to achieve the above object, the present invention further provides a dynamic escape guidance system, which includes a processor, a path safety state identification model, and a dynamic escape guidance program module;
the path safety state recognition model is obtained by off-line training in an off-line training stage in the dynamic escape guidance method according to any one of the previous items;
when the dynamic escape guidance program module is called by the processor, the steps 1-3 of the dynamic escape guidance method are executed.
Further, comprising: the camera, the indicator light and the wireless connection module;
the cameras are distributed in a monitoring area corresponding to the escape route, are used for shooting images in a monitoring range and upload the images to the processor through the wireless connection module;
the plurality of indicator lamps are distributed in the escape passage and receive the pointing instruction of the processor through the wireless connection module;
and the processor issues pointing instructions to the indicating lamps in real time according to the path planning result, so that dynamic planning and guidance of the escape path are realized.
In general, compared with the prior art, the above technical solution contemplated by the present invention can obtain the following beneficial effects:
(1) the method of the invention judges the path safety based on the off-line training path safety state recognition model, and modifies the scene undirected graph in real time aiming at the unsafe path, thereby matching with the shortest path algorithm, avoiding the dangerous path in real time and selecting a new safe path, realizing the real-time dynamic planning and guidance of the escape path, having the characteristics of intellectualization, dynamism and real-time, being suitable for various large, medium and small scenes, and being especially suitable for large traffic hubs with complex paths and escape personnel conditions.
(2) Compared with the non-basic training, the off-line training efficiency can be improved based on the migration training, and the path safety state recognition model with high accuracy can be quickly obtained.
(3) Due to the fact that the hidden layer design is increased, a large number of training sets are needed, overfitting is prone to happen more and more, and massive computing resources are needed; however, if the design of the hidden layer is less, the accuracy is not high, and under-fitting is easy to occur, so that on the basis of the migration training, the use number of the training set can be reduced and the training efficiency can be further improved on the premise of not reducing the accuracy by further reducing the learning rate of the underlying network and increasing the learning rate of the upper network.
(4) The multiple different path safety state identification modes can be used independently aiming at the use properties of different channels, can also be matched for use, and are more flexible and comprehensive in identification.
(5) The scene is abstracted into an undirected graph based on a graph theory method, an entrance and an exit and various feasible paths in the scene can be simply and visually displayed, the subsequent path calculation process is simplified, the complexity of the dijkstra algorithm is low, the calculation delay is low, the near-real-time path planning can be realized, and the escape efficiency is improved.
(6) The system of the invention is based on an image processing technology, automatically distinguishes whether a space and a channel in a traffic junction are available in real time through deep learning of a computer, further plans an optimal path, controls a display terminal through a wireless or wired data link (preferably the wireless data link to avoid monitoring failure caused by line damage), and finally displays an escape direction or a route and the like on a terminal screen, thereby rapidly and correctly guiding people to escape from an accident site. In addition, due to the adoption of an image recognition mode, the monitoring cameras in the transportation junction can be directly compatible and called (namely, no new other monitoring equipment is required to be added). Therefore, the system has the outstanding characteristics of novel conception, simple structure, economy, practicability, strong robustness, high automation degree and the like.
Drawings
FIG. 1 is a block flow diagram of the system of the present invention;
FIG. 2 is a schematic diagram of the structure of a migration training neural network employed in the present invention;
FIG. 3 is a presentation of an undirected graph abstracted based on graph theory, using a subterranean complex as an example;
FIG. 4 is an undirected graph example in a simplified scenario;
fig. 5 is a schematic view of a camera layout, in which the positions of triangles represent camera mounting positions, the left side triangle is a camera 1, and the right side triangle is a camera 2;
FIG. 6 is a schematic of the shortest path from the departure point 8 to the destination point 6 when the camera 1 detects a fire in the aisle 9 → 10; inf represents that the weight of the corresponding channel is infinite;
FIG. 7 is a schematic of the shortest path from the departure point 8 to the destination point 6 when the camera 2 detects a fire in the aisle 5 → 6;
fig. 8 is a schematic of the shortest path from the departure point 8 to the destination point 6 when two cameras simultaneously detect a fire.
Detailed Description
In order to make the objects, technical solutions and advantages of the present invention more apparent, the present invention is described in further detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention. In addition, the technical features involved in the embodiments of the present invention described below may be combined with each other as long as they do not conflict with each other.
The invention utilizes a deep learning algorithm to identify and classify images, and mainly aims to identify the conditions of open fire, smoke, local building collapse, channel blockage, people crowding and the like, and then dynamically generate an escape path according to the information to guide people to escape.
The Convolutional Neural Network (CNN) is the most common and effective deep learning network for image classification, and as shown in fig. 2, local features are mainly extracted through convolution operation, and then a plurality of low-order features are integrated to identify overall features. Of course, in actual practice, for better and more accurate classification purposes, it is preferable that the present invention also adds many different types of layers to enhance the accuracy of their discrimination, such as an image input layer, a convolutional layer, a full-pooling layer, a ReLU layer, a softmax layer, a dropout layer, a full-link layer, an image output layer, and so on.
However, the following difficulties still exist in the application process: due to the fact that the hidden layer design is increased, a large number of training sets are needed, overfitting is prone to happen more and more, and massive computing resources are needed. However, if the design of the hidden layer is small, the accuracy is not high, and under-fitting is easy to occur. By combining the two points, the invention selects a transfer learning mode, applies the trained underlying network (such as AlexNet, vgg and GoogLeNet), modifies the rear full-connection layer and the output layer at the same time, slows down the learning rate of the previous layers (i.e. the underlying network), and accelerates the learning rate of the last layers (i.e. the upper network). In this way, we only need a relatively small training set to get better training results.
Based on the above concept, as shown in fig. 1, a dynamic escape guidance method according to a preferred embodiment of the present invention includes an offline training phase and an online dynamic escape guidance phase.
In the off-line training stage, the neural network is trained by adopting at least one combination of a flame picture and a fireless picture, an image with people and an image without people, a person falling down picture, a congestion picture and a person normal passing picture, and a collapse picture, a collapse and congestion picture and a normal scene picture, so as to obtain a path safety state recognition model. Preferably, the neural network performs migration training by using an existing underlying network, so as to quickly obtain a path security state recognition model. Preferably, the learning rate of the underlying network is adjusted to be small, while the learning rate of the upper network is adjusted to be large.
The online dynamic escape guiding stage comprises the following steps:
step 1: and (3) acquiring pictures of the passing area in real time, inputting the path safety state identification model, judging whether the path is safe or not, and if the path is not safe, turning to the step 2.
Preferably, the identification mode of unsafe paths caused by people jam adopts at least one of the following modes:
(1) if the images with people and the images without people are adopted for training in the off-line training stage, the number of people passing through the traffic area identified by the path safety state identification model in unit time is recorded in step 1, the number is compared with the maximum traffic capacity designed in the corresponding traffic area, and if the maximum traffic capacity is exceeded, the path of the traffic area is regarded as unsafe.
(2) If the person falling picture and the person normal passing picture are adopted for training in the offline training stage, in the step 1, the corresponding path is regarded as unsafe when a person falls.
(3) If the person congestion picture and the person normal traffic picture are adopted for training in the offline training stage, in the step 1, whether the person congestion exists or not is directly identified.
Step 2: in an undirected graph abstracted from the scene, the weight of the corresponding unsafe path is changed to be infinite in real time; the undirected graph is abstractly obtained as follows: abstracting all intersections needing path selection into points, connecting adjacent points according to paths capable of being passed under normal conditions to form an undirected graph, and setting the weight of the connecting line as the distance between two adjacent intersections, so that the undirected graph comprises node numbers of all intersections from a starting point to a terminal point, the distance between the intersections and the passable paths between the intersections.
Preferably, the escape starting point position in the undirected graph is represented by a decimal matrix s, the escape ending point position is represented by a decimal matrix t, and the distance between the starting point and the ending point is represented by a decimal matrix weight.
The shortest path algorithm comprises a Dijkstra algorithm, a Bellman-Ford algorithm, a Floyd algorithm, an SPFA algorithm and the like. The optimal shortest path algorithm is dijkstra algorithm; and updating the weight of the corresponding path in the decimal matrix weights in real time according to the path safety state identification result, and then quickly calculating the shortest safety path bypassing the unsafe path by using a dijkstra algorithm.
And step 3: and (3) carrying out shortest path planning for avoiding unsafe paths on the undirected graph modified in the step (2) in real time based on a shortest path algorithm, and dynamically carrying out escape guidance in real time according to the planned shortest paths.
The method and system of the present invention are described in more detail below using a more specific example.
An off-line training stage: in a specific case, a convolutional neural network with a 25-layer architecture is adopted, and the structural hierarchy of the convolutional neural network is shown as follows:
because the training of the convolutional neural network is completed off line, the specific network structure is not particularly limited, and an engineer can build a brand new network structure by himself to perform foundation-free training as long as the online use is not influenced.
Taking flame recognition as an example, the actual training result of the neural network model is as follows:
iterative training achievement table
It can be seen that from the 250 th iteration, the flame identification accuracy reaches 100%, and the training efficiency and accuracy are extremely high. Other training ideas such as personnel identification, collapse blockage (blockage caused by local collapse of a building) identification, collapse identification, congestion identification, fall identification and the like are similar and are not repeated.
With respect to path planning: the key point of the online dynamic escape lies in the planning of the path, and the basis of the path planning is necessarily all known entrances and exits and passages inside the building structure, so the abstraction and simplification of the building structure are introduced firstly:
abstracting all intersections needing path selection in the building into points, forming an undirected graph by connecting the points, wherein the weight of a connecting line is the distance between the two intersections to complete the digitization of the path information of the building, taking an underground complex as an example, the process of abstracting the intersections into the undirected graph is shown in figure 3, and the weight is omitted in the graph. In an undirected graph, any intersection can be used as a starting point or an end point, so that the path planning mode is very flexible.
In the process of path abstraction simplification, a computer can grasp the core characteristics of a building, ignore useless information irrelevant to the path and skillfully convert a three-dimensional building into a plane or three-dimensional dot-line graph in a graph theory mode. Therefore, the method is suitable for all large, medium and small scenes and even large and complex transportation hubs, especially for the simulation of increasingly complex underground spaces and ultra-large buildings at present. When the path planning is needed, the map matrixes of different buildings are constructed according to the structural characteristics of the different buildings.
For ease of illustration and understanding, this example selects a simpler planar undirected graph to replace the underground complex of FIG. 3, as shown in FIG. 4. Fig. 4 is represented as follows using a digitizing matrix:
s=[1 1 1 2 2 2 3 3 4 5 5 6 6 6 7 8 9 9 10 11 11 13 14]
t=[2 3 4 3 5 8 4 6 7 6 9 7 10 11 12 9 10 14 11 12 13 15 15]
weights=[30 10 30 30 20 20 30 20 20 20 20 20 10 20 20 10 10 10 10 2020 20 20]
in the matrix, the escape starting point position in the undirected graph is represented by a matrix s, the escape end point position is represented by a matrix t, and the distance between the starting point and the end point is represented by a matrix weight. If the elements in s, t and weights are s in sequence from left to righti、ti、weightsiWeights are theniDenotes siTo tiI is 1, 2, …, 23.
The undirected graph under the normal state of the building is generated and stored in advance, and then can be directly called in the on-line dynamic escape guidance stage.
An online dynamic escape guiding stage:
step 1: taking the arrangement of the cameras in fig. 5 as an example (generally, cameras can be arranged in all channels, and this example only takes two cameras as an example for simple description), acquiring pictures of the path 9 → 10 and the path 5 → 6 in real time through the camera 1 and the camera 2, respectively, inputting the path security state identification model, and determining whether the path is secure, if a certain path is not secure, going to step 2;
step 2: in an undirected graph abstracted from the scene, the weight of the corresponding unsafe path is changed to infinity in real time, and as shown in FIGS. 6-8, the unsafe path is marked by inf;
and step 3: and (3) carrying out shortest path planning for avoiding unsafe paths on the undirected graph modified in the step (2) in real time based on a shortest path algorithm, and dynamically carrying out escape guidance in real time according to the planned shortest paths, wherein the Dijkstra algorithm is taken as an example below.
In the dynamic escape guidance process, the system is generally required to timely and effectively complete information collection, processing and feedback, so that the ant colony algorithm is not used for calculating the shortest path in the system, the situation that the escaper cannot timely give a correct path indication due to too low processing speed is avoided, and the Dijkstra algorithm is selected to be used, so that the system can be ensured to quickly complete path planning and indication. During specific operation, Dijkstra algorithm can be realized through MATLAB software, and planning of an optimal path is completed.
As shown in fig. 5 to 8, the case of the path 9 → 10 and the path 5 → 6 being separately and simultaneously impassable, and the shortest escape path which is solved in the corresponding case and needs to bypass the unsafe path when reaching the intersection 6 from the intersection 8. When a camera in a certain channel detects a fire, the channel length will immediately become infinite (Inf shown in the figure), i.e. cannot pass through. Meanwhile, the computer calculates the shortest and optimal path (the escape path shown by the thick line arrow in the figure) to bypass the accident section according to the updated weight for people to escape.
In other embodiments, in order to ensure that the escape route can be globally optimized as much as possible, the invention also provides a dynamic escape guidance method which is combined with the characteristic that offline training can be carried out when the scene is completely normal, and can plan a plurality of globally optimal escape routes leading to the same or different exits for each intersection in advance by taking each intersection as a starting point when no dangerous case occurs. At this time, the method for planning the globally optimal escape route is not limited, and any algorithm may be used, for example, various genetic algorithms, particle swarm algorithms, ant colony algorithms, shortest path algorithms, and the like. The global optimal escape route does not necessarily require the absolute shortest escape distance, and can also be shortest and second shortest, so that the escape routes of crowds can be reasonably dispersed as soon as possible, and the crowds are reduced in congestion.
During emergency evacuation, a plurality of planned escape routes are all opened or multiple paths are opened to disperse escape personnel, meanwhile, the method of the preferred embodiment is adopted to monitor the safety of the paths among all intersection nodes on each escape route in real time, so that the secondary planning and guidance of the escape routes are dynamically carried out in real time according to road condition changes on the basis of the planned global optimal escape route, and therefore the unsafe routes are bypassed.
The mode of combining offline global optimization and online dynamic planning guidance can dynamically guide people to escape in real time, and can keep the global optimality of escape routes as much as possible in emergency, so that the escape probability is comprehensively improved.
The invention also provides a dynamic escape guidance system which comprises a processor, a path safety state identification model, a dynamic escape guidance program module, a camera, a display terminal and a wireless connection module. The path safety state recognition model is obtained through off-line training; and when the dynamic escape guidance program module is called by the processor, the dynamic escape guidance program module performs dynamic planning guidance of an escape path. The wireless connection module can be a 5G network module, a wifi module, a Bluetooth module and the like.
Preferably, the display terminal adopts an Arduino controlled LED dot matrix. Arduino is a single chip microcomputer for the purpose of opening sources. On one hand, the wireless data transmission module can be connected to form a wireless communication link to communicate with an upper computer; on the other hand, the LED dot matrix can be controlled to generate dynamic indication marks. Arduino may also be powered using a battery. Even if the power is cut off in the building and the communication line is damaged, the system can still work normally. When accidents such as fire disasters occur in a large-scale transportation junction, the camera captures the occurrence condition of the fire disasters in time, transmits the fire disasters to a central computer to process information and plan an optimal escape path, and then utilizes Arduino to receive the information through a wireless data transmission module and control the on-off of each display point in the LED display screen so that the LED display screen presents arrows or other information for guiding the correct escape direction.
In other embodiments, the pictures obtained by photographing the camera inside the building can be used as a training set to increase the accuracy and the applicability of neural network recognition, and especially in the fire drill and the escape drill processes, the training samples suitable for the current scene are very easy to collect. And can also infer the position of stranded crowd through categorised discernment portrait and corresponding camera mounted position, upload the fire control center with information automatically simultaneously, supplementary emergency personnel salvage.
In other embodiments, camera image recognition can be used as a main part, and meanwhile, the temperature sensor and the smoke sensor are used as auxiliary parts to collect more comprehensive information, so that the safety state of the path can be judged more accurately, and path planning can be performed more reasonably.
In other embodiments, the total evacuation time can be minimized by comprehensively considering the channel capacity and the evacuation speed from the pure pursuit of the shortest path plan, for example, the number of the channel persons is determined by counting the persons through the typical characteristics (such as the head) of the human body based on the image recognition technology, and compared with the maximum channel capacity, and the time of the escape person appearing at different designated positions (which can be referred to the installation position of the camera) is recorded in real time based on the face recognition technology (also realized through the convolutional neural network offline training), so that the evacuation speed is calculated, and then the crowd is guided to the path with small channel capacity and high evacuation speed.
In other embodiments, the display terminal for safety guidance may also be a display screen or a ground projection lamp, a holographic projection device, etc., rather than simply relying on an LED lamp to indicate a path. For example, the real-time disaster situation of the building can be simply and intuitively reflected and the graph can be drawn directly through a wall-mounted display screen or a ground projection lamp, holographic projection and the like, and a voice reminding function can be set, so that the situation that people neglect the display terminal due to factors such as confusion and smoke in the escape process is avoided.
In general, the dynamic escape guidance system has strong practicability, takes practical application as guidance, completes all designs from theoretical concepts and algorithms to terminals, and has strong real scene adaptability. The invention firstly proposes to use the image recognition technology to carry out dynamic evacuation. The existing escape guidance systems mostly adopt complex smoke and temperature sensors to identify fire, use of existing cameras is omitted, and the conditions that the paths are invalid due to people jam, building collapse and the like cannot be identified. The invention introduces the image recognition based on the deep neural network into the system, thereby not only reducing the use of the sensor, but also being directly connected to the existing camera in the scene for use, improving the efficiency, simplifying the system and enhancing the reliability. In addition, the image recognition is very scalable and can be adapted to a variety of different tasks, such as: the ability of positioning crowds with congestion, identifying the road sections blocked by the collapsed objects and the like. The invention also establishes the abstract map of the building by grasping the key information in the building and utilizing the related knowledge concept of the graph theory, thereby facilitating the dynamic design of path planning by combining the change of the path safety state.
It will be understood by those skilled in the art that the foregoing is only a preferred embodiment of the present invention, and is not intended to limit the invention, and that any modification, equivalent replacement, or improvement made within the spirit and principle of the present invention should be included in the scope of the present invention.
Claims (10)
1. A dynamic escape guidance method is characterized by comprising an off-line training stage and an on-line dynamic escape guidance stage:
in the off-line training stage, a neural network is trained by adopting at least one combination of a flame picture and a fireless picture, an manned picture and an unmanned picture, a person falling, a congestion picture and a person normal traffic picture, and a collapse picture and a collapse and congestion picture and a normal scene picture to obtain a path safety state identification model;
the online dynamic escape guiding stage comprises the following steps:
step 1: acquiring pictures of the passing area in real time, inputting the path safety state identification model, judging whether a path is safe or not, and if a certain path is unsafe, turning to the step 2;
step 2: in an undirected graph abstracted from the scene, the weight of the corresponding unsafe path is changed to be infinite in real time;
and step 3: and (3) carrying out shortest path planning for avoiding unsafe paths on the undirected graph modified in the step (2) in real time based on a shortest path algorithm, and dynamically carrying out escape guidance in real time according to the planned shortest paths.
2. The dynamic escape guidance method of claim 1, wherein the neural network is migration trained using an existing underlying network, thereby rapidly obtaining the path safety state recognition model.
3. The dynamic escape guidance method according to claim 2, wherein the learning rate of the underlying network is adjusted to be small, and the learning rate of the overlying network is adjusted to be large.
4. A dynamic escape guidance method according to any one of claims 1 to 3, wherein in step 1, the identification of the unsafe route due to the congestion of the person is performed by at least one of the following methods:
(1) if the person picture and the unmanned picture are adopted for training in the off-line training stage, recording the number of the passers identified by the path safety state identification model in unit time in a certain passing area in step 1, comparing the number with the maximum passing capacity designed in the corresponding passing area, and determining that the path of the passing area is unsafe if the maximum passing capacity is exceeded;
(2) if the person falling picture and the person normal passing picture are adopted for training in the offline training stage, in the step 1, the corresponding path is regarded as unsafe when a person falls down;
(3) if the person congestion picture and the person normal traffic picture are adopted for training in the offline training stage, in the step 1, whether the person congestion exists or not is directly identified.
5. A dynamic escape guidance method according to any one of claims 1 to 3, wherein the undirected graph in step 2 is abstracted as follows: abstracting all intersections needing path selection into points, connecting adjacent points according to paths capable of being passed under normal conditions to form an undirected graph, and setting the weight of the connecting line as the distance between two adjacent intersections, so that the undirected graph comprises node numbers of all intersections from a starting point to a terminal point, the distance between the intersections and the passable paths between the intersections.
6. The dynamic escape guidance method according to claim 5, wherein the escape starting point position in the undirected graph is represented by a decimal matrix s, the escape ending point position is represented by a decimal matrix t, the distance between the starting point and the ending point is represented by a decimal matrix weights, and the shortest path algorithm used is dijkstra algorithm; and (3) updating the decimal matrix weights in real time, and quickly calculating the shortest safe path bypassing the unsafe path by using a dijkstra algorithm.
7. A dynamic escape guidance method according to any one of claims 1 to 3, wherein the step 1 further comprises receiving information of flame and smoke from a temperature and/or smoke sensor in the scene, so as to judge the safety state of the path more accurately and comprehensively.
8. A dynamic escape guidance method is characterized in that when no dangerous case occurs, a plurality of global optimal escape routes leading to the same or different exits are planned for each intersection in advance by taking each intersection as a starting point;
during emergency evacuation, a plurality of pre-planned escape routes are opened completely or in multiple ways to disperse evacuees, meanwhile, the dynamic escape guidance method according to any one of claims 1 to 7 is adopted to monitor the safety of the routes between the intersection nodes on each escape route in real time, so that on the basis of the pre-planned global optimal escape route, secondary planning and guidance of the escape routes are carried out dynamically in real time according to road condition changes, and therefore unsafe routes are bypassed.
9. A dynamic escape guidance system is characterized by comprising a processor, a path safety state identification model and a dynamic escape guidance program module;
the path safety state recognition model is obtained by performing off-line training in an off-line training stage in the dynamic escape guidance method according to any one of claims 1 to 7;
when the dynamic escape guidance program module is called by the processor, the steps 1-3 of the dynamic escape guidance method according to any one of claims 1-7 are executed.
10. The dynamic escape guidance system of claim 9, comprising: the camera, the display terminal and the wireless or wired connection module;
the cameras are distributed in a monitoring area corresponding to the escape route, are used for shooting images in a monitoring range and upload the images to the processor through the wireless connection module;
the display terminals are distributed in the escape passage and receive the pointing instruction of the processor through the wireless or wired connection module;
and the processor issues pointing instructions to each display terminal in real time according to the path planning result, so that dynamic planning and guidance of the escape path are realized.
Priority Applications (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911365696.7A CN111080027A (en) | 2019-12-26 | 2019-12-26 | Dynamic escape guiding method and system |
Applications Claiming Priority (1)
Application Number | Priority Date | Filing Date | Title |
---|---|---|---|
CN201911365696.7A CN111080027A (en) | 2019-12-26 | 2019-12-26 | Dynamic escape guiding method and system |
Publications (1)
Publication Number | Publication Date |
---|---|
CN111080027A true CN111080027A (en) | 2020-04-28 |
Family
ID=70318060
Family Applications (1)
Application Number | Title | Priority Date | Filing Date |
---|---|---|---|
CN201911365696.7A Pending CN111080027A (en) | 2019-12-26 | 2019-12-26 | Dynamic escape guiding method and system |
Country Status (1)
Country | Link |
---|---|
CN (1) | CN111080027A (en) |
Cited By (6)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112006678A (en) * | 2020-09-10 | 2020-12-01 | 齐鲁工业大学 | Electrocardiogram abnormity identification method and system based on combination of AlexNet and transfer learning |
CN112461253A (en) * | 2020-11-05 | 2021-03-09 | 深圳慧拓无限科技有限公司 | Emergency evacuation path planning method, device, medium and electronic equipment |
CN113485206A (en) * | 2021-08-06 | 2021-10-08 | 时代云英(重庆)科技有限公司 | Extensible Internet of things system and method |
CN114067507A (en) * | 2021-11-08 | 2022-02-18 | 深圳正为格智能科技有限公司 | Intelligent fire control sign control system adapting to environment for transformation |
CN114550624A (en) * | 2021-08-18 | 2022-05-27 | 成都华迈通信技术有限公司 | Escape guiding system for riot and terrorist environment |
CN114627601A (en) * | 2022-02-23 | 2022-06-14 | 广东交通职业技术学院 | Subway passenger emergency escape guiding method, system, device and storage medium |
Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102011083023A1 (en) * | 2011-09-20 | 2013-03-21 | Robert Bosch Gmbh | Evacuation system for planning escape routes in a building, procedures and computer program |
CN103776452A (en) * | 2014-01-13 | 2014-05-07 | 华中科技大学 | Personnel characteristic-based fire evacuation path navigation method |
CN105561492A (en) * | 2014-11-07 | 2016-05-11 | 开利公司 | Dynamic acquisition terminal for behavior statistical information of humans as well as evacuation system and method |
CN106845679A (en) * | 2016-12-12 | 2017-06-13 | 梁健宁 | A kind of emergency evacuation automated intelligent generates system |
CN108692728A (en) * | 2018-04-26 | 2018-10-23 | 哈尔滨工业大学深圳研究生院 | Indoor navigation method based on CAD architectural drawings and Computer Vision Recognition and system |
CN109101694A (en) * | 2018-07-16 | 2018-12-28 | 山东师范大学 | A kind of the crowd behaviour emulation mode and system of the guidance of safe escape mark |
CN109543631A (en) * | 2018-11-28 | 2019-03-29 | 公安部沈阳消防研究所 | A kind of fire image detection alarm method based on machine learning |
-
2019
- 2019-12-26 CN CN201911365696.7A patent/CN111080027A/en active Pending
Patent Citations (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
DE102011083023A1 (en) * | 2011-09-20 | 2013-03-21 | Robert Bosch Gmbh | Evacuation system for planning escape routes in a building, procedures and computer program |
CN103776452A (en) * | 2014-01-13 | 2014-05-07 | 华中科技大学 | Personnel characteristic-based fire evacuation path navigation method |
CN105561492A (en) * | 2014-11-07 | 2016-05-11 | 开利公司 | Dynamic acquisition terminal for behavior statistical information of humans as well as evacuation system and method |
CN106845679A (en) * | 2016-12-12 | 2017-06-13 | 梁健宁 | A kind of emergency evacuation automated intelligent generates system |
CN108692728A (en) * | 2018-04-26 | 2018-10-23 | 哈尔滨工业大学深圳研究生院 | Indoor navigation method based on CAD architectural drawings and Computer Vision Recognition and system |
CN109101694A (en) * | 2018-07-16 | 2018-12-28 | 山东师范大学 | A kind of the crowd behaviour emulation mode and system of the guidance of safe escape mark |
CN109543631A (en) * | 2018-11-28 | 2019-03-29 | 公安部沈阳消防研究所 | A kind of fire image detection alarm method based on machine learning |
Non-Patent Citations (1)
Title |
---|
张斌 等: "基于多路径负载均衡的智能楼宇火灾应急疏散策略研究", 《研究与发展》 * |
Cited By (7)
Publication number | Priority date | Publication date | Assignee | Title |
---|---|---|---|---|
CN112006678A (en) * | 2020-09-10 | 2020-12-01 | 齐鲁工业大学 | Electrocardiogram abnormity identification method and system based on combination of AlexNet and transfer learning |
CN112461253A (en) * | 2020-11-05 | 2021-03-09 | 深圳慧拓无限科技有限公司 | Emergency evacuation path planning method, device, medium and electronic equipment |
CN113485206A (en) * | 2021-08-06 | 2021-10-08 | 时代云英(重庆)科技有限公司 | Extensible Internet of things system and method |
CN114550624A (en) * | 2021-08-18 | 2022-05-27 | 成都华迈通信技术有限公司 | Escape guiding system for riot and terrorist environment |
CN114550624B (en) * | 2021-08-18 | 2024-02-06 | 成都华迈通信技术有限公司 | Escape guiding system |
CN114067507A (en) * | 2021-11-08 | 2022-02-18 | 深圳正为格智能科技有限公司 | Intelligent fire control sign control system adapting to environment for transformation |
CN114627601A (en) * | 2022-02-23 | 2022-06-14 | 广东交通职业技术学院 | Subway passenger emergency escape guiding method, system, device and storage medium |
Similar Documents
Publication | Publication Date | Title |
---|---|---|
CN111080027A (en) | Dynamic escape guiding method and system | |
CN110047240B (en) | Intelligent fire-fighting warning guidance system and method based on Internet of things | |
Gu et al. | A metaverse-based teaching building evacuation training system with deep reinforcement learning | |
CN109185743B (en) | Intelligent fire-fighting emergency lamp system | |
CN105404231A (en) | Internet of things-based intelligent building monitoring managing system | |
CN114353804B (en) | Fire emergency evacuation path planning method and device, intelligent terminal and storage medium | |
CN109697807A (en) | A kind of Internet of Things escaping guidance system of intelligence fire behavior detection analysis | |
CN1963878A (en) | Intelligence inspection prewarning forecasting apparatus for fire of high-rise building | |
CN115752490B (en) | Safe trip path optimization method and system based on big data and positioning technology | |
CN115545359B (en) | Dynamic intelligent evacuation method and device for complex building fire | |
Liu et al. | A survey of crowd evacuation on passenger ships: Recent advances and future challenges | |
CN113947014A (en) | BIM-based tunnel fire emergency rescue method, terminal and storage medium | |
CN116402243A (en) | Emergency evacuation path planning method, device, terminal and storage medium | |
CN112488401A (en) | Fire escape route guiding method and system | |
CN114972639A (en) | Fire monitoring system suitable for wisdom garden | |
Bahamid et al. | Intelligent robot-assisted evacuation: A review | |
CN115018283A (en) | Intelligent fire-fighting management method for industrial park | |
CN116805444A (en) | Distributed named navigation pedestrian driving early warning method, system and device | |
CN105096661B (en) | Air traffic Workflow Management System and method based on vertical sectional view | |
KR102592101B1 (en) | A system and methods for guiding evacuation route based on artificial intelligence | |
Sun et al. | Study on safe evacuation routes based on crowd density map of shopping mall | |
Zhang et al. | An optimized multisource bilinear convolutional neural network model for flame image identification of coal mine | |
CN115544267A (en) | Knowledge graph-based simulation dynamic verification method for multiple emergency treatments | |
CN112150040B (en) | Scenic spot integrated dispatching system | |
CN114067065A (en) | Escape route visualization system and escape method |
Legal Events
Date | Code | Title | Description |
---|---|---|---|
PB01 | Publication | ||
PB01 | Publication | ||
SE01 | Entry into force of request for substantive examination | ||
SE01 | Entry into force of request for substantive examination | ||
RJ01 | Rejection of invention patent application after publication |
Application publication date: 20200428 |
|
RJ01 | Rejection of invention patent application after publication |