Panic behavior recognition method and system based on panic semantic model and medium

Abstract

The invention relates to a panic behavior recognition method and system based on a panic semantic model, and a medium. The method comprises the following steps: calling an AI interface, recognizing an audio stream as semantic information in real time, performing hierarchical division through the semantic information and a basic concept, and recognizing key phrases; according to the positioning coordinate matrix, determining a specific level and a position of each key phrase; according to the specific hierarchy and position, using the description matrix to obtain the weight of the key phrase, carrying out consistency matching on the panic degree in the scene, and judging whether a panic behavior occurs or not; the panic semantic reasoning network model is constructed by the following steps: selecting a panic scene as a description object; defining a basic concept in the selected panic scene by using a network ontology language; supplementing knowledge elements in a panic scene to perfect a panic semantic model; and defining an inference rule, and constructing a panic semantic inference network model. Compared with the prior art, the accuracy of pedestrian panic behavior recognition is improved.

Description

Technical Field

[0001] The present invention relates to the field of panic behavior recognition, and in particular to a panic behavior recognition method, system and medium based on a panic semantic model. Background Art

[0002] Various public places are hosting more and more crowd gathering activities (such as transportation, religion, sports, business, culture and entertainment, etc.), and it is crucial to ensure the stability of the crowd. Pedestrian panic behavior is an important factor affecting crowd stability. How to identify pedestrian panic behavior is of great significance to the scientific management and emergency evacuation of crowds in public places. At present, the recognition of panic behavior mainly focuses on the recognition of abnormal postures of pedestrians. For complex group scenes, there are few related studies on the panic behavior recognition method through panic semantic reasoning network model.

[0003] There are still some deficiencies in the current research: 1) At present, most of the research on panic behavior recognition focuses on pedestrian abnormal posture recognition, and there are few related studies on identifying panic scenes through panic semantic models. 2) Panic scenes are diverse, and different panic scene knowledge elements have different weights in the recognition process of panic scenes, which will bring certain obstacles to the recognition of panic scenes. At present, there are few related studies on eliminating this obstacle to improve the accuracy of panic scene recognition. Summary of the invention

[0004] The purpose of the present invention is to overcome the defects of the above-mentioned prior art and to provide a panic behavior recognition method, system and medium based on a panic semantic model to improve the recognition accuracy of panic behavior.

[0005] The purpose of the present invention can be achieved by the following technical solutions:

[0006] A panic behavior recognition method based on a panic semantic model comprises the following steps:

[0007] Calling the AI ​​interface to recognize the audio stream as semantic information in real time, performing hierarchical division based on the semantic information and basic concepts, and identifying key phrases in the semantic information;

[0008] Determining the specific level and position of each of the key phrases in the panic semantic model according to the positioning coordinate matrix;

[0009] The weight of the key phrase is obtained according to the specific level and position of the key phrase using the description matrix of the panic semantic reasoning network model, and the panic degree in the scene is matched consistently to determine whether panic behavior occurs in the scene;

[0010] The steps for constructing the panic semantic reasoning network model are as follows:

[0011] Selecting a panic scene, taking the panic scene as a description object;

[0012] Using a web ontology language to define selected basic concepts of the panic scenario;

[0013] Supplementing the knowledge elements in the panic scenario to improve the panic semantic model in the panic scenario;

[0014] Based on the panic semantic model, inference rules are defined, and a panic semantic reasoning network model with the ability to analyze panic semantics and reason about panic events is constructed.

[0015] Furthermore, the semantic information includes shouting sounds, cries for help and conversation contents in a crowd.

[0016] Furthermore, the panic scenes include multiple types of medical disturbance scenes, natural disaster scenes and crowded scenes.

[0017] Furthermore, the knowledge elements in the panic scenario include multiple ones of key phrases, crowd status and occurrence conditions.

[0018] Furthermore, the hierarchical positioning coordinate matrix is ​​as follows:

[0019]

[0020] Where i = col r -1,col r Indicates the total length of the key paragraph, A i It means that when the total length of the segments is the same, only the last level is different. q+1 indicates the total number of keywords in the level where the key segment is located.

[0021] Furthermore, the description matrix of the panic semantic reasoning network model is as follows:

[0022]

[0023] In the formula, Indicates the position information of the q+1th key phrase in the hierarchy. Indicates the weight of the q+1th key phrase in the level,

[0024] Furthermore, the description matrix includes the key phrase information and weight information.

[0025] Furthermore, when the panic semantic reasoning network model matches a panic-related key phrase and the weighted structure exceeds a threshold, the matching result γ of the panic semantic reasoning network model is 1, otherwise, γ is 0.

[0026] According to another aspect of the present invention, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, a panic behavior recognition method based on a panic semantic model can be implemented.

[0027] According to another aspect of the present invention, a panic behavior recognition system based on a panic semantic model is provided, comprising:

[0028] A key phrase identification module is used to call the AI ​​interface to identify the audio stream as semantic information in real time, perform hierarchical division based on the semantic information and basic concepts, and identify key phrases in the semantic information;

[0029] A hierarchical positioning module, used to determine the specific level and position of each key phrase in the panic semantic model according to a positioning coordinate matrix;

[0030] A panic behavior judgment module is used to obtain the weight of the key segment according to the specific level and position of the key segment using the description matrix of the panic semantic reasoning network model, perform consistency matching on the panic degree in the scene, and judge whether panic behavior occurs in the scene;

[0031] The steps for constructing the panic semantic reasoning network model are as follows:

[0032] Selecting a panic scene, taking the panic scene as a description object;

[0033] Using a web ontology language to define selected basic concepts of the panic scenario;

[0034] Supplementing the knowledge elements in the panic scenario to improve the panic semantic model in the panic scenario;

[0035] Based on the panic semantic model, inference rules are defined, and a panic semantic reasoning network model with the ability to analyze panic semantics and reason about panic events is constructed.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] 1. When panic behavior occurs, the present invention uses real-time voice recognition to determine whether a disaster event, terrorist attack or other panic event occurs in the scene, and issues an early warning, thereby effectively avoiding the occurrence of panic behavior and increasing the accuracy of identifying panic behavior of pedestrians.

[0038] 2. The present invention divides the hierarchy by key semantic information and basic concepts, and greatly reduces the performance requirements of the computer for key segment recognition through the hierarchical design structure, thereby enhancing the immediacy of key segment detection.

[0039] 3. The present invention optimizes weights through statistical methods, which can accurately identify whether panic behavior occurs in a scene, and can judge other panic scenes by changing the weights of key phrases, and also provide new ideas for subsequent research on crowd stability analysis and prevention of the spread of panic behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 A flowchart of a panic behavior recognition method based on a panic semantic model is proposed in the present invention;

[0041] Figure 2 This is a schematic diagram of the structure of the panic semantic reasoning network model;

[0042] Figure 3 Derive the logistic graph for the description matrix of the panic semantic reasoning network model. DETAILED DESCRIPTION

[0043] The present invention is described in detail below in conjunction with the accompanying drawings and specific embodiments. This embodiment is implemented based on the technical solution of the present invention, and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0044] Example 1

[0045] This embodiment provides a panic behavior recognition method based on a panic semantic model. Figure 1 As shown, the following steps are included:

[0046] S1. Calling an AI interface to recognize the audio stream as semantic information in real time, performing hierarchical division based on the semantic information and basic concepts, and identifying key phrases in the semantic information.

[0047] This embodiment takes a wounding incident as an example and verifies the panic semantic reasoning network model through a real wounding incident video. The video records the panic incident caused by the wounding, including crowd chaos, cries for help, panic behavior, etc.

[0048] The audio in the video of the wounding incident is called to the Baidu AI interface to realize the conversion of speech to text, and the semantic information such as shouting, cries for help, and conversation content in the crowd is captured and recognized. This semantic information is input into the panic semantic reasoning network model. From the text obtained by speech recognition, the model recognizes key phrases related to panic scenes, including "cut people", "help", "run", "someone is injured", etc.

[0049] The hierarchical division is carried out through semantic information and basic concepts. The first field in the longer key semantic information is at a higher level. For example, in "I want to kill people" and "murder", the computer retrieves semantic information word by word (identified from the first word in sequence). Assuming that "I want to kill people" is located, "I" will first be located at the i-th layer, "want" will be found from the i+1th layer, and the subsequent "murder" will be found from the i+2th layer, while "murder" will be directly located at the i+2th layer, and the previous layers will be filled with 0 as the position code. Of course, due to the different initial positioning information, the degree of influence of "I want to kill people" and "murder" on whether panic behavior occurs in the scene will also be different. We found that such a design structure can greatly reduce the performance requirements of the computer and enhance the instant detection. When the computer recognizes a potential key phrase, if the subsequent fields are not matched, it is regarded as a non-key phrase, such as "I want to eat" is not regarded as key semantic information.

[0050] S2. Determine the specific level and position of each key phrase in the panic semantic model based on the positioning coordinate matrix.

[0051] The positioning coordinate matrix is ​​shown as follows:

[0052]

[0053] Where i = col r -1,col r Indicates the total length of the key paragraph, A i It means that when the total length of the segments is the same, only the last level is different. q+1 indicates the total number of keywords in the level where the key segment is located.

[0054] Based on the four panic scenarios of disaster events, medical incidents, stampedes, and terrorist attacks as case studies, a random survey was conducted. The total number of participants was 300, and the groups were divided as follows:

[0055] (1) Elderly group (under 60 years old): 62 people, including 29 males and 33 females;

[0056] (2) Middle-aged group (40 to 60 years old): 98 people, including 61 males and 37 females;

[0057] (3) Youth group (16 to 40 years old): 140 participants, including 65 males and 75 females.

[0058] Statistical methods were used to optimize the survey results and weights. The weight information table is shown in Table 1.

[0059] Table 1. Statistics of key phrase weights in semantic models

[0060]

[0061]

[0062] S3. According to the specific level and position of the key paragraphs, the description matrix of the panic semantic reasoning network model is used to obtain the weight of the key paragraphs, and the panic degree in the scene is matched consistently to determine whether panic behavior occurs in the scenario.

[0063] The steps to build the panic semantic reasoning network model are as follows:

[0064] 1) Select panic scenarios. First, consider whether the panic scenario is common. Try to select scenarios that occur frequently and are prone to panic, such as medical disturbances, natural disasters, and crowded scenes. Secondly, consider whether the panic scenario can be identified and prevented in advance. The construction of an inference network model for such scenarios has more practical application value. Finally, consider whether there is sufficient data support for the scenario for training and verification.

[0065] 2) Use OWL (Web Ontology Language) to define the basic concepts of the selected panic scenario, such as panic semantic expression is one of the sources of panic psychology, and panic psychology will induce the occurrence of panic behavior.

[0066] 3) Improve the panic semantic model in this scenario and supplement the knowledge elements in this panic scenario, including key phrases, crowd status and occurrence conditions.

[0067] 4) In order to enable the model to have reasoning capabilities, corresponding reasoning rules are defined. For example, in a crowded stampede, when identifying key sentences such as "I can't breathe", "Help", "Someone fell down", and "Someone was trampled to death", the panic level in the scene will be consistently judged based on the weight of the sentence. When it exceeds a certain threshold, it is judged that panic behavior has occurred in the scene.

[0068] The specific reasoning network model is as follows Figure 2 shown.

[0069] The description matrix of the panic semantic reasoning network model is as follows:

[0070]

[0071] In the formula, Indicates the position information of the q+1th key phrase in the hierarchy. Indicates the weight of the q+1th key phrase in the level,

[0072] The description matrix derivation logic diagram of the panic semantic reasoning network model is as follows Figure 3 shown.

[0073] The matching results of the panic semantic reasoning network model are as follows:

[0074] γ∈(0,1)

[0075] Where γ is the matching result of the panic semantic reasoning network model.

[0076] Furthermore, when the panic semantic reasoning network model matches a panic-related key phrase and the weighted structure exceeds a threshold, the matching result γ of the panic semantic reasoning network model is 1, otherwise, γ is 0.

[0077] In this embodiment, key phrases such as "cutting people" are identified, and the panic semantic reasoning network model determines the specific level and position of each key phrase in the panic semantic reasoning network model based on the positioning coordinate matrix. Then, based on the weight information provided in the description matrix, the impact of each key phrase on the panic level of the current scene is calculated. Key phrases with large weights have a greater impact on the judgment of panic scenes. By combining the weights and reasoning rules of multiple key phrases, the current scene is inferred and judged. The system matches multiple high-weight key phrases at the same time, reaching a certain threshold. The matching result of the panic semantic reasoning network model is γ=1, so it can be determined that panic behavior exists in the input video.

[0078] Example 2

[0079] This embodiment provides a computer-readable storage medium. A computer program is stored on the computer-readable storage medium. When the computer program is executed by a processor, a panic behavior recognition method based on a panic semantic model can be implemented.

[0080] The rest is the same as in Example 1.

[0081] Example 3

[0082] This embodiment provides a panic behavior recognition system based on a panic semantic model, including:

[0083] The key phrase recognition module is used to call the AI ​​interface to recognize the audio stream as semantic information in real time, divide the levels according to the semantic information and basic concepts, and recognize the key phrases in the semantic information;

[0084] A hierarchical positioning module is used to determine the specific level and position of each key segment in the panic semantic model according to the positioning coordinate matrix;

[0085] The panic behavior judgment module is used to obtain the weight of the key paragraphs according to the specific level and position of the key paragraphs using the description matrix of the panic semantic reasoning network model, to perform consistency matching on the panic degree in the scene, and to judge whether panic behavior occurs in the scene;

[0086] The steps for building the panic semantic reasoning network model are as follows:

[0087] Select panic scene and use panic scene as the description object;

[0088] A web ontology language is used to define the basic concepts of the selected panic scenarios;

[0089] Supplement the knowledge elements in panic scenarios to improve the panic semantic model in panic scenarios;

[0090] Based on the panic semantic model, inference rules are defined, and a panic semantic reasoning network model with the ability to analyze panic semantics and reason about panic events is constructed.

[0091] The rest is the same as in Example 1.

Claims

1. A panic behavior recognition method based on a panic semantic model, characterized in that: The following steps are involved: Calling the AI ​​interface to recognize the audio stream as semantic information in real time, performing hierarchical division based on the semantic information and basic concepts, and identifying key phrases in the semantic information; Determining the specific level and position of each of the key phrases in the panic semantic model according to the positioning coordinate matrix; The weight of the key phrase is obtained according to the specific level and position of the key phrase using the description matrix of the panic semantic reasoning network model, and the panic degree in the scene is matched consistently to determine whether panic behavior occurs in the scene; The steps for constructing the panic semantic reasoning network model are as follows: Selecting a panic scene, taking the panic scene as a description object; Using a web ontology language to define selected basic concepts of the panic scenario; Supplementing the knowledge elements in the panic scenario to improve the panic semantic model in the panic scenario; Based on the panic semantic model, inference rules are defined, and a panic semantic reasoning network model with the ability to analyze panic semantics and reason about panic events is constructed.

2. The panic behavior recognition method based on the panic semantic model according to claim 1 is characterized in that: The semantic information includes shouts from a crowd, cries for help, and various conversation contents.

3. The panic behavior recognition method based on the panic semantic model according to claim 1 is characterized in that: The panic scenes include multiple ones of medical disturbance scenes, natural disaster scenes and crowded scenes.

4. The panic behavior recognition method based on panic semantic model according to claim 1 is characterized in that: The knowledge elements in the panic scenario include multiple key phrases, crowd status and occurrence conditions.

5. The panic behavior recognition method based on panic semantic model according to claim 1 is characterized in that: The hierarchical positioning coordinate matrix is ​​as follows: Where i = col r -1,col r Indicates the total length of the key paragraph, A i It means that when the total length of the segments is the same, only the last level is different. q+1 indicates the total number of keywords in the level where the key segment is located.

6. The panic behavior recognition method based on panic semantic model according to claim 1 is characterized in that: The description matrix of the panic semantic reasoning network model is as follows: In the formula, Indicates the position information of the q+1th key phrase in the hierarchy. Indicates the weight of the q+1th key phrase in the level, 7. The panic behavior recognition method based on panic semantic model according to claim 6 is characterized in that: The description matrix includes the key phrase information and weight information.

8. The panic behavior recognition method based on panic semantic model according to claim 1 is characterized in that: When the panic semantic inference network model matches a panic-related key phrase and the weighted structure exceeds a threshold, the matching result γ of the panic semantic inference network model is 1, otherwise, γ is 0.

9. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the panic behavior recognition method based on a panic semantic model according to any one of claims 1 to 8 can be implemented.

10. A panic behavior recognition system based on a panic semantic model, characterized in that: include: A key phrase identification module is used to call the AI ​​interface to identify the audio stream as semantic information in real time, perform hierarchical division based on the semantic information and basic concepts, and identify key phrases in the semantic information; A hierarchical positioning module, used to determine the specific level and position of each key phrase in the panic semantic model according to a positioning coordinate matrix; A panic behavior judgment module is used to obtain the weight of the key segment according to the specific level and position of the key segment using the description matrix of the panic semantic reasoning network model, perform consistency matching on the panic degree in the scene, and judge whether panic behavior occurs in the scene; The steps for constructing the panic semantic reasoning network model are as follows: Selecting a panic scene, taking the panic scene as a description object; Using a web ontology language to define selected basic concepts of the panic scenario; Supplementing the knowledge elements in the panic scenario to improve the panic semantic model in the panic scenario; Based on the panic semantic model, inference rules are defined, and a panic semantic reasoning network model with the ability to analyze panic semantics and reason about panic events is constructed.

Images