Group behavior identification method based on video data

Abstract

The invention provides a group behavior recognition method based on video data, and relates to the technical field of group behavior recognition, and the method comprises the steps: obtaining a video image set; extracting behavior characteristics of targets from the video image, wherein the targets comprise biological objects and non-biological objects; determining behavior characteristics of the group according to the behavior characteristics of the biological object; determining an influence weight of the biological object on group behaviors through behavior characteristics of the biological object and the group to obtain a first weight; tracking the target by using a tracking algorithm to obtain a behavior trajectory of the target; determining the influence weight of the non-biological object on the group behavior through the behavior track of the target, and obtaining a second weight; and obtaining group behavior identification data through the behavior characteristics of the target, the first weight and the second weight. The method not only considers biological objects, but also considers non-biological objects, improves the comprehensiveness of group behavior recognition, and improves the recognition precision of group behaviors.

Description

Technical Field

[0001] The present invention relates to the technical field of group behavior recognition, and in particular to a group behavior recognition method based on video data. Background Art

[0002] Group behavior recognition has been widely used in social places such as sports game video analysis, surveillance video recognition, and social behavior understanding. By identifying group behavior in videos, dangerous incidents can be prevented.

[0003] Currently, most efforts are devoted to constructing interactive relationships between individuals in multi-person scenarios to infer group behavior. However, in actual scenarios, not only individual relationships affect group behavior, but external interference events also affect group behavior. Due to the complexity of group relationships, each individual and external interference event have different degrees of influence on group behavior. If only individual characteristics are used to infer group behavior, the recognition accuracy will be affected. Summary of the invention

[0004] The purpose of the present invention is to provide a group behavior recognition method based on video data, and the technical problem to be solved is how to improve the recognition accuracy of group behavior.

[0005] The present invention is achieved through the following technical solutions:

[0006] A method for group behavior recognition based on video data comprises the following steps:

[0007] S100, obtaining a group behavior video, sampling the group behavior video, and obtaining a video image set consisting of a plurality of video images;

[0008] S200, extracting the behavioral characteristics of the target from the video image, the target including biological objects and non-biological objects; wherein a group is composed of a plurality of the biological objects, and determining the behavioral characteristics of the group according to the behavioral characteristics of the biological objects;

[0009] Determine the influence weight of the biological object on the group behavior through the above-mentioned behavior characteristics of the biological object and the group, and obtain a first weight;

[0010] S300, tracking the target using a tracking algorithm to obtain a behavior trajectory of the target;

[0011] Through the behavioral trajectory of the above targets, the influence weight of the non-biological objects on the group behavior is determined to obtain the second weight;

[0012] S400: Obtain group behavior recognition data through the above-mentioned target's behavior characteristics, the first weight and the second weight.

[0013] By evaluating the impact of biological objects and non-biological objects on group behavior, not only the behavioral characteristics of biological objects are taken into account, but also the impact of non-biological objects on group behavior, which improves the comprehensiveness of group behavior recognition and thus improves the recognition accuracy of group behavior. In addition, the tracking algorithm is used to obtain the target's behavior trajectory and understand the target's behavior pattern in time and space, which helps to identify the dynamic changes of group behavior.

[0014] Furthermore, in S200, the behavior characteristics of the group are determined according to the behavior characteristics of the biological objects, and the specific steps include:

[0015] S211, classifying the behavioral characteristics of the biological objects to obtain biological behavior categories;

[0016] S212: Count the number of biological objects in the above biological behavior categories, and sort the biological behavior categories from large to small according to the number of biological objects to obtain a biological behavior sequence. ;in, represents the first biological behavior category, Indicates Biobehavioral categories, , Indicates Biobehavioral Behavioral characteristics of biological objects;

[0017] S213, selecting the first biological behavior category from the above biological behavior sequence The behavioral characteristics of the group are used as the behavioral characteristics of the group.

[0018] Through classification, biological objects with similar behavioral characteristics are grouped together to identify different behavioral patterns, which helps to grasp the overall behavior of the group in the future; by counting the number of biological objects in each category, the main behavioral patterns in the group are identified, and the behavioral category with the largest number often represents the mainstream behavior of the group; the biological behavior sequence obtained by sorting by quantity not only highlights the mainstream behavior, but also reflects the relative importance of other behaviors, which helps to grasp the hierarchy of group behavior more accurately in subsequent analysis; selecting the behavioral characteristics of the most numerous behavioral categories as the behavioral characteristics of the group can reflect the overall behavioral trend of the group to the greatest extent. In a dynamically changing group, the mainstream behavior may change with time and environmental changes. Through real-time statistics and sorting, the behavioral characteristics of the group can be dynamically updated, thereby enhancing the adaptability of recognition.

[0019] Furthermore, in S200, the influence weight of the biological object on the group behavior is determined based on the above-mentioned behavior characteristics of the biological object and the group. The specific steps include:

[0020] S221. The behavioral characteristics of the biological objects include movements, postures, and positions in the group;

[0021] S222, extracting a group distribution range from the video image, dividing the group distribution range into a decision area, a following area, and a non-following area, and assigning corresponding weights to the decision area, the following area, and the non-following area;

[0022] S223, determining an additional weight of the biological object according to the position of the biological object in the group;

[0023] S224, comparing the posture of the biological object with that of the group, if the postures are consistent, then the posture consistency number of the biological object is increased by 1; otherwise, the posture inconsistency number of the biological object is increased by 1; comparing the action of the biological object with that of the group, if the actions are consistent, then the action consistency number of the biological object is increased by 1; otherwise, the action inconsistency number of the biological object is increased by 1;

[0024] Traversing the video image set, determining the posture consistency rate of the biological object and the group based on the posture consistency number and posture inconsistency number of the biological object; determining the action consistency rate of the biological object and the group based on the action consistency number and action inconsistency number of the biological object;

[0025] The behavior consistency rate between the biological object and the group is determined through the above-mentioned posture consistency rate and action consistency rate;

[0026] S225. Determine the influence weight of the biological object on the group behavior through the additional weight of the biological object and the behavior consistency rate.

[0027] An additional weight is assigned to the biological object according to its position in the group. The more important the position, the higher the additional weight. By comparing postures and movements, the behavioral consistency between the biological object and the group is determined. If the behavior of an individual is highly consistent with the group behavior and can drive the group to act, then the biological object has a higher influence.

[0028] Furthermore, in S300, the influence weight of the non-biological object on the group behavior is determined through the behavior trajectory of the above target, and the specific steps include:

[0029] S311, obtaining the behavior trajectories of the above-mentioned non-biological objects and the behavior trajectories of the biological objects in the same period, predicting the behavior trajectories of the non-biological objects in the next period and the behavior trajectories of the biological objects when they are not affected by the non-biological objects; determining the population distribution range in the next period according to the predicted behavior trajectories of the above-mentioned biological objects;

[0030] S312, determining the position of the non-biological object passing through the group distribution range in the next time period according to the predicted behavior trajectory and group distribution range of the non-biological object;

[0031] Determine the additional weight of the abiotic object according to the position of the abiotic object in the distribution range of the population;

[0032] S313, determining the movement speed of the non-living object according to the obtained behavior trajectory and time period of the non-living object; wherein the non-living object is provided with a plurality of speed intervals, and each speed interval is assigned a corresponding weight;

[0033] Determining a speed weight of the non-living object according to the speed interval of the movement speed of the non-living object;

[0034] S314. Determine the influence weight of the non-living object on the group behavior through the additional weight and speed weight of the non-living object.

[0035] The prediction, spatial analysis and weight distribution are combined to determine the weight of the impact of non-biological objects on group behavior. First, collect the behavior trajectories of non-biological objects and biological objects in the same time period; use machine learning or physical models to predict the behavior trajectory of non-biological objects in the next time period, and at the same time, predict the behavior trajectory of biological objects when they are not affected by non-biological objects. According to the predicted behavior trajectory of biological objects, determine the group distribution range of the next time period (that is, identify the area where biological objects may gather in the next time period). Then, according to the predicted behavior trajectory of non-biological objects, determine the location of the group distribution range that they will pass through in the next time period, and evaluate the possible impact of the behavior of non-biological objects on group behavior, that is, the additional weight. According to the obtained behavior trajectory of non-biological objects and the duration of the time period, calculate their movement speed, and determine their speed weight according to the speed range of the movement speed of non-biological objects. Finally, by combining the additional weight and speed weight of non-biological objects, calculate their influence weight on group behavior.

[0036] Furthermore, the behavior consistency rate between the above biological object and the group is calculated using the following formula:

[0037] ,

[0038] in, Indicates The rate of consistency between the behavior of the biological object and the group; Indicates The rate of agreement between the biological object and the group’s posture; Indicates The number of pose congruences between the biological object and the group; Indicates The number of posture inconsistencies between biological objects and the group; Indicates The consistency rate of the actions of the biological object and the group; Indicates The number of consistent actions between the biological object and the group; Indicates The number of inconsistent actions between biological objects and groups; represents the weight of posture in the behavior consistency rate; Represents the weight of the action in the behavior consistency rate.

[0039] Furthermore, the influence weight of the above biological objects on group behavior is calculated using the following formula:

[0040] ,

[0041] in, Indicates The weight of the biological object's influence on group behavior; represents the additional weight of the biological object; represents the decision area weight; Indicates the weight of the follow area; Indicates the non-following area weight.

[0042] Furthermore, the additional weight of the above non-biological objects in the next period is calculated using the following formula:

[0043] ,

[0044] in, Indicates the T+1 period Additional weights for inanimate objects; represents the decision area weight; Indicates the weight of the follow area; Indicates the T+1 period The area of ​​the decision region covered by the inanimate objects; Represents the total area of ​​the decision-making region during the T+1 period; Indicates the T+1 period The area of ​​the following region covered by the non-biological object; Indicates the total area of ​​the follow-up region during the T+1 period.

[0045] Furthermore, the following formula is used to calculate the movement speed of the above non-biological object:

[0046] ,in, Indicates The speed of movement of inanimate objects; Indicates the T period behavioral trajectories of inanimate objects; Indicates the duration of the T period;

[0047] Judging from the above The speed interval to which the non-biological object's movement speed belongs is retrieved, and the corresponding speed weight is obtained. Velocity weight for non-mob objects.

[0048] Furthermore, the following formula is used to calculate the influence weight of the above non-biological objects on group behavior:

[0049] ,

[0050] Indicates The weight of the impact of non-biological objects on group behavior; Indicates Velocity weight for non-mob objects.

[0051] Compared with the prior art, the present invention has the following advantages and beneficial effects:

[0052] By evaluating the impact of biological objects and non-biological objects on group behavior, not only the behavioral characteristics of biological objects are taken into account, but also the impact of non-biological objects on group behavior, which improves the comprehensiveness of group behavior recognition and thus improves the recognition accuracy of group behavior. In addition, the tracking algorithm is used to obtain the target's behavior trajectory and understand the target's behavior pattern in time and space, which helps to identify the dynamic changes of group behavior. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] In order to more clearly illustrate the technical solutions of the exemplary embodiments of the present invention, the following briefly introduces the drawings required for use in the embodiments. It should be understood that the following drawings only illustrate certain embodiments of the present invention and should not be regarded as limiting the scope. For those of ordinary skill in the art, other relevant drawings can be obtained based on these drawings without creative work. In the drawings:

[0054] Figure 1 Main flow chart. DETAILED DESCRIPTION

[0055] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with embodiments and drawings. The exemplary embodiments of the present invention and their description are only used to explain the present invention and are not intended to limit the present invention.

[0056] First embodiment:

[0057] Combination Figure 1 A first aspect provides a method for group behavior recognition based on video data, comprising the following steps:

[0058] S100, obtaining a group behavior video, sampling the group behavior video, and obtaining a video image set consisting of a plurality of video images;

[0059] S200, extracting the behavioral characteristics of the target from the video image, the target including biological objects and non-biological objects; wherein a group is composed of a plurality of the biological objects, and determining the behavioral characteristics of the group according to the behavioral characteristics of the biological objects;

[0060] Determine the influence weight of the biological object on the group behavior through the above-mentioned behavior characteristics of the biological object and the group, and obtain a first weight;

[0061] The biological objects mentioned above can be people, fish, birds, etc., and the non-biological objects can be buildings, vehicles, balls, etc. By determining the influence weight of biological objects on group behavior, the accuracy of recognition can be improved; for example, in a crowd, the behavior of the decision maker has a significant impact on the behavior of the entire group.

[0062] S300, tracking the target using a tracking algorithm to obtain a behavior trajectory of the target;

[0063] Through the behavioral trajectory of the above targets, the influence weight of the non-biological objects on the group behavior is determined to obtain the second weight;

[0064] Consider the impact of non-biological objects (such as road layout, building location, etc.) on group behavior to improve the comprehensiveness of recognition. For example, in traffic scenes, road layout can significantly affect the flow pattern of vehicles and people.

[0065] S400, obtaining group behavior recognition data through the behavior characteristics, first weight and second weight of the above target. Comprehensively consider the information of biological objects and non-biological objects to improve the recognition accuracy of group behavior. For example, in a complex scene, the behavior of a single biological object may be difficult to accurately identify, but after combining the information of its non-biological object, the behavioral intention and group behavior pattern of the biological object can be determined.

[0066] By evaluating the impact of biological objects and non-biological objects on group behavior, not only the behavioral characteristics of biological objects are taken into account, but also the impact of non-biological objects on group behavior, which improves the comprehensiveness of group behavior recognition and thus improves the recognition accuracy of group behavior. In addition, the tracking algorithm is used to obtain the target's behavior trajectory and understand the target's behavior pattern in time and space, which helps to identify the dynamic changes of group behavior.

[0067] A usage scenario for reference is the group activity monitoring scenario in a public square, which includes both biological objects (crowds) and non-biological objects (mobile public facilities, such as vending machines, mobile screens, etc.). The video data of this scenario is used to identify and analyze group behavior, such as the flow direction of the crowd, aggregation patterns, reactions to specific events, etc.; by identifying the behavioral characteristics of biological objects (crowds) and non-biological objects (public facilities) and their impact on group behavior, we can understand group dynamics and optimize the management and safety measures of public places.

[0068] Specifically, the surveillance camera records a day's worth of activity video on a public square, and extracts a frame from the video every 5 seconds to form a video image set consisting of hundreds of frames, which includes the location and status of the crowd and public facilities on the square in different time periods; the movement speed of each crowd in each frame is calculated, and the areas where the crowd gathers are identified through spatial density analysis; the overall flow direction is determined based on the movement direction of the majority of the crowd, and the relationship between the change in crowd speed and the group flow direction is analyzed to obtain the first weight; if the crowd speed increases and the direction is consistent, it is considered that the crowd has a greater impact on the group behavior and is given a higher weight.

[0069] Use target detection and tracking algorithms (such as YOLO+SORT) to track people and public facilities between consecutive frames to obtain their behavior trajectories. By analyzing the trajectory, record the number of people in front of the vending machine and their usage in each frame, and detect whether the content played on the screen attracts people to stop. If a large number of people gather in front of the vending machine, it indicates that it has a certain impact on group behavior. Weights are assigned according to the number of people gathered and the time to obtain the second weight; if a mobile screen attracts people to stay for a long time, its weight should be increased accordingly.

[0070] Combine the behavioral characteristics, first weight and second weight of biological objects and non-biological objects for comprehensive analysis; if the direction of group flow is mainly affected by the speed of the crowd (high first weight), and the vending machine only affects the distribution of the crowd in a small range (low second weight), then the group behavior recognition data will mainly reflect the flow pattern of the crowd. Generate a report containing the group behavior recognition results, including the direction of group flow, gathering hotspots, and the speed of response to specific events (such as promotions, emergencies), etc., to provide decision support for public place management.

[0071] In volleyball, the trajectory of the volleyball usually determines the position and behavioral tendencies of the players, and plays an important role in judging the group behavior of athletes.

[0072] Second embodiment:

[0073] On the basis of the first embodiment, in S200, the behavior characteristics of the group are determined according to the behavior characteristics of the biological objects, and the specific steps include:

[0074] S211, classifying the behavioral characteristics of the biological objects to obtain biological behavior categories;

[0075] S212: Count the number of biological objects in the above biological behavior categories, and sort the biological behavior categories from large to small according to the number of biological objects to obtain a biological behavior sequence. ;in, represents the first biological behavior category, Indicates Biobehavioral categories, , Indicates Biobehavioral Behavioral characteristics of biological objects;

[0076] S213, selecting the first biological behavior category from the above biological behavior sequence The behavioral characteristics of the group are used as the behavioral characteristics of the group.

[0077] Through classification, biological objects with similar behavioral characteristics are grouped together to identify different behavioral patterns, which helps to grasp the overall behavior of the group in the future; by counting the number of biological objects in each category, the main behavioral patterns in the group are identified, and the behavioral category with the largest number often represents the mainstream behavior of the group; the biological behavior sequence obtained by sorting by quantity not only highlights the mainstream behavior, but also reflects the relative importance of other behaviors, which helps to grasp the hierarchy of group behavior more accurately in subsequent analysis; selecting the behavioral characteristics of the most numerous behavioral categories as the behavioral characteristics of the group can reflect the overall behavioral trend of the group to the greatest extent. In a dynamically changing group, the mainstream behavior may change with time and environmental changes. Through real-time statistics and sorting, the behavioral characteristics of the group can be dynamically updated, thereby enhancing the adaptability of recognition.

[0078] Third embodiment:

[0079] On the basis of any of the above embodiments, in S200, the influence weight of the biological object on the group behavior is determined by the above-mentioned behavior characteristics of the biological object and the group. The specific steps include:

[0080] S221. The behavioral characteristics of the biological objects include movements, postures, and positions in the group;

[0081] S222, extracting a group distribution range from the video image, dividing the group distribution range into a decision area, a following area, and a non-following area, and assigning corresponding weights to the decision area, the following area, and the non-following area;

[0082] S223, determining an additional weight of the biological object according to the position of the biological object in the group;

[0083] S224, comparing the posture of the biological object with that of the group, if the postures are consistent, then the posture consistency number of the biological object is increased by 1; otherwise, the posture inconsistency number of the biological object is increased by 1; comparing the action of the biological object with that of the group, if the actions are consistent, then the action consistency number of the biological object is increased by 1; otherwise, the action inconsistency number of the biological object is increased by 1;

[0084] Traversing the video image set, determining the posture consistency rate of the biological object and the group based on the posture consistency number and posture inconsistency number of the biological object; determining the action consistency rate of the biological object and the group based on the action consistency number and action inconsistency number of the biological object;

[0085] The behavior consistency rate between the biological object and the group is determined through the above-mentioned posture consistency rate and action consistency rate;

[0086] S225. Determine the influence weight of the biological object on the group behavior through the additional weight of the biological object and the behavior consistency rate.

[0087] An additional weight is assigned to the biological object according to its position in the group. The more important the position, the higher the additional weight. By comparing postures and movements, the behavioral consistency between the biological object and the group is determined. If the behavior of an individual is highly consistent with the group behavior and can drive the group to act, then the biological object has a higher influence.

[0088] A reference usage scenario is the on-site monitoring scenario of a large music festival, in which biological objects are mainly audiences participating in the music festival, and non-biological objects include the stage, audio equipment, etc. Monitoring video data is used to identify and analyze the behavioral characteristics of the audience and their impact on group behavior in order to optimize on-site management and improve the audience experience. At the music festival, the audience moves in different areas such as the stage, in front of the stage, and in the rest area. The actions (such as dancing, walking, standing), postures (such as front, side, back to the stage) and positions in the group (such as the stage, in front of the stage, and rest area) of each audience are extracted from the video images; through spatial density analysis, high-density areas (such as the stage) are identified as decision-making areas, medium-density areas (such as in front of the stage) are followed areas, and low-density areas (such as rest areas) are non-followed areas. According to the position of the audience in the group, the audience in the decision-making area has a higher additional weight, followed by the following area, and the non-following area has the lowest. In addition, the influence of the audience in the group can also be considered, such as the additional weight of obvious leaders (such as those holding flags or leading slogans).

[0089] Traverse the video image set, and for each audience member, compare their posture with the surrounding group. If the posture is similar (such as dancing facing the stage), then the posture consistency number is increased by 1; otherwise, the posture inconsistency number is increased by 1. Similarly, compare the movements of the audience and the group. If the movements are the same (such as dancing to the rhythm of the music), then the movement consistency number is increased by 1; otherwise, the movement inconsistency number is increased by 1. The posture consistency rate is calculated based on the posture consistency number and the posture inconsistency number, and the movement consistency rate is calculated based on the movement consistency number and the movement inconsistency number. Then, the two are combined to obtain the behavior consistency rate of the biological object and the group.

[0090] The influence weight of the biological object on the group behavior is calculated by combining the additional weight of the biological object and the behavior consistency rate. For example, an audience located in the decision-making area with highly consistent behavior will have a greater impact on the group behavior, so its influence weight is higher.

[0091] In terms of application, by identifying audiences with high influence weights, we can give priority to their needs and behaviors. If the consistency rate of audience behavior in a certain area suddenly drops, it may mean that an emergency has occurred or the audience is emotionally fluctuating, and timely measures must be taken to ensure safety. By analyzing the audience's behavioral characteristics and their impact on group behavior, we can adjust the on-site layout, event arrangements, etc. to improve the audience's overall experience.

[0092] In a specific embodiment, the behavior consistency rate between the biological object and the group is calculated using the following formula:

[0093] ,

[0094] in, Indicates The rate of consistency between the behavior of the biological object and the group; Indicates The rate of agreement between the biological object and the group’s posture; Indicates The number of pose congruences between the biological object and the group; Indicates The number of posture inconsistencies between biological objects and the group; Indicates The consistency rate of the actions of the biological object and the group; Indicates The number of consistent actions between the biological object and the group; Indicates The number of inconsistent actions between biological objects and groups; represents the weight of posture in the behavior consistency rate; Represents the weight of the action in the behavior consistency rate.

[0095] In a specific embodiment, the influence weight of the above biological objects on group behavior is calculated using the following formula:

[0096] ,

[0097] in, Indicates The weight of the biological object's influence on group behavior; represents the additional weight of the biological object; represents the decision area weight; Indicates the weight of the follow area; Indicates the non-following area weight.

[0098] Fourth embodiment:

[0099] On the basis of any of the above embodiments, in S300, the influence weight of the non-biological object on the group behavior is determined by the behavior trajectory of the above target, and the specific steps include:

[0100] S311, obtaining the behavior trajectories of the above-mentioned non-biological objects and the behavior trajectories of the biological objects in the same period, predicting the behavior trajectories of the non-biological objects in the next period and the behavior trajectories of the biological objects when they are not affected by the non-biological objects; determining the population distribution range in the next period according to the predicted behavior trajectories of the above-mentioned biological objects;

[0101] S312, determining the position of the non-biological object passing through the group distribution range in the next time period according to the predicted behavior trajectory and group distribution range of the non-biological object;

[0102] Determine the additional weight of the abiotic object according to the position of the abiotic object in the distribution range of the population;

[0103] S313, determining the movement speed of the non-living object according to the obtained behavior trajectory and time period of the non-living object; wherein the non-living object is provided with a plurality of speed intervals, and each speed interval is assigned a corresponding weight;

[0104] Determining a speed weight of the non-living object according to the speed interval of the movement speed of the non-living object;

[0105] S314. Determine the influence weight of the non-living object on the group behavior through the additional weight and speed weight of the non-living object.

[0106] The prediction, spatial analysis and weight distribution are combined to determine the weight of the impact of non-biological objects on group behavior. First, the behavior trajectories of non-biological objects and biological objects in the same period are collected. These behavior trajectories can be obtained through video surveillance, sensor data or other tracking technologies; machine learning or physical models are used to predict the behavior trajectory of non-biological objects in the next period, and at the same time, the behavior trajectory of biological objects when they are not affected by non-biological objects is predicted. According to the predicted behavior trajectory of biological objects, the group distribution range of the next period is determined (that is, the area where biological objects may gather in the next period is identified). Then, according to the predicted behavior trajectory of non-biological objects, the location of the group distribution range that they will pass through in the next period is determined, and the possible impact of the behavior of non-biological objects on group behavior is evaluated, that is, the additional weight. According to the obtained behavior trajectory of non-biological objects and the duration of the time period, their movement speed is calculated, and their speed weight is determined according to the speed range of the movement speed of non-biological objects. Finally, by combining the additional weight and speed weight of non-biological objects, their influence weight on group behavior is calculated.

[0107] A reference usage scenario is to study the impact of different vehicles (such as cars, buses, bicycles, etc.) on crowd behavior in an open square in order to provide decision support for urban planning and management. The square often has crowds and vehicles passing through. The behavior trajectories of vehicles and crowds in the same period are collected through the surveillance cameras in the square. The behavior trajectory data includes information such as vehicle type, speed, position, and walking path and speed of the crowd. The machine learning algorithm (such as deep learning model) is used to predict the behavior trajectory of the vehicle in the next period, and at the same time, the possible walking path of the crowd is predicted when it is not affected by the vehicle. According to the predicted crowd behavior trajectory, the area where the crowd may gather in the next period is determined. The gathering area may include the entrance, exit, rest area, etc. of the square. According to the predicted vehicle behavior trajectory, the crowd gathering area that the vehicle will pass through in the next period is determined. If a high-speed bus passes near the crowd gathering area, its additional weight is higher; in addition, the determination of the additional weight can also consider factors such as vehicle type (such as the impact of cars on the crowd may be less than that of buses), speed, and relative position to the crowd.

[0108] Based on the collected vehicle behavior trajectory data, the average speed of each vehicle is calculated, and multiple speed intervals (such as low speed, medium speed, and high speed) are set for the vehicle. Each interval is assigned a corresponding weight, and the weight of the high speed interval is greater than that of the low speed interval. The speed weight of each vehicle is determined according to the speed interval in which the speed of the vehicle is located. Combined with the vehicle's additional weight and speed weight, the weight of each vehicle's impact on crowd behavior is calculated.

[0109] Through the above process, we can get the weights of the impact of different vehicles on crowd behavior. These weights can be used to evaluate the impact of vehicles on the flow and safety of square crowds, and provide decision support for urban planners and managers. For example, we can adjust the vehicle route, limit vehicle speed, or increase crowd protection measures based on these weights.

[0110] In a specific embodiment, the additional weight of the non-biological object in the next period is calculated using the following formula:

[0111] ,

[0112] in, Indicates the T+1 period Additional weights for inanimate objects; represents the decision area weight; Indicates the weight of the follow area; Indicates the T+1 period The area of ​​the decision region covered by the inanimate objects; It represents the total area of ​​the decision-making region in the T+1 period; Indicates the T+1 period The area of ​​the following region covered by the non-biological object; Indicates the total area of ​​the follow-up region during the T+1 period.

[0113] In a specific embodiment, the following formula is used to calculate the movement speed of the non-biological object:

[0114] ,in, Indicates The speed of movement of inanimate objects; Indicates the T period behavioral trajectories of inanimate objects; Indicates the duration of the T period;

[0115] Judging from the above The speed interval to which the non-biological object's movement speed belongs is retrieved, and the corresponding speed weight is obtained. Velocity weight for non-mob objects.

[0116] In a specific embodiment, the following formula is used to calculate the influence weight of the above-mentioned non-biological objects on group behavior:

[0117] ,

[0118] Indicates The weight of the impact of non-biological objects on group behavior; Indicates Velocity weight for non-mob objects.

[0119] The specific implementation methods described above further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above description is only a specific implementation method of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for group behavior recognition based on video data, characterized in that: The following steps are involved: S100, acquiring a group behavior video, sampling the group behavior video, and obtaining a video image set consisting of a plurality of video images; S200, extracting behavioral features of targets from the video image, wherein the targets include biological objects and non-biological objects; wherein a group is composed of a number of the biological objects, and the behavioral features of the group are determined according to the behavioral features of the biological objects; Determine the influence weight of the biological object on the group behavior based on the behavior characteristics of the biological object and the group, and obtain a first weight; S300, tracking the target using a tracking algorithm to obtain a behavior trajectory of the target; Determine the influence weight of the non-biological object on the group behavior through the behavior trajectory of the target, and obtain a second weight; S400: Obtain group behavior recognition data through the behavior characteristics, the first weight and the second weight of the target.

2. The group behavior recognition method according to claim 1, characterized in that: In S200, the behavior characteristics of the group are determined according to the behavior characteristics of the biological objects, and the specific steps include: S211, classifying the behavior characteristics of the biological object to obtain a biological behavior category; S212: Count the number of biological objects in the biological behavior category, sort the biological behavior categories from large to small according to the number of biological objects, and obtain a biological behavior sequence. ;in, represents the first biological behavior category, Indicates Biobehavioral categories, , Indicates Biobehavioral Behavioral characteristics of biological objects; S213, selecting a first biological behavior category from the biological behavior sequence The behavioral characteristics of the group are used as the behavioral characteristics of the group.

3. The group behavior recognition method according to claim 1, characterized in that: In S200, the influence weight of the biological object on the group behavior is determined based on the behavior characteristics of the biological object and the group. The specific steps include: S221, the behavioral characteristics of the biological object include action, posture and position in the group; S222, extracting a group distribution range from the video image, dividing the group distribution range into a decision area, a following area, and a non-following area, and assigning corresponding weights to the decision area, the following area, and the non-following area; S223, determining an additional weight of the biological object according to the position of the biological object in the group; S224, comparing the posture of the biological object with that of the group, if the postures are consistent, then the posture consistency number of the biological object is increased by 1; otherwise, the posture inconsistency number of the biological object is increased by 1; comparing the action of the biological object with that of the group, if the actions are consistent, then the action consistency number of the biological object is increased by 1; otherwise, the action inconsistency number of the biological object is increased by 1; Traversing the video image set, determining the posture consistency rate of the biological object and the group based on the number of consistent postures and the number of inconsistent postures of the biological object; determining the action consistency rate of the biological object and the group based on the number of consistent actions and the number of inconsistent actions of the biological object; Determining the behavior consistency rate between the biological object and the group through the posture consistency rate and the action consistency rate; S225. Determine the influence weight of the biological object on the group behavior through the additional weight of the biological object and the behavior consistency rate.

4. The group behavior recognition method according to claim 1, characterized in that: In S300, the influence weight of the non-biological object on the group behavior is determined according to the behavior trajectory of the target. The specific steps include: S311, obtaining the behavior trajectories of the non-biological objects and the behavior trajectories of the biological objects in the same time period, predicting the behavior trajectories of the non-biological objects in the next time period and the behavior trajectories of the biological objects when they are not affected by the non-biological objects; determining the population distribution range in the next time period according to the predicted behavior trajectories of the biological objects; S312, determining the position of the non-biological object passing through the group distribution range in the next time period according to the predicted behavior trajectory and group distribution range of the non-biological object; Determining an additional weight of the non-biological object according to the position through which the non-biological object passes in the population distribution range; S313, determining the movement speed of the non-living object according to the obtained behavior trajectory and time period of the non-living object; wherein the non-living object is provided with a plurality of speed intervals, and each speed interval is assigned a corresponding weight; Determining a speed weight of the non-living object according to the speed interval of the movement speed of the non-living object; S314: Determine the influence weight of the non-living object on the group behavior by using the additional weight and speed weight of the non-living object.

5. The group behavior recognition method according to claim 3, characterized in that: The behavior consistency rate between the biological object and the group is calculated using the following formula: ， in, Indicates The rate of consistency between the behavior of the biological object and the group; Indicates The rate of agreement between the biological object and the group’s posture; Indicates The number of pose congruences between the biological object and the group; Indicates The number of posture inconsistencies between biological objects and the group; Indicates The consistency rate of the biological object's actions with the group; Indicates The number of consistent actions between the biological object and the group; Indicates The number of inconsistent actions between biological objects and groups; represents the weight of posture in the behavior consistency rate; Represents the weight of the action in the behavior consistency rate.

6. The group behavior recognition method according to claim 5, characterized in that: The influence weight of the biological object on the group behavior is calculated using the following formula: ， in, Indicates The weight of the biological object's influence on group behavior; represents the additional weight of the biological object; represents the decision area weight; Indicates the weight of the follow area; Indicates the non-following area weight.

7. The group behavior recognition method according to claim 4, characterized in that: The additional weight of the non-biological object in the next period is calculated using the following formula: ， in, Indicates the T+1 period Additional weights for inanimate objects; represents the decision area weight; Indicates the weight of the follow area; Indicates the T+1 period The area of ​​the decision region covered by the inanimate objects; Represents the total area of ​​the decision-making region during the T+1 period; Indicates the T+1 period The area of ​​the following region covered by the non-biological object; Indicates the total area of ​​the follow-up region during the T+1 period.

8. The group behavior recognition method according to claim 7, characterized in that: The movement speed of the non-biological object is calculated using the following formula: ,in, Indicates The speed of movement of inanimate objects; Indicates the T period behavioral trajectories of inanimate objects; Indicates the duration of the T period; Determine the The speed interval to which the non-biological object's movement speed belongs is retrieved, and the corresponding speed weight is obtained. Velocity weight for non-mob objects.

9. The group behavior recognition method according to claim 8, characterized in that: The influence weight of the non-biological object on the group behavior is calculated using the following formula: ， Indicates The weight of the impact of non-biological objects on group behavior; Indicates Velocity weight for non-mob objects.

Images