A method, device, equipment and medium for predicting human flow density distribution

Abstract

Embodiments of the present application disclose a kind of crowd density distribution prediction method, device, equipment and medium, involve artificial intelligence technical field.The method comprises: obtaining scene environment information and the historical trajectory information of multiple individuals;Scene environment information and the historical trajectory information of multiple individuals are input into space-time decoupling differential equation network, obtain the crowd density distribution of future arbitrary time output by space-time decoupling differential equation network.From this, since space-time decoupling differential equation network is from real trajectory information learns the motion mode of crowd, rather than relying on preset simplified rule, therefore space-time decoupling differential equation network can automatically capture and fit the complex nonlinear interaction and randomness in actual scene Behavior, so that the prediction result of crowd density distribution is more real, significantly reduces the prediction error of crowd density distribution.

Description

Technical Field

[0001] This application relates to the field of artificial intelligence technology, and in particular to a method, apparatus, device and medium for predicting the distribution of pedestrian density. Background Technology

[0002] With the acceleration of urbanization, the distribution of pedestrian density, as an indicator for quantifying the gathering and movement of people, is crucial for accurate prediction in areas such as urban public safety management and the formulation of emergency evacuation plans.

[0003] Currently, the approach typically relies on physical model-driven frameworks (such as social force models) to simulate the individual movement of a single pedestrian based on pre-defined microscopic physical rules (such as attraction and repulsion between individuals), and then uses aggregation strategies to deduce the movement of the group based on the individual movement, thereby predicting the distribution of pedestrian density.

[0004] However, physical model-driven frameworks based on fixed microscopic physical rules are difficult to effectively capture and learn the nonlinearity and randomness of crowd interactions in real-world scenarios, resulting in low accuracy in predicting crowd density distribution. Summary of the Invention

[0005] To address the aforementioned issues, this application provides a method, apparatus, device, and medium for predicting pedestrian density distribution, which can improve the accuracy of pedestrian density distribution prediction.

[0006] The embodiments of this application disclose the following technical solutions:

[0007] In a first aspect, this application discloses a method for predicting pedestrian density distribution, the method comprising:

[0008] Acquire scene environment information and historical trajectory information of multiple individuals;

[0009] The scene environment information and the historical trajectory information of the multiple individuals are input into a spatiotemporal decoupled differential equation network to obtain the predicted pedestrian density distribution at any future time output by the spatiotemporal decoupled differential equation network.

[0010] Optionally, the spatiotemporal decoupling differential equation network includes a microscopic trajectory prediction network and a density-velocity coupled graph learning module;

[0011] The step of inputting the scene environment information and the historical trajectory information of the multiple individuals into a spatiotemporal decoupled differential equation network to obtain the predicted pedestrian density distribution at any future time output by the spatiotemporal decoupled differential equation network includes:

[0012] The scene environment information and the historical trajectory information of the multiple individuals are input into the micro-trajectory prediction network to obtain the predicted velocity information and predicted position information of the multiple individuals at the prediction start time;

[0013] The predicted velocity and location information of the multiple individuals at the predicted start time are input into the density-velocity coupled graph learning module to obtain the predicted population density distribution at any future time.

[0014] Optionally, the density-velocity coupled graph learning module includes: a differentiable density mapping module, a continuous cross-lattice detection module, and a neural differential equation module;

[0015] The step of inputting the predicted velocity and predicted location information of the multiple individuals at the predicted start time into the density-velocity coupled graph learning module to obtain the predicted pedestrian density distribution at any future time includes:

[0016] The predicted location information of the multiple individuals at the predicted start time is input into the differentiable density mapping module to generate the crowd density distribution at the predicted start time.

[0017] The predicted velocity information of the multiple individuals at the prediction start time and the pedestrian density distribution at the prediction start time are input into the continuous cross-grid detection module to determine the density flux change at the grid boundary;

[0018] The pedestrian density distribution at the predicted starting time and the density flux change are input into the neural differential equation module to obtain the pedestrian density distribution at any future time.

[0019] Optionally, the method further includes:

[0020] The micro-trajectory prediction network determines the predicted trajectory information of the multiple individuals at any future time based on their predicted velocity and position information at the prediction start time.

[0021] Optionally, the spatiotemporal decoupling differential equation network is trained in the following manner:

[0022] The training environment information and the historical trajectory information of multiple training individuals are used as inputs. The real trajectory information of the multiple training individuals at any future time and the real pedestrian density distribution are used as training labels. The neural network model is trained through a joint loss function. The joint loss function is used to simultaneously constrain the trajectory error and the pedestrian density distribution error.

[0023] Optionally, determining the density flux change at the grid boundary includes:

[0024] By using a node embedding mechanism, the predicted speed information of the multiple individuals at the prediction start time and the crowd density distribution at the prediction start time are dimensionality reduced to obtain the dimensionality-reduced predicted speed information and crowd density distribution.

[0025] Based on the predicted velocity information after dimensionality reduction and the population density distribution, the density flux change on the grid boundary is determined.

[0026] Secondly, this application discloses a device for predicting the distribution of pedestrian flow density, the device comprising: an information acquisition module and a distribution prediction module;

[0027] The information acquisition module is used to acquire scene environment information and historical trajectory information of multiple individuals;

[0028] The distribution prediction module is used to input the scene environment information and the historical trajectory information of the multiple individuals into a spatiotemporal decoupled differential equation network to obtain the predicted pedestrian density distribution at any future time output by the spatiotemporal decoupled differential equation network.

[0029] Optionally, the spatiotemporal decoupling differential equation network includes a microscopic trajectory prediction network and a density-velocity coupled graph learning module;

[0030] The distribution prediction module includes: a first prediction module and a second prediction module;

[0031] The first prediction module is used to input the scene environment information and the historical trajectory information of the multiple individuals into the micro-trajectory prediction network to obtain the predicted velocity information and predicted position information of the multiple individuals at the prediction start time;

[0032] The second prediction module is used to input the predicted velocity information and predicted location information of the multiple individuals at the prediction start time into the density-velocity coupled graph learning module to obtain the predicted population density distribution at any future time.

[0033] Optionally, the density-velocity coupled graph learning module includes: a differentiable density mapping module, a continuous cross-lattice detection module, and a neural differential equation module;

[0034] The second prediction module includes: a first prediction submodule, a second prediction submodule, and a third prediction submodule;

[0035] The first prediction submodule is used to input the predicted location information of the multiple individuals at the prediction start time into the differentiable density mapping module to generate the crowd density distribution at the prediction start time.

[0036] The second prediction submodule is used to input the predicted velocity information of the multiple individuals at the prediction start time and the population density distribution at the prediction start time into the continuous cross-grid detection module to determine the density flux change on the grid boundary;

[0037] The third prediction submodule is used to input the pedestrian density distribution at the prediction start time and the density flux change into the neural differential equation module to obtain the pedestrian density distribution at any future time.

[0038] Optionally, the device further includes: a third prediction module;

[0039] The third prediction module is used to determine the predicted trajectory information of the multiple individuals at any future time by using the micro-trajectory prediction network, based on the predicted velocity information and predicted position information of the multiple individuals at the prediction start time.

[0040] Optionally, the spatiotemporal decoupled differential equation network is trained using the following units:

[0041] The training unit is used to take training environment information and historical trajectory information of multiple training individuals as input, and the real trajectory information and real pedestrian density distribution of the multiple training individuals at any future time as training labels, and train the neural network model through a joint loss function; the joint loss function is used to simultaneously constrain trajectory error and pedestrian density distribution error.

[0042] Optionally, the second prediction submodule is specifically used to: reduce the dimensionality of the predicted velocity information of the multiple individuals at the prediction start time and the pedestrian density distribution at the prediction start time through a node embedding mechanism to obtain the dimensionality-reduced predicted velocity information and pedestrian density distribution; and determine the density flux change on the grid boundary based on the dimensionality-reduced predicted velocity information and pedestrian density distribution.

[0043] Thirdly, this application discloses a device for predicting the distribution of pedestrian flow density, the device comprising: a memory and a processor;

[0044] The memory is used to store programs;

[0045] The processor is configured to execute the program to implement the various steps of the method for predicting the distribution of pedestrian flow density as described in the first aspect.

[0046] Fourthly, this application discloses a computer-readable medium having a computer program stored thereon, which, when executed by a processor, implements the various steps of the method for predicting the distribution of pedestrian flow density as described in the first aspect.

[0047] Compared with the prior art, this application has the following advantages:

[0048] This application provides a method, apparatus, device, and medium for predicting pedestrian density distribution. The method includes: acquiring scene environment information and historical trajectory information of multiple individuals; inputting the scene environment information and the historical trajectory information of multiple individuals into a spatiotemporal decoupled differential equation network to obtain the pedestrian density distribution at any future time output by the spatiotemporal decoupled differential equation network. Therefore, since the spatiotemporal decoupled differential equation network learns the movement patterns of crowds from real trajectory information, rather than relying on preset simplified rules, it can automatically capture and fit complex nonlinear interactions and random behaviors in real-world scenarios, thus making the predicted pedestrian density distribution more realistic and significantly reducing the prediction error. Attached Figure Description

[0049] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0050] Figure 1 A flowchart illustrating a method for predicting pedestrian density distribution provided in an embodiment of this application;

[0051] Figure 2 A flowchart of an application of a spatiotemporally decoupled differential equation network is provided for an embodiment of this application;

[0052] Figure 3 A flowchart of an application density-velocity coupled graph learning module provided in an embodiment of this application;

[0053] Figure 4 A schematic diagram of a device for predicting pedestrian density distribution provided in an embodiment of this application;

[0054] Figure 5 This is a schematic diagram of a computer-readable medium provided in an embodiment of this application. Detailed Implementation

[0055] As described above, currently, the approach typically relies on physical model-driven frameworks (such as social force models) to simulate the individual movement of a single pedestrian based on pre-defined micro-physical rules (such as attraction and repulsion between individuals), and then uses aggregation strategies to deduce the movement of the group based on the individual movement, thereby predicting the distribution of pedestrian density.

[0056] However, physical model-driven frameworks based on fixed microscopic physical rules are difficult to effectively capture and learn the nonlinearity and randomness of crowd interactions in real-world scenarios, resulting in low accuracy in predicting crowd density distribution.

[0057] Through research, the inventors proposed a method, device, equipment, and medium for predicting pedestrian density distribution. Because the spatiotemporal decoupled differential equation network learns the movement patterns of crowds from real trajectory information, rather than relying on pre-defined simplified rules, it can automatically capture and fit complex nonlinear interactions and random behaviors in real-world scenarios. This results in more realistic predictions of pedestrian density distribution and significantly reduces prediction errors. Furthermore, the spatiotemporal decoupled differential equation network incorporates the fluid dynamics continuity equation as a strong physical constraint during training, ensuring that the prediction results adhere to the law of conservation of mass even in long-term simulations. This avoids the irrationalities that may occur with purely data-driven models, enhancing the physical rationality and long-term stability of the predicted pedestrian density distribution.

[0058] To enable those skilled in the art to better understand the present application, the technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present application, and not all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present application.

[0059] See Figure 1 The figure is a flowchart of a method for predicting pedestrian density distribution provided in an embodiment of this application. The method includes:

[0060] S101: Obtain scene environment information and historical trajectory information of multiple individuals.

[0061] Scene environment information includes map layout and obstacle locations.

[0062] Map layout refers to a two-dimensional (2D) or three-dimensional (3D) digital map of an indoor environment, capable of defining walkable areas (such as passageways, lobbies, corridors, etc.) and non-walkable areas (such as walls, fixed columns, enclosed rooms, etc.). Understandably, two-dimensional digital maps are suitable for most planar indoor scenes (such as subway station halls, shopping mall floors, etc.). Three-dimensional digital maps are suitable for complex three-dimensional scenes (such as staircases, escalator locations, etc.) and can support the learning of cross-floor pedestrian flow patterns.

[0063] Obstacle location refers to the coordinates of fixed or temporary objects within a walkable area that impede pedestrian passage. Fixed objects are those that are permanently located, such as indoor potted plants, information desk counters, fixed shelves, and seating areas. Temporary objects are those that exist only for a short period, such as temporary promotional displays, cleaning equipment, and stacked goods (which need to be dynamically obtained based on the update frequency of the scene environment).

[0064] Understandably, by acquiring scene environment information, the model can learn how pedestrians interact with the environment. For example, it can learn the "linear pedestrian flow formation pattern" in passage areas, the "diffusion movement pattern" in spacious areas such as halls, and the "turning and avoidance pattern" at corners, thereby avoiding predicting results that do not conform to physical logic, such as "pedestrians passing through walls" or "passing through walls", and ensuring the scene adaptability of the prediction.

[0065] Historical trajectory information refers to continuous location coordinates formed by collecting location data at fixed time intervals within a historical period. The historical trajectory information of each individual is represented in the form of a coordinate sequence, such as [(x1,y1,t1),(x2,y2,t2),...,(xn,yn,tn)]. Here, (x,y) refers to the individual's planar position coordinates at the corresponding time step t (consistent with the dimension indicated by the scene environment information).

[0066] In one specific implementation, a target detection algorithm (such as YOLO) can first be used to detect targets in each frame of the acquired video image, identifying all pedestrian targets in each frame. Then, a multi-target tracking algorithm (such as DeepSORT) can be used to associate the same pedestrian target in consecutive frames to obtain the individual pedestrian's motion trajectory. Finally, based on the individual pedestrian's motion trajectory and the camera's calibration parameters, the individual pedestrian's historical trajectory information is obtained.

[0067] Understandably, by acquiring historical trajectory information of multiple individuals, the model can learn the rules of social interaction between individuals (for example, when two pedestrians are walking towards each other, they will move to the sides when they are close to each other; for example, when multiple pedestrians are moving in the same direction, they will form a following queue and maintain similar walking speeds) and individual movement habits (such as the average walking speed, turning preferences, and stopping habits of different pedestrians).

[0068] S102: Input scene environment information and historical trajectory information of multiple individuals into the spatiotemporal decoupled differential equation network to obtain the predicted pedestrian density distribution at any future time output by the spatiotemporal decoupled differential equation network.

[0069] See Figure 2 This figure is a flowchart of an application of a spatiotemporal decoupling differential equation network provided in an embodiment of this application. Figure 2 As shown, the Spatial-Temporal Decoupled Differential Network (STDDN) includes a connected microscopic trajectory prediction network and a density-velocity coupled graph learning module.

[0070] A1: Input scene environment information and historical trajectory information of multiple individuals into the micro trajectory prediction network to obtain the predicted velocity information and predicted position information of multiple individuals at the prediction start time.

[0071] On the one hand, by inputting scene environment information, the micro-trajectory prediction network can perceive and understand the physical structure of space and the location of obstacles. On the other hand, by inputting historical trajectory information of multiple individuals, the micro-trajectory prediction network can learn the interaction between individuals and surrounding pedestrians and the environment. Thus, the micro-trajectory prediction network can deduce the predicted velocity and predicted position information of multiple individuals at the start of the prediction, which conforms to both physical constraints and interaction logic.

[0072] In one specific implementation, the method further includes: using a micro-trajectory prediction network, based on the predicted velocity information (v0) and predicted position information (p0) of multiple individuals at the prediction start time (e.g., t=0), to determine the predicted trajectory information of multiple individuals at any future time in an autoregressive manner.

[0073] A2: Input the predicted velocity and location information of multiple individuals at the predicted start time into the density-velocity coupled graph learning module to obtain the predicted pedestrian density distribution at any future time.

[0074] See Figure 3 This figure is a flowchart of an application density-velocity coupled graph learning module provided in an embodiment of this application. Figure 3 As shown, the Density-Velocity Coupled Graph (DVCG) learning module includes: Differentiable Density Mapping (DDM), Continuous Grid-crossing Detection (CGD), and Neural Differential Equation module.

[0075] A21: Input the predicted location information of multiple individuals at the predicted start time into the differentiable density mapping module to generate the predicted population density distribution at the predicted start time.

[0076] The core function of the differentiable density mapping module is to transform the predicted location information of multiple individuals at the predicted start time, output by step A1, into the predicted population density distribution at the predicted start time through a probability distribution model, thereby solving the problem of "gradient discontinuity" caused by existing discrete modeling (such as hard mesh partitioning).

[0077] Specifically, the differentiable density mapping module constructs a probability distribution function (e.g., Gaussian distribution) for each individual's predicted location, and calculates the individual's contribution to the surrounding grid, where the contribution is negatively correlated with the distance between the individual and the surrounding grid. Subsequently, by aggregating the contribution values ​​of all individuals to the same grid, the crowd density distribution (ρ0) at the prediction start time is generated.

[0078] A22: Input the predicted velocity information of multiple individuals at the predicted start time and the population density distribution at the predicted start time into the continuous cross-grid detection module to determine the density flux change on the grid boundary.

[0079] Density flux change refers to the rate of change of the number of people passing through a grid boundary per unit time, and the unit is: people / (meter·second).

[0080] The core function of the continuous cross-grid detection module is to dynamically simulate the movement of individuals on the grid boundary based on the predicted velocity information of multiple individuals at the predicted start time output by step A1 and the crowd density distribution at the predicted start time output by step A21, and to calculate the density flux change at the grid boundary.

[0081] It should be noted that when dealing with extremely large crowds (such as stadiums, train station plazas, etc.), the computational overhead and memory consumption can be excessive. Therefore, the continuous cross-grid detection module introduces a node embedding mechanism to optimize computation. Specifically, through the node embedding mechanism, the predicted velocity information of multiple individuals at the prediction start time and the crowd density distribution at the prediction start time are dimensionality-reduced (e.g., high-dimensional coordinate vectors are transformed into 256-dimensional embedding vectors) to obtain the dimensionality-reduced predicted velocity information and crowd density distribution. Subsequently, based on the dimensionality-reduced predicted velocity information and crowd density distribution, the density flux change at the grid boundary is determined.

[0082] This significantly reduces the memory footprint and computational complexity of subsequent graph calculations, enabling the model to be scaled up to ultra-large scenes such as stadiums and train station plazas, while maintaining accuracy without significant loss due to the preservation of core features.

[0083] A23: Input the predicted pedestrian density distribution and density flux change at the initial time into the neural differential equation module to obtain the pedestrian density distribution at any future time.

[0084] The neural differential equation uses the continuity equation in fluid mechanics as an inherent strong physical constraint. It takes the predicted population density distribution at the starting moment output by step A21 and the density flux change on the grid boundary output by step A22 as inputs. Through numerical integration, it deduces and outputs the population density distribution at any future moment (ρt, t≥0).

[0085] Understandably, introducing the physical law of continuity has the following beneficial effects: Firstly, the continuity equation is essentially the differential form of the law of conservation of mass, which mandates that "the increase or decrease in population within a region must equal the net flux of population flowing into and out of that region." This fundamentally eliminates phenomena that violate physical common sense, such as "population appearing out of thin air or disappearing without cause" in the prediction results. Secondly, because the model evolves according to stable physical laws, rather than purely relying on data-driven statistical extrapolation, even in long-term predictions (such as after 30 seconds or 1 minute), the prediction results will not exhibit unreasonable divergence or collapse. This effectively suppresses the problem of gradual error accumulation and amplification in traditional autoregressive prediction methods, ensuring the long-term reliability and robustness of the prediction results.

[0086] It should be noted that the prediction accuracy of step S102 depends on the effective training of STDDN. Specifically, the spatiotemporal decoupling differential equation network is trained in the following way:

[0087] B1: Obtain the training dataset; the training dataset includes: training environment information, historical trajectory information of multiple training individuals, real trajectory information of multiple training individuals at any future time, and real pedestrian density distribution.

[0088] B2: The training environment information and the historical trajectory information of multiple training individuals are used as inputs. The real trajectory information of multiple training individuals at any future time and the real pedestrian density distribution are used as training labels. The neural network model is trained through a joint loss function to obtain the STDDN model.

[0089] It should be noted that the joint loss function, through backpropagation and gradient descent algorithms, simultaneously minimizes micro-errors (the difference between the model's predicted individual trajectories and the actual trajectories) and macro-errors (the difference between the model's predicted density distribution and the actual density distribution), thereby optimizing the parameters of all modules of STDDN. This ensures that the final trained STDDN simultaneously satisfies the goals of "reasonable individual trajectories" and "accurate overall density".

[0090] It is understandable that in emergency evacuation scenarios in subway stations, the crowd density distribution prediction method provided in this application embodiment can predict passenger movement trends in real time, identify potential congestion areas in advance, and provide decision support for managers.

[0091] It is understood that the method for predicting pedestrian density distribution provided in this application has achieved better results than existing methods on multiple public datasets (such as GC, UCY, ETH, HOTEL, etc.).

[0092] In summary, this application discloses a method for predicting pedestrian density distribution. This method includes: acquiring scene environment information and historical trajectory information of multiple individuals; inputting the scene environment information and the historical trajectory information of multiple individuals into a spatiotemporal decoupled differential equation network to obtain the pedestrian density distribution at any future time output by the spatiotemporal decoupled differential equation network. Therefore, since the spatiotemporal decoupled differential equation network learns the movement patterns of crowds from real trajectory information, rather than relying on preset simplified rules, it can automatically capture and fit complex nonlinear interactions and random behaviors in real-world scenarios, thus making the predicted pedestrian density distribution more realistic and significantly reducing the prediction error. Furthermore, the spatiotemporal decoupled differential equation network introduces the fluid dynamics continuity equation as a strong physical constraint during training, ensuring that the prediction results follow the law of conservation of mass even in long-term extrapolation, avoiding the unreasonable phenomena that may occur in purely data-driven models, and enhancing the physical rationality and long-term stability of the predicted pedestrian density distribution.

[0093] See Figure 4 The figure is a schematic diagram of a pedestrian density distribution prediction device provided in an embodiment of this application. The pedestrian density distribution prediction device 400 includes: an information acquisition module 401 and a distribution prediction module 402.

[0094] The information acquisition module 401 is used to acquire scene environment information and historical trajectory information of multiple individuals;

[0095] The distribution prediction module 402 is used to input scene environment information and historical trajectory information of multiple individuals into the spatiotemporal decoupled differential equation network to obtain the predicted human flow density distribution at any future time output by the spatiotemporal decoupled differential equation network.

[0096] In one specific implementation, the spatiotemporal decoupled differential equation network includes a microscopic trajectory prediction network and a density-velocity coupled graph learning module;

[0097] The distribution prediction module 402 includes: a first prediction module and a second prediction module;

[0098] The first prediction module is used to input scene environment information and historical trajectory information of multiple individuals into the micro trajectory prediction network to obtain the predicted velocity information and predicted position information of multiple individuals at the prediction start time.

[0099] The second prediction module is used to input the predicted velocity and location information of multiple individuals at the prediction start time into the density-velocity coupled graph learning module to obtain the predicted population density distribution at any future time.

[0100] In one specific implementation, the density-velocity coupled graph learning module includes: a differentiable density mapping module, a continuous cross-lattice detection module, and a neural differential equation module;

[0101] The second prediction module includes: a first prediction submodule, a second prediction submodule, and a third prediction submodule;

[0102] The first prediction submodule is used to input the predicted location information of multiple individuals at the prediction start time into the differentiable density mapping module to generate the crowd density distribution at the prediction start time.

[0103] The second prediction submodule is used to input the prediction velocity information of multiple individuals at the prediction start time and the population density distribution at the prediction start time into the continuous cross-grid detection module to determine the density flux change on the grid boundary.

[0104] The third prediction submodule is used to input the pedestrian density distribution and density flux change at the prediction start time into the neural differential equation module to obtain the pedestrian density distribution at any future time.

[0105] In one specific implementation, the pedestrian density distribution prediction device 400 further includes: a third prediction module;

[0106] The third prediction module is used to determine the predicted trajectory information of multiple individuals at any future time by using a micro-trajectory prediction network based on the predicted velocity and predicted position information of multiple individuals at the prediction start time.

[0107] In one specific implementation, the spatiotemporal decoupled differential equation network is trained using the following units:

[0108] The training unit takes training environment information and historical trajectory information of multiple training individuals as input, and uses the real trajectory information of multiple training individuals and the real pedestrian density distribution at any future time as training labels to train the neural network model through a joint loss function. The joint loss function is used to simultaneously constrain trajectory error and pedestrian density distribution error.

[0109] In one specific implementation, the second prediction submodule is specifically used to: reduce the dimensionality of the prediction velocity information of multiple individuals at the prediction start time and the pedestrian density distribution at the prediction start time through a node embedding mechanism to obtain the dimensionality-reduced prediction velocity information and pedestrian density distribution; and determine the density flux change on the grid boundary based on the dimensionality-reduced prediction velocity information and pedestrian density distribution.

[0110] In summary, this application discloses a device for predicting pedestrian density distribution. Because the spatiotemporal decoupled differential equation network learns the movement patterns of crowds from real trajectory information, rather than relying on pre-defined simplified rules, it can automatically capture and fit complex nonlinear interactions and random behaviors in real-world scenarios. This results in more realistic predictions of pedestrian density distribution and significantly reduces prediction errors. Furthermore, the spatiotemporal decoupled differential equation network incorporates the fluid dynamics continuity equation as a strong physical constraint during training, ensuring that the prediction results adhere to the law of conservation of mass even in long-term simulations. This avoids the irrational phenomena that may occur with purely data-driven models, enhancing the physical rationality and long-term stability of the predicted pedestrian density distribution.

[0111] This application also provides a corresponding device for predicting pedestrian density distribution and a computer-readable medium for implementing the method for predicting pedestrian density distribution provided in this application.

[0112] The device for predicting pedestrian density distribution includes a memory and a processor. The memory is used to store instructions or code, and the processor is used to execute the instructions or code so that the device performs a method for predicting pedestrian density distribution according to any embodiment of this application.

[0113] See Figure 5 This figure is a schematic diagram of a computer-readable medium provided in an embodiment of this application. The computer-readable medium 500 stores a computer program 511, which, when executed by a processor, implements the above-described... Figure 1 The steps of a method for predicting the distribution of pedestrian density.

[0114] It should be noted that, in the context of this application, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. Machine-readable media can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0115] It should be noted that the machine-readable medium described above in this application can be a computer-readable signal medium or a computer-readable storage medium, or any combination of the two. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0116] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device.

[0117] Although the subject matter has been described using language specific to structural features and / or methodological logic, it should be understood that the subject matter defined in the appended claims is not necessarily limited to the specific features or actions described above. Rather, the specific features and actions described above are merely illustrative examples of implementing the claims.

[0118] While several specific implementation details are included in the foregoing discussion, these should not be construed as limiting the scope of this application. Certain features described in the context of individual embodiments may also be implemented in combination in a single embodiment. Conversely, various features described in the context of a single embodiment may also be implemented individually or in any suitable sub-combination in multiple embodiments.

[0119] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of disclosure in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for predicting pedestrian density distribution, characterized in that, The method includes: Acquire scene environment information and historical trajectory information of multiple individuals; the scene environment information includes map layout and obstacle locations; The scene environment information and the historical trajectory information of the multiple individuals are input into the micro-trajectory prediction network to obtain the predicted velocity information and predicted position information of the multiple individuals at the prediction start time. The predicted location information of the multiple individuals at the predicted start time is input into the differentiable density mapping module. By constructing a probability distribution function for the predicted locations of the multiple individuals and aggregating the contribution values ​​of each individual to the surrounding grid, the population density distribution at the predicted start time is generated. The predicted velocity information of the multiple individuals at the prediction start time and the pedestrian density distribution at the prediction start time are input into the continuous cross-grid detection module. The predicted velocity information of the multiple individuals at the prediction start time and the pedestrian density distribution at the prediction start time are dimensionality reduced through the node embedding mechanism. The density flux change on the grid boundary is determined based on the dimensionality-reduced predicted velocity information and pedestrian density distribution. By inputting the predicted pedestrian density distribution at the initial time and the density flux change into the neural differential equation module, the pedestrian density distribution at any future time can be obtained.

2. The method according to claim 1, characterized in that, The method further includes: The micro-trajectory prediction network determines the predicted trajectory information of the multiple individuals at any future time based on their predicted velocity and position information at the prediction start time.

3. A device for predicting pedestrian density distribution, characterized in that, The device includes: an information acquisition module and a distribution prediction module; The information acquisition module is used to acquire scene environment information and historical trajectory information of multiple individuals; the scene environment information includes map layout and obstacle locations; The distribution prediction module is used to input the scene environment information and the historical trajectory information of the multiple individuals into a micro-trajectory prediction network to obtain the predicted velocity information and predicted position information of the multiple individuals at the prediction start time; input the predicted position information of the multiple individuals at the prediction start time into a differentiable density mapping module, and generate the crowd density distribution at the prediction start time by constructing a probability distribution function for the predicted positions of the multiple individuals and aggregating the contribution values ​​of each individual to the surrounding grid; input the predicted velocity information of the multiple individuals at the prediction start time and the crowd density distribution at the prediction start time into a continuous cross-grid detection module, and perform dimensionality reduction processing on the predicted velocity information of the multiple individuals at the prediction start time and the crowd density distribution at the prediction start time through a node embedding mechanism, and determine the density flux change on the grid boundary based on the dimensionality-reduced predicted velocity information and crowd density distribution; input the crowd density distribution at the prediction start time and the density flux change into a neural differential equation module to obtain the crowd density distribution at any future time.

4. A device for predicting pedestrian density distribution, characterized in that, The device includes: a memory and a processor; The memory is used to store programs; The processor is used to execute the program to implement the various steps of the method for predicting the distribution of pedestrian flow density as described in claim 1 or 2.

5. A computer-readable medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the various steps of the method for predicting the distribution of pedestrian flow density as described in claim 1 or 2.

Images