A scenic spot passenger flow prediction method, system, medium, device and terminal

Abstract

The present application belongs to the technical field of public safety management and analysis of the spatiotemporal distribution characteristics of crowd gathering behavior in public places, and discloses a scenic spot passenger flow prediction method, system, medium, equipment and terminal. The method uses a face recognition camera to identify the identity of visitors entering the scenic spot, installs a camera on a tourist trail, analyzes the touring habits of visitors based on deep learning, estimates the travel position of visitors, predicts the dynamic distribution of scenic spot passenger flow according to the estimated travel position of visitors at each time point, predicts the personnel congestion period of special scenic spots and popular spots in the scenic spot, and starts a visitor diversion plan. The face recognition method has high accuracy; the behavior characteristics of visitors are identified and transmitted to the server as shared information, improving the accuracy of the entire system in predicting the number of visitors at each scenic spot in the scenic spot and the forwardness of visitor diversion.

Description

Technical Field

[0001] This invention belongs to the field of spatiotemporal distribution characteristics analysis of crowd gathering behavior in public places and the field of public safety management technology, and particularly relates to a method, system, medium, equipment and terminal for predicting the flow of people in scenic areas. Background Technology

[0002] Currently, with the improvement of people's living standards and the pursuit of a high-quality life, leisure tourism has become one of the main ways for people to relax and enjoy life. Due to time, regional, and other limitations, traveling during winter and summer vacations, "Golden Weeks," and short holidays has become the mainstream, causing the number of visitors to major scenic spots to far exceed their designed capacity during special periods. Especially under the requirements of various new infectious disease prevention and control, preventing excessive crowding in scenic spots is a necessary means to protect the lives and property of tourists and reduce the risk of infectious disease transmission. It is also an important measure to strengthen the intelligent management of scenic spots and enhance the tourist experience. Using infrared sensor technology, radio frequency identification technology, wireless communication positioning technology, and video image recognition technology to count and analyze the flow of tourists in scenic spots is the main method for monitoring the distribution of tourist flow and tourist behavior. However, the completeness and maturity of the technology still need to be improved. It cannot meet the needs of refined monitoring of tourist flow and multi-dimensional description of tourist behavior in scenic spots, nor can it predict the gathering behavior of tourists at special attractions and popular spots in scenic spots. It has not fundamentally enhanced the intelligence of scenic spot management and effectively improved the satisfaction of tourist experience.

[0003] Based on the above analysis, the existing technologies have the following problems and shortcomings: the completeness and maturity of existing scenic area visitor flow monitoring methods need to be improved; they cannot meet the needs of refined monitoring of visitor flow and multi-dimensional description of visitor behavior; they cannot predict visitor gathering behavior at special attractions and popular spots within the scenic area; and they cannot fundamentally enhance the intelligence level of scenic area management or effectively improve visitor satisfaction. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a method, system, medium, equipment, and terminal for predicting visitor flow in scenic areas, and particularly relates to a method, system, medium, equipment, and terminal for predicting visitor flow in scenic areas based on deep learning.

[0005] This invention is implemented as follows: a method for predicting visitor flow in scenic areas. The method includes: using facial and behavioral recognition cameras to identify groups and recognize the identities of tourists entering the scenic area; analyzing tourists' travel habits based on deep learning to estimate their future locations at different times; predicting the dynamic distribution of visitor flow at various points in the scenic area at different times based on the estimated time and location of tourists throughout their visit, predicting peak periods for crowds at special attractions and popular spots in the scenic area, and determining whether to activate visitor flow control plans.

[0006] Furthermore, the method for predicting visitor flow in scenic areas includes the following steps:

[0007] Step 1: Deploy facial and behavioral recognition cameras according to the scenic area's tour routes. Collect and identify faces through video and image data and perform identity verification. Analyze the dynamic characteristics of tourists to distinguish their behavior.

[0008] Step two: The camera's image data is transmitted to the server in real time via fiber optic cable. The server analyzes the tourist information collected by multiple cameras and identifies tourist groups.

[0009] Step 3: The server analyzes the data uploaded by all cameras in the scenic area to determine the specific time interval of each group entering the scenic area.

[0010] Step 4: The server determines the number of tourists at each high-traffic attraction at the current time and in the next few moments by determining the current location of the tourist groups and the behavioral characteristics of the tourists in each group, and determines whether to activate the tourist diversion plan.

[0011] Step 5: Based on the actual movement data of tourists, the server updates the estimated number of tourists at each attraction along the tour route in real time, adjusts the tourist flow management plan in real time, and determines whether to provide voice prompts.

[0012] Furthermore, in step four, the server determines the estimated number of tourists at each scenic spot and popular attraction in the next time period by determining the current location of the tourist group and the behavioral characteristics of the tourists in each group. If the estimated number of tourists at a scenic spot or popular attraction exceeds the threshold for the scenic spot's capacity, an early warning message is sent to the scenic area's operation and management department, and an application is made to activate the tourist diversion plan.

[0013] The length of the next time period can be set arbitrarily, with a typical value of 0.5 hours.

[0014] Furthermore, in step four, the visitor management plan is a pre-set procedure at the scenic area's control center. This involves using voice IP broadcasts along the tour route to remind visitors to adjust their tour schedule, including:

[0015] ① Target crowds gathered at popular tourist spots or groups that visit slowly and provide voice IP broadcast reminders;

[0016] ②Provide voice IP broadcast reminders for attractions that tourists do not pay much attention to, miss, or have ample capacity, or for tour groups that visit at a fast pace.

[0017] Furthermore, in step five, the server updates the estimated number of tourists at each attraction in real time based on the actual movement data of the tourists, and adjusts the tourist flow management plan in real time; the flow management prompts broadcast by each voice IP are adjusted in real time according to the behavioral characteristics of the tourists in the current group. If the estimated number of tourists at all attractions is lower than the tourist capacity threshold of the attraction, then no more voice prompts will be given.

[0018] Another objective of this invention is to provide a scenic area visitor flow prediction system that applies the aforementioned scenic area visitor flow prediction method. The scenic area visitor flow prediction system includes face and behavior recognition cameras, fiber optic networks, fiber optic switches, storage servers, and application servers.

[0019] Face and behavior recognition cameras are installed at scenic area entrances, walkways, special locations, attractions, and exits, including fixed-angle bullet cameras and PTZ cameras with pan-tilt heads. PTZ cameras are used for aerial shooting and tracking shooting.

[0020] Fiber optic switches are used to connect all facial and behavioral recognition cameras in the scenic area to the server;

[0021] The server is divided into application server and storage server, which are directly connected to fiber optic switches;

[0022] The storage server includes sufficient disk arrays to meet the scenic area's storage requirements for video data;

[0023] The application server has sufficient computing speed and stores programs to identify the facial information collected by each face and behavior recognition camera, calculate the estimated number of tourists for each attraction at each moment in the next period of time in real time, and compare it with the tourist capacity threshold determined in advance for the attraction.

[0024] The application server broadcasts all the identified identity numbers and facial information for the day to each facial recognition camera periodically via a fiber optic switch, with an update cycle of no more than 10 seconds.

[0025] Furthermore, the face and behavior recognition cameras calibrate the face information, and immediately upload it to the server after successful calibration; the server assigns an identity number to all calibrated face information of the day in real time, and broadcasts the identity number and face image information to each face recognition camera, with an update interval of no more than 10 seconds per update;

[0026] The face and behavior recognition camera stores the facial information and identification number broadcast by the server in its local storage. When a new face enters the frame, it captures the visitor's behavior and compares it with the facial information in the storage. If the face already exists, the camera stores the visitor's behavioral characteristics according to the identification number and facial information stored in the local storage, and transmits the video and recognition data to the server in real time via fiber optic cable. If the visitor's facial information and identification number are not in the local storage, the camera directly transmits the video, facial recognition data, and visitor behavioral characteristics to the server. The local storage only stores the visitor's identification code, the corresponding facial image information, and the visitor's behavioral characteristics, and is completely cleared at 24:00 every day.

[0027] Another object of the present invention is to provide a computer device, the computer device including a memory and a processor, the memory storing a computer program, and when the computer program is executed by the processor, causing the processor to perform the steps of the scenic area visitor flow prediction method.

[0028] Another object of the present invention is to provide a computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the scenic area visitor flow prediction method.

[0029] Another objective of this invention is to provide an information data processing terminal for implementing the scenic area visitor flow prediction system.

[0030] Based on the above technical solutions and the technical problems solved, the advantages and positive effects of the technical solution to be protected by this invention are as follows:

[0031] This invention provides a deep learning-based system and method for monitoring visitor flow in scenic areas. It employs facial recognition cameras to identify visitors entering the scenic area. By installing cameras at a certain density along the walkways, it analyzes visitors' habits using deep learning to estimate their positions at all time points. Based on this, it predicts the dynamic distribution of visitor flow in the scenic area, anticipates peak hours at special attractions and popular spots, takes preventative safety measures, and provides scientific, reasonable, and beneficial guidance to visitors, thereby enhancing visitor satisfaction and the reputation of the scenic area's management.

[0032] The deep learning-based scenic area visitor flow monitoring system provided by this invention uses face and behavior recognition cameras. The cameras process the raw video and image information they collect, achieving high accuracy in face recognition. They also identify visitor behavior characteristics and transmit this information to the server as shared information. The cameras and servers communicate via fiber optic cables, providing high communication speed, large bandwidth, and strong anti-interference capabilities. In the central control center, the storage server and application server are separated, improving the computing speed of the application server.

[0033] The deep learning-based scenic area visitor flow prediction system provided by this invention adopts a fiber optic network architecture, which features high reliability and strong stability, and improves the accuracy of the entire system in predicting the number of visitors to each scenic spot.

[0034] (1) The expected benefits and commercial value of the technical solution of the present invention after transformation are as follows: The product corresponding to the technical solution of the present invention is an essential choice for all provincial-level scenic area management and can be promoted and deployed in 3A and above scenic areas across the country, with broad market prospects.

[0035] (2) The technical solution of the present invention fills the technical gap in the industry at home and abroad: The technical solution of the present invention adopts deep learning method in predicting the behavior of tourists in scenic spots and estimating the number of tourists in each scenic spot. It is the first of its kind in China in the subdivided field and fills the technical gap in China.

[0036] (3) The technical solution of the present invention solves the technical problem that people have been eager to solve but have never been able to succeed: The technical solution of the present invention solves the problems of the existing scenic spot tourist flow monitoring and management system being imprecise, having poor real-time performance, and lacking predictive data.

[0037] (4) The technical bias overcome by the technical solution of the present invention is that the existing technical solution mainly solves the real-time data of the distribution of people in various scenic spots in the scenic area, while the technical solution of the present invention mainly targets the number of tourists gathered in various scenic spots in the scenic area in the next period of time, and mainly solves the problem of tourist congestion by guiding tourists in advance according to the prediction results.

[0038] Traditional scenic area management methods are inadequate in monitoring and forecasting visitor flow, relying mainly on manual statistics and simple monitoring equipment, which makes it difficult to achieve real-time and accurate collection and analysis of visitor flow data. Especially during peak tourist seasons, the lack of an efficient visitor flow forecasting system can easily lead to problems such as congestion and safety hazards, affecting visitor experience and scenic area operations.

[0039] This invention proposes a visitor flow prediction system based on facial recognition technology and deep learning. It deploys facial recognition cameras at key locations within scenic areas to collect visitor image data in real time. The system utilizes the YOLOv8-MP model and MobileNet network for efficient facial feature recognition and learning, while employing K-means clustering to optimize the recognition of small-sized facial features. Combined with a deep learning-based visitor flow prediction model (DL-TFP), the system can accurately predict visitor flow at various scenic spots, providing a scientific basis for scenic area management.

[0040] The significant technological advancement of this invention lies in its realization of high-precision real-time monitoring and prediction of tourist flow in scenic areas. Through the integration of deep learning and computer vision technologies, the system can not only effectively identify tourist facial features but also predict future tourist flow trends based on historical data, providing dynamic management strategy support for scenic areas. Furthermore, the introduced region-enhanced attention mechanism further improves the model's ability to process spatiotemporal sequence data, making the prediction results more accurate and reliable.

[0041] The application of this visitor flow prediction system has significantly improved the management efficiency and visitor experience of the scenic area. Through real-time monitoring and early warning mechanisms, the scenic area can respond promptly to peak visitor flows, effectively avoiding congestion and safety hazards. At the same time, the system's multi-dimensional data analysis capabilities provide strong data support for scenic area planning, resource allocation, and marketing strategy formulation, contributing to the sustainable development and intelligent upgrading of the scenic area. Attached Figure Description

[0042] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the embodiments of the present invention will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0043] Figure 1 This is a flowchart of the scenic area visitor flow prediction method provided in an embodiment of the present invention;

[0044] Figure 2 This is a structural block diagram of the scenic area visitor flow prediction system provided in an embodiment of the present invention;

[0045] Figure 3 This is a network structure diagram of the tourist flow prediction model provided in an embodiment of the present invention;

[0046] In the diagram: 001, face and behavior recognition camera; 002, optical fiber; 003, optical fiber switch / router; 004, storage server; 005, application server. Detailed Implementation

[0047] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0048] To address the problems existing in the prior art, the present invention provides a method, system, medium, equipment, and terminal for predicting visitor flow in scenic areas. The present invention will be described in detail below with reference to the accompanying drawings.

[0049] Example 1: Facial Recognition and Visitor Flow Prediction System for Large Tourist Attractions

[0050] Equipment configuration:

[0051] 1) Facial recognition camera:

[0052] At the entrance: Deploy no fewer than 3 facial recognition cameras to cover the queuing area and ticket gate area. If the scenic area has multiple ticket gates, increase the number of cameras as needed to ensure that each entrance is covered by more than 3 cameras.

[0053] Exit: Install the appropriate number of facial recognition cameras according to the number of exit gates, ensuring that each exit gate has at least one camera.

[0054] Inside the scenic area: The cameras are distributed according to the length of the tour route, the density of attractions, and the carrying capacity of tourists. Facial recognition cameras must be installed at important attractions (popular spots). The initial section of the trail has a higher density of cameras, while other sections are randomly distributed.

[0055] 2) Network transmission:

[0056] The facial recognition camera is connected to fiber optic switches, routers, and servers via fiber optic cables to ensure efficient and stable data transmission.

[0057] 3) Face recognition algorithm:

[0058] The face recognition camera uses the YOLOv8-MP model and the MobileNet lightweight deep neural network, and uses K-means clustering to optimize the anchors, thereby improving the speed and accuracy of face feature recognition.

[0059] 4) Data storage and analysis:

[0060] Storage server: Configured according to the number of cameras, video parameters, and storage time requirements, with a storage time of not less than 180 days.

[0061] Application server: Stores and runs the visitor flow prediction program, based on the deep learning-based visitor flow prediction model (DL-TFP), to predict and analyze visitor flow within the scenic area.

[0062] Workflow:

[0063] 1) Data acquisition: Facial recognition cameras capture images of tourists and transmit the data to the central control center via fiber optic network.

[0064] 2) Data processing: In the central control center, face recognition is performed using the YOLOv8-MP model and MobileNet to extract facial features and perform matching and statistics.

[0065] 3) Tourist flow prediction: Based on the DL-TFP model, ConvLSTM and regional enhanced attention mechanism are used to predict tourist flow in real time.

[0066] 4) Results Display: The current and predicted number of tourists are displayed on the central control center platform using heat maps, line graphs, bar charts, etc., and early warning information is provided.

[0067] Example 2: Facial Recognition and Visitor Flow Prediction System for Medium-Sized City Parks

[0068] Equipment configuration:

[0069] 1) Facial recognition camera:

[0070] At the entrance: Deploy no fewer than 3 facial recognition cameras at the main entrance and 2 at the secondary entrance of the park, covering the queuing area and the ticket gate area.

[0071] Exits: Depending on the number of exits, install 1-2 facial recognition cameras at each exit to ensure coverage of all exits.

[0072] Inside the park: Cameras are densely deployed at key scenic spots (such as gardens and lakesides) and the starting points of main walking trails, while cameras are randomly deployed in other areas.

[0073] 2) Network transmission:

[0074] All cameras are connected to fiber optic switches, routers, and servers via fiber optic cables to ensure efficient data transmission.

[0075] 3) Face recognition algorithm:

[0076] By using the YOLOv8-MP model and the MobileNet deep neural network, combined with K-means clustering to optimize anchors, the speed and accuracy of face recognition are improved.

[0077] 4) Data storage and analysis:

[0078] Storage server: Configured according to the number of cameras and storage requirements, ensuring that video data is stored for no less than 180 days.

[0079] Application server: Runs a visitor flow prediction program, using the DL-TFP model to predict and analyze visitor flow.

[0080] Workflow:

[0081] 1) Data collection: Facial recognition cameras collect images of visitors at various entrances, exits and inside the park, and transmit them to the central control center via fiber optic network.

[0082] 2) Data processing: In the central control center, face recognition is performed using the YOLOv8-MP model and MobileNet, features are extracted and statistically analyzed.

[0083] 3) Visitor flow prediction: Based on the DL-TFP model, combined with ConvLSTM and regional enhanced attention mechanism, visitor flow in the park is predicted in real time.

[0084] 4) Results Display: The central control center display platform displays the current and predicted number of tourists through heat maps, line graphs and bar charts, and provides early warning information.

[0085] These two specific examples demonstrate how to deploy facial recognition cameras in scenic areas of different sizes and use advanced deep learning algorithms to achieve real-time monitoring and prediction of visitor flow, ensuring efficient scenic area management and visitor safety.

[0086] like Figure 1 As shown, the scenic area visitor flow prediction method provided in this embodiment of the invention includes the following steps:

[0087] S101 uses facial and behavioral recognition cameras to identify and recognize tourists' facial information;

[0088] S102, based on deep learning analysis of tourists' travel habits, to estimate the tourists' positions at various times in the next period of time;

[0089] S103 estimates the actual location of tourists based on their movement positions at various times over a period of time, predicts the dynamic distribution of tourist flow in the scenic area, anticipates peak periods of crowding at special attractions and popular spots within the scenic area, and determines whether to activate the tourist evacuation plan.

[0090] As a preferred embodiment, the scenic area visitor flow prediction method provided by this invention includes the following steps:

[0091] 1. First, based on the tour routes of the scenic area, appropriately deploy cameras with facial and behavioral recognition functions. The cameras can identify faces and perform identity verification through video and image capture, and can also distinguish tourist behavioral characteristics through dynamic analysis.

[0092] 2. The camera's image data is transmitted to the server in real time via fiber optic cable. The server analyzes the tourist information collected by multiple cameras, identifies tourist groups, and determines which group a specific tourist belongs to. The number of people in a group can be 1, 2, 3, ... N, where N is a positive integer.

[0093] 3. The server analyzes the data uploaded by all cameras within the scenic area to determine the specific time interval within the scenic area for all groups entering the area;

[0094] 4. By determining the current location of tourist groups and the behavioral characteristics of tourists in each group, the server determines the estimated number of tourists for each scenic spot and popular spot in the next time period (typically 0.5 hours). If the estimated number of tourists for a scenic spot or popular spot exceeds the threshold for the scenic spot's capacity, an early warning message is sent to the scenic area's operation and management department, requesting the activation of the tourist diversion plan.

[0095] 5. The visitor flow management plan is a pre-set procedure at the scenic area's control center. Specifically, it uses voice IP broadcasts along the tour route to remind visitors to adjust their pace. There are two methods: Method 1: For crowds gathering at popular spots or groups moving slowly, the voice IP broadcast reminds them, "The scenery ahead is even more beautiful, keep going!" Method 2: For attractions that visitors are not paying much attention to, have missed, or have ample capacity, or groups moving quickly, the voice IP broadcast reminds them, "Tired? Stop and breathe in the pure air with its high concentration of negative oxygen ions!"

[0096] 6. The server updates the estimated visitor volume for each attraction in real time based on the current location data of all visitors in the park, and adjusts the visitor flow management plan in real time. The flow management prompts broadcast by each voice IP are adjusted in real time according to the behavioral characteristics of the current group of visitors. If the estimated visitor volume for all attractions is lower than the visitor capacity threshold for that attraction, no more voice prompts will be given.

[0097] like Figure 2 As shown, the scenic area visitor flow prediction system provided in this embodiment of the invention includes: facial and behavioral recognition cameras installed at the scenic area entrance, walkways, special locations, attractions (including popular spots), and exits; an optical fiber network; an optical fiber switch; a storage server; and an application server.

[0098] Among them, face and behavior recognition cameras include fixed-angle bullet cameras and PTZ cameras with pan-tilt units. PTZ cameras are mainly used for high-altitude shooting and tracking shooting.

[0099] Face and behavior recognition cameras can identify face information. Once the identification is successful, the information is immediately uploaded to the server. The server assigns a unique identifier to all identified face information of the day and broadcasts the identifier and face image information to each face recognition camera. The update interval is no more than 10 seconds per update.

[0100] The face and behavior recognition camera stores the facial information and identification number broadcast by the server in its local storage. When a new face enters the lens and the visitor's behavior is captured, it is compared with the facial information in the storage. If the face already exists, the visitor's behavioral characteristics are stored according to the identification number and facial information saved in the local storage, and the video and recognition data are transmitted to the server in real time via optical fiber. If the visitor's facial information and identification number are not in the local storage, the video, facial recognition, and visitor behavioral characteristic information are directly transmitted to the server. The local storage only stores the visitor's identification code, the corresponding facial image information, and the visitor's behavioral characteristic information, and is completely cleared at 24:00 every day.

[0101] Fiber optic switches are used to connect all facial and behavioral recognition cameras in the scenic area to the server;

[0102] The server is divided into application server and storage server, which are directly connected to fiber optic switches;

[0103] The storage server has sufficient disk arrays to meet the scenic area's storage requirements for video data;

[0104] The application server has sufficient computing speed and stores programs to identify the facial information collected by each face and behavior recognition camera, and to calculate the estimated number of tourists for each attraction at each moment in the next period of time in real time, and compare it with the tourist capacity threshold determined in advance for that attraction.

[0105] The application server broadcasts all the identified identity numbers and facial information for the day to each facial recognition camera periodically via a fiber optic switch, with an update cycle of no more than 10 seconds.

[0106] The technical solution of this invention is mainly used for monitoring tourist flow in scenic areas, specifically for predicting the number of tourists at each attraction within the scenic area over a period of time, providing data support for the scientific and effective management of tourist flow in the scenic area.

[0107] To facilitate visitor access, scenic areas typically design tour routes in one-way or loop directions. Therefore, it's relatively easy to deploy cameras along these routes for image capture. Initially, at least three facial recognition cameras should be deployed from the queuing area at the entrance to the ticket gates. If the scenic area is large, additional cameras may be needed. For each ticket gate, no fewer than [number] facial recognition cameras must be installed in this area. There are [number] exit gates at the scenic area exit. Install Personal facial recognition cameras. These are installed and deployed within the scenic area, based on the length of the tour route, the density of attractions, and the tourist capacity. Personal facial recognition cameras, the total number of cameras is indivual.

[0108] The embodiments of the present invention provide an installation inside the scenic area. The facial recognition cameras are not uniformly distributed, requiring that facial recognition cameras be installed in important scenic spots (popular tourist spots). The density of cameras installed in the initial section of the trail is higher than that in other sections, while the cameras in other sections are randomly distributed.

[0109] The facial recognition camera is connected to fiber optic switches, routers, and servers via fiber optic cables.

[0110] The face recognition camera uses the YOLOv8-MP model for face feature learning and recognition. It is a lightweight deep neural network MobileNet built on depthwise separable convolutions and uses K-means clustering to reset the anchors to speed up the model convergence, thereby improving the recognition speed of face features.

[0111] The output size of the YOLOv8-MP model used in the face recognition camera It can be represented as:

[0112]

[0113] in, Enter the dimensions of the image. The size of the convolution kernel. This represents the number of padding elements needed for the convolution. The step size.

[0114] The face recognition camera provided in this embodiment of the invention uses the lightweight deep neural network MobileNet, which decomposes the standard convolution operation into a depthwise convolution and a depthwise convolution. Point convolution.

[0115] To avoid the difficulty of recognizing small facial features due to factors such as camera angle and depth of field, this embodiment of the invention uses the K-means clustering algorithm to obtain the optimal number of anchors and facial feature dimensions. The clustering distance is determined by the overlap ratio (Intersection over Union). This is used to reflect the error between the eigenvalue and the true value (the initial region uses the eigenvalue), and the error formula is: .

[0116] in, As the center of all clusters, The results of clustering based on sample features Indicates all sample features and The intersection and union ratio.

[0117] The application server provided in this embodiment of the invention is loaded with a crowd flow prediction program. Its main data source is facial recognition cameras installed at scenic area entrances, walkways, and exits. The crowd flow prediction program is based on a deep learning-based crowd density prediction model. The hardware system framework required by the model has a three-layer structure: the first layer is the data acquisition layer, mainly composed of facial recognition cameras installed at scenic area entrances, walkways, and exits; the second layer is the network transmission layer, mainly composed of optical fibers, optical fiber switches, and routers; and the third layer is the application layer, mainly composed of storage servers and application servers.

[0118] The storage capacity of the storage server is calculated based on the number of cameras, video parameters, and storage time requirements. To achieve better data analysis results, the storage time should be no less than 180 days.

[0119] The application server stores a program for predicting visitor flow, which uses a deep learning-based tourist flow prediction model (DL-TFP).

[0120] The network structure diagram of the deep learning-based tourist flow prediction model is as follows: Figure 3 As shown.

[0121] The visitor flow prediction system provided in this embodiment of the invention numbers all facial recognition cameras in a scenic area based on their installation location and IP address, totaling [number missing]. There are [number] cameras, and the image sequence for each camera is [number]. The image sequence is the input to the system.

[0122] The tourist flow prediction model (DL-TFP) provided in this embodiment of the invention uses convolution operations to capture the spatial correlation between different image acquisition points, i.e., the lens areas of face recognition cameras, and uses a ConvLSTM model to calculate their temporal correlation. The tourist flow prediction system treats the entire scenic area as an independent space and considers the interrelationship of the spatial areas of all camera acquisition areas.

[0123]

[0124] in, For the first The area contained in the image of a personal facial recognition camera The output of the convolutional layer, For the first The input of a convolutional layer, express The weight matrix, Indicates the bias term. This indicates that the modified linear unit is used as the activation function.

[0125] The tourist flow prediction system provided in this embodiment of the invention uses a Convolutional Short-Term Memory (ConvLSTM) model, which is a recurrent neural network capable of processing high-dimensional spatial sequence data and modeling the temporal dependencies in the sequence.

[0126] The tourist flow prediction system provided in this embodiment of the invention employs a region-enhanced attention mechanism to describe the spatial correlation between frames of different time steps.

[0127]

[0128]

[0129]

[0130] in, This represents the attention score function. and Represents the learning parameters, This represents the rate of change in the number of people output based on a single-head attention mechanism. Represents the query vector. Represents the key vector. Represents a value vector. This represents the numerical value of a person based on a multi-head attention mechanism. This represents the learning parameters.

[0131] The tourist flow prediction system provided in this embodiment of the invention calculates the attention scores of the learned features using weighted summaries to predict the first... indivual( (The total number of attractions to be predicted is...) (The point has passed from the current time) Number of tourists after the time Warning time The value range of h is (0, 24).

[0132]

[0133]

[0134] The application server equipped with the visitor flow prediction system provided in this embodiment of the invention is connected to the scenic area's centralized control center. The centralized control center's display platform shows the current number of visitors and warning time for each attraction in the scenic area through heat maps, line graphs, bar charts, and other methods. (Default value is 0.5h) The number of visitors to each attraction at any given time. For a specific attraction, you can view the curve of the number of visitors to the attraction from the opening time in the morning (which can be set by the user) to the current time, as well as the estimated number of visitors at each time point from the current time to the closing time of the same day (which can be set by the user), and the three curves of the optimal number of visitors and the maximum number of visitors for the attraction.

[0135] It should be noted that embodiments of the present invention can be implemented in hardware, software, or a combination of both. The hardware portion can be implemented using dedicated logic; the software portion can be stored in memory and executed by a suitable instruction execution system, such as a microprocessor or dedicated-design hardware. Those skilled in the art will understand that the above-described devices and methods can be implemented using computer-executable instructions and / or included in processor control code, for example, such code provided on a carrier medium such as a disk, CD, or DVD-ROM, a programmable memory such as read-only memory (firmware), or a data carrier such as an optical or electronic signal carrier. The devices and modules of the present invention can be implemented by hardware circuitry such as very large-scale integrated circuits or gate arrays, semiconductors such as logic chips, transistors, or programmable hardware devices such as field-programmable gate arrays, programmable logic devices, etc., or by software executed by various types of processors, or by a combination of the above-described hardware circuitry and software, such as firmware.

[0136] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any modifications, equivalent substitutions, and improvements made by those skilled in the art within the scope of the technology disclosed in the present invention, and within the spirit and principles of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for predicting visitor flow in scenic areas, characterized in that, The methods for predicting visitor flow in scenic areas include: using facial and behavioral recognition cameras to identify groups and recognize the identities of tourists entering the scenic area; analyzing tourists' travel habits based on deep learning to estimate their location at each subsequent moment; predicting the dynamic distribution of visitor flow in the scenic area, anticipating peak periods for crowds at attractions and popular spots, and determining whether to activate visitor evacuation plans. The method for predicting visitor flow in scenic areas includes the following steps: Step 1: Deploy facial and behavioral recognition cameras according to the scenic area's tour routes. Collect and identify faces through video and image data and perform identity verification. Analyze the dynamic characteristics of tourists to distinguish their behavior. Step two: The camera's image data is transmitted to the server in real time via fiber optic cable. The server analyzes the tourist information collected by multiple cameras and identifies tourist groups. Step 3: The server analyzes the data uploaded by all cameras in the scenic area to determine the specific time interval of each group entering the scenic area. Step 4: The server determines the number of tourists at each high-density attraction at the current time and in the next few moments by determining the current location of the tourist groups and the behavioral characteristics of the tourists in each group, and determines whether to activate the tourist diversion plan. Step 5: The server updates the estimated number of tourists at each attraction in real time based on the actual location data of the tourists, adjusts the tourist flow management plan in real time, and determines whether to issue voice prompts. In step four, the server determines the estimated number of tourists at each scenic spot and popular attraction in the next time period by identifying the current location of the tourist groups and the behavioral characteristics of the tourists in each group. If the estimated number of tourists at a scenic spot or popular attraction exceeds the threshold for the attraction's capacity, an early warning message is sent to the scenic area's operation and management department, requesting the activation of the tourist diversion plan.

2. The method for predicting visitor flow in scenic areas as described in claim 1, characterized in that, In step four, the visitor management plan is a pre-set procedure at the scenic area's control center. It involves using voice IP broadcasts along the tour route to remind visitors to adjust their tour schedule, including: (1) Announce voice IP broadcasts to remind people who gather at popular tourist spots or groups that visit slowly; (2) Provide voice IP broadcast reminders for attractions that tourists do not pay much attention to, miss, or have sufficient capacity, or for groups that visit too quickly.

3. The method for predicting visitor flow in scenic areas as described in claim 1, characterized in that, In step five, the server updates the estimated number of tourists at each attraction in real time based on the actual movement data of the tourists, and adjusts the tourist flow management plan in real time; the flow management prompts broadcast by each voice IP are adjusted in real time according to the behavioral characteristics of the tourists in the current group. If the estimated number of tourists at all attractions is lower than the tourist capacity threshold of the attraction, the voice prompts will no longer be given.

4. A scenic area visitor flow prediction system applying the scenic area visitor flow prediction method as described in any one of claims 1 to 3, characterized in that, The scenic area visitor flow prediction system includes facial and behavioral recognition cameras, fiber optic networks, fiber optic switches, storage servers, and application servers; Face and behavior recognition cameras are installed at scenic area entrances, walkways, venues, attractions, and exits, including fixed-angle bullet cameras and PTZ cameras with pan-tilt heads. PTZ cameras are used for aerial shooting and tracking shooting. Fiber optic switches are used to connect all facial and behavioral recognition cameras in the scenic area to the server; The server is divided into application server and storage server, which are directly connected to fiber optic switches; The storage server includes a disk array to meet the scenic area's requirements for storing video data; The application server has computing speed and stores programs to identify the facial information collected by each face and behavior recognition camera, calculate the estimated number of tourists for each attraction at each moment in the next period of time in real time, and compare it with the tourist capacity threshold determined in advance for the attraction. The application server broadcasts all the identified identity numbers and facial information for the day to each facial recognition camera periodically via a fiber optic switch, with an update cycle of no more than 10 seconds.

5. The scenic area visitor flow prediction system as described in claim 4, characterized in that, The face and behavior recognition cameras calibrate face information and upload it to the server immediately after successful calibration. The server assigns an identity number to all calibrated face information of the day in real time and broadcasts the identity number and face image information to each face recognition camera. The update interval is no more than 10 seconds. The face and behavior recognition camera stores the facial information and identification number broadcast by the server in its local storage. When a new face enters the lens, it captures the visitor's behavior and compares it with the facial information in the storage. If the face already exists, the camera stores the visitor's behavioral characteristics according to the identification number and facial information stored in the local storage, and transmits the video and recognition data to the server in real time via fiber optic cable. If the visitor's facial information and identification number are not in the local storage, the camera directly transmits the video, facial recognition, and visitor behavioral characteristic information to the server. The local storage only stores the visitor's identification code, the corresponding facial image information, and the visitor's behavioral characteristic information, and is completely cleared at 24:00 every day.

6. A computer device, characterized in that, The computer device includes a memory and a processor. The memory stores a computer program. When the computer program is executed by the processor, it causes the processor to perform the steps of the scenic area visitor flow prediction method as described in any one of claims 1 to 3.

7. A computer-readable storage medium storing a computer program, which, when executed by a processor, causes the processor to perform the steps of the scenic area visitor flow prediction method as described in any one of claims 1 to 3.

8. An information data processing terminal, characterized in that, The information data processing terminal is used to implement the scenic area visitor flow prediction system as described in any one of claims 4 to 5.

Images