A Smart Wayfinding Method and System for Urban Parks

Abstract

This invention relates to the field of smart park technology, and more particularly to a smart guidance method and system for urban parks. The system includes: responding to a guardian activation command by collecting temporary characteristic information of children to generate a temporary feature code; monitoring video streams in real time and sending an alert when an unaccompanied child enters an electronic fence area; responding to a lost child assistance command by retrieving the temporary feature code and defining a search area; retrieving video streams within the search area and generating a list of suspected targets based on the temporary feature code using structured human body analysis; responding to an administrator's confirmation command for suspected targets by obtaining the target's real-time location; planning a dynamic navigation route for visitors based on the real-time location and sending assistance commands to nearby administrators; and automatically deleting the temporary feature information in response to a search ending command or the end of a park visit period. This invention, through pre-registered features, proactive alerts, intelligent search, and two-way guidance, achieves proactive prevention and accurate search for lost children, improving search efficiency and privacy protection.

Description

Technical Field

[0001] This invention relates to the field of smart park technology, and in particular to a smart wayfinding method and system for urban parks. Background Technology

[0002] With the acceleration of urbanization and the deepening of smart city construction, urban parks, as important public spaces for citizens' leisure and recreation, are seeing continuous improvement in their intelligent management. Currently, many parks have deployed infrastructure such as smart navigation screens, official mini-programs, and surveillance cameras to provide visitors with services such as attraction navigation, information inquiry, and visitor flow statistics. During holidays, park visitor numbers surge, and incidents of children getting separated from their parents occur frequently. Existing smart navigation systems mainly focus on attraction navigation and information dissemination, lacking proactive protection and precise search functions for emergency scenarios such as children getting lost.

[0003] Because lost children are usually young and have limited cognitive abilities, they may have difficulty understanding announcements or finding their parents in crowded and noisy outdoor environments, leading to inefficient search efforts. Furthermore, parents, in their panic, may verbally describe their child's features to staff, resulting in unclear descriptions and omissions of crucial information, further complicating the search. This passive and vague search method fails to meet the urgent need for child safety protection during peak holiday periods. Summary of the Invention

[0004] To overcome the above shortcomings, this invention provides a smart wayfinding method and system for urban parks, aiming to improve the existing methods of finding people in parks, which rely on manual broadcasts, make it difficult for children to understand the broadcast content in noisy environments, and cause low search efficiency due to unclear descriptions from parents.

[0005] In a first aspect, the present invention provides the following technical solution: a smart wayfinding method for urban parks, comprising the following steps: S1. In response to the guardian activation command triggered by tourists through the park's official mini-program or smart guide screen, collect and bind the child's temporary characteristic information, and generate a temporary characteristic code associated with the tourist's account. S2. Based on the surveillance cameras deployed in the park, the system acquires real-time video streams of visitors and uses the background AI analysis system to detect whether unaccompanied children have entered the preset electronic fence area. If so, the system sends an early warning message containing the scene to the administrator's handheld terminal for manual review and intervention. S3. In response to a lost person request triggered by a tourist through a mini-program or smart guide screen, retrieve the temporary feature code bound to the tourist's account and delineate a search area starting from the location where the request was triggered. S4. Retrieve the real-time video stream from the surveillance cameras within the search area, perform a structured human body search based on the temporary feature code, generate a list of suspected targets, and push it to the administrator's handheld terminal. S5. In response to the administrator's confirmation command for the suspected target on the handheld terminal, obtain the real-time location of the target; S6. Based on the real-time location, plan and display a dynamic navigation route for the tourist's mini-program, and at the same time send an assistance instruction to the administrator's handheld terminal near the real-time location. S7. In response to a visitor's command to end the search for a person triggered by a mini-program or smart guide screen, or after the preset park visit period ends, automatically delete the temporary feature information related to this park visit.

[0006] Preferably, in step S1, the step of collecting and binding the child's temporary characteristic information and generating a temporary feature code associated with the visitor account specifically includes: In response to the guardian activation command triggered by tourists through the park's official mini-program or smart guide screen, a feature collection menu is displayed on the interactive interface, which includes at least one clothing color option. In response to the visitor's instruction to select the clothing color option, obtain the child's clothing color information; In response to a visitor's photo upload command triggered through the interactive interface, a photo of the child's outfit for the day is obtained; The clothing color information and the photo of the outfit worn on the day are used as temporary feature information, which are then bound to the current visitor's account to generate a temporary feature code, and the current time is recorded as the entry time.

[0007] Preferably, in step S2, the step of detecting whether an unaccompanied child has entered the preset electronic fence area through the background AI analysis system specifically includes: Based on the surveillance cameras deployed in the park, real-time video streams of park visitors are obtained from the preset electronic fence area and its surroundings; The background AI analysis system analyzes the park visit video stream in real time to identify child and adult targets in the video footage. Based on the relative positions of the child target and the adult target in the image, determine whether there is an associated adult target within a preset range around the child target; When there is no associated adult target within a preset range around the child target, the child is determined to be unaccompanied by an adult, and an early warning message containing the child's location and the scene is generated.

[0008] Preferably, in step S3, the step of defining the search area starting from the location that triggered the distress call specifically includes: In response to lost and need-help requests triggered by visitors through the park's official mini-program or smart guide screen, the system obtains the current visitor's account information and the location coordinates of the location that triggered the request. Based on the visitor account information, a temporary feature code bound to the account is retrieved from the backend database; the temporary feature code contains the color information of the children's clothing collected during the pre-registration for park entry. Using the location coordinates of the triggered distress call as the center, a circular search area is defined according to a preset radius distance, or the park area where the location coordinates are located and adjacent areas are defined as the search area.

[0009] Preferably, in step S4, the step of generating a list of suspected targets and pushing it to the administrator's handheld terminal specifically includes: Based on the defined search area, obtain the device identifiers of all surveillance cameras within that area and retrieve their real-time video streams; The color information of the child's clothing is extracted from the temporary feature code and used as the target feature for human body structured search; The background AI analysis system performs human detection and attribute analysis on the real-time video stream to identify all child targets in the video and their corresponding clothing color attributes. The clothing color attribute of the child target is compared with the target features, and child targets with matching clothing color attributes are selected as suspected targets. Obtain the real-time location and on-site screenshot of each suspected target, and generate a list of suspected targets containing target images, location information and timestamps; The list of suspected targets is pushed to the administrator's handheld terminal for manual confirmation.

[0010] Preferably, in step S5, the step of obtaining the real-time location of the target specifically includes: In response to the administrator's confirmation click command on a suspected target image in the list of suspected targets on the handheld terminal, the target identifier corresponding to the suspected target is obtained; Based on the target identifier, real-time video stream analysis data associated with the target is retrieved from the background AI analysis system; Extract the current frame of the target from the real-time video stream analysis data, and parse the image coordinates of the target in the frame; Based on the device identifier of the surveillance camera that captured the current frame and its preset spatial location parameters, the image coordinates are converted into actual geographical location coordinates on the park map, which serve as the real-time location of the target.

[0011] Preferably, in step S6, the step of planning and displaying a dynamic navigation route for the tourist's mini-program, and simultaneously sending assistance instructions to the administrator's handheld terminal near the real-time location, specifically includes: Based on the real-time location of the target and the current location of the tourist account that triggered the missing person assistance request, the optimal walking route from the current location to the real-time location is calculated by the background path planning engine. The optimal walking route is pushed to the tourist's mini-program in the form of dynamic navigation instructions, and the navigation path and direction guidance are rendered and displayed in real time on the map interface of the mini-program. Based on the real-time location, query the handheld terminals of administrators who are on duty within a preset range around the location, and generate an assistance instruction that includes the real-time location of the target, a screenshot of the scene, and a description of the assistance task. The assistance command is pushed to the queried administrator's handheld terminal, and the real-time location and recommended route are marked on the terminal's map interface.

[0012] Preferably, in step S7, the step of automatically deleting the temporary feature information related to this visit specifically includes: In response to a visitor's command to end the search for a person triggered by the park's official mini-program or smart guide screen, the system obtains the visitor's account information and the time the command was triggered. Based on the visitor account information, locate the temporary feature information bound to the account and its storage path from the backend database; Send a data deletion request to the database management system to clear the temporary feature information under the storage path and unbind the information from the visitor account; or when the preset park visit period ends, the system automatically scans all temporary feature information in the database whose creation time is earlier than the end time, performs batch deletion operations, and releases storage space.

[0013] Secondly, the present invention provides the following technical solution: a smart guidance system for urban parks, the system comprising: The park entry pre-registration module is used to respond to the guardian activation command triggered by tourists through the park's official mini-program or smart guide screen, collect and bind the child's temporary characteristic information, and generate a temporary characteristic code associated with the tourist's account. The park visitor monitoring module is used to acquire real-time video streams of visitors based on surveillance cameras deployed in the park. The back-end AI analysis system detects whether unaccompanied children have entered the preset electronic fence area. If so, it sends an early warning message containing the scene to the administrator's handheld terminal for manual review and intervention. The lost and found assistance module is used to respond to lost and found assistance commands triggered by tourists through mini-programs or smart guide screens, retrieve the temporary feature code bound to the tourist's account, and delineate the search area starting from the location where the assistance was triggered. The intelligent search module is used to retrieve real-time video streams from surveillance cameras within the search area, perform structured human body searches based on the temporary feature codes, generate a list of suspected targets, and push it to the administrator's handheld terminal. The target confirmation module is used to respond to the administrator's confirmation command for a suspected target on a handheld terminal and obtain the real-time location of the target; The navigation guidance module is used to plan and display dynamic navigation routes for tourists' mini-program terminals based on the real-time location, and at the same time send assistance instructions to the handheld terminals of administrators near the real-time location. The data clearing module is used to automatically delete the temporary feature information related to the current visit in response to a visitor's command to end the search via a mini-program or smart guide screen, or after the preset visit period has ended.

[0014] The present invention has the following beneficial effects: 1. In this invention, children's clothing color information and photos of their attire on the day of their visit are collected during the pre-registration process, generating a temporary feature code linked to the visitor's account. When a child goes missing, the system directly retrieves the accurate features from the pre-registration for an AI video search, avoiding the problems of unclear verbal descriptions and missing information by parents in their panic. Simultaneously, the background AI analysis system performs a structured human body search on the real-time video streams from surveillance cameras within the search area, quickly filtering out suspected targets that match the features. After manual confirmation by the administrator, the real-time location is obtained, upgrading the traditional passive and vague broadcast search to an active and precise video search, significantly improving the success rate of finding missing persons and the response speed.

[0015] 2. In this invention, an electronic fence area is preset, and surveillance cameras monitor in real time whether children enter dangerous areas such as water bodies, entrances / exits, and steep slopes alone. Once an unaccompanied child is detected approaching, an early warning message containing the scene is immediately sent to the administrator's handheld terminal so that staff can intervene in a timely manner. This proactive early warning mechanism shifts the focus of safety protection from passive searching after a child gets lost to proactive prevention before they get lost, effectively reducing the probability of children getting lost and safety accidents.

[0016] 3. In this invention, after confirming the real-time location of a lost child, a dynamic navigation route is simultaneously planned and displayed on the parent's mini-program, guiding the parent to quickly reach the child's location. At the same time, an assistance instruction containing the target's real-time location, a screenshot of the scene, and a description of the assistance task is sent to the handheld terminal of an administrator near that location, notifying security personnel to go to the scene to care for the child. This avoids situations where parents search blindly or security guards passively wait, significantly improving the collaborative efficiency of emergency response. Attached Figure Description

[0017] Figure 1This is a flowchart illustrating a smart wayfinding method for urban parks proposed in this invention. Figure 2 This is a schematic diagram of the architecture of a smart wayfinding system for urban parks proposed in this invention. Detailed Implementation

[0018] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1: In a first embodiment of the present invention, the present invention provides a smart wayfinding method for urban parks, such as... Figure 1 As shown, it includes the following steps: S1. In response to the guardian activation command triggered by tourists through the park's official mini-program or smart guide screen, collect and bind the child's temporary characteristic information, and generate a temporary characteristic code associated with the tourist's account.

[0020] Specifically, visitors can activate the "Child Protection" button via the park's official mini-program or smart navigation screen. Upon receiving the activation command, the system displays a feature collection menu on the interactive interface. The menu includes a clothing color option area and a photo upload portal. The clothing color option area displays common color options such as red, yellow, blue, green, white, and black in color blocks, each corresponding to a preset color label such as "red" or "blue." The photo upload portal provides "Take a Photo and Upload" and "Select from Album" buttons for visitors to upload photos of their child's clothing for the day. When a visitor clicks on a color option, the system retrieves the corresponding color label as the child's clothing color information. After the visitor selects and confirms the photo, the system retrieves a photo of the child's clothing for the day, in JPEG or PNG format with a resolution of at least 720P. It should be noted that photo upload is an optional step; visitors can complete the feature pre-registration solely based on clothing color information.

[0021] The system uses the acquired clothing color information and the photo of the outfit worn that day as temporary feature information, binding it to the current visitor's account. The specific implementation is as follows: Data encapsulation: The temporary feature information is encapsulated into a structured data object, containing fields such as: account ID, clothing color label, photo storage path, and creation timestamp; Generating a unique identifier: A 32-bit unique string is generated using the UUID algorithm as the temporary feature code, ensuring that each registration record has global uniqueness. The formula for generating the UUID is: ; Where timestamp is the current timestamp in milliseconds. This is the node identifier in the server cluster that handles the request, and the sequence is an auto-incrementing sequence number within the same millisecond. Data storage: The temporary feature code is used as the primary key and stored together with the aforementioned structured data objects on the park's local server or in the government cloud database. The current time is also recorded as the entry time, accurate to the second.

[0022] The system returns a successful registration message to the visitor's interface and displays a summary of the child's information (including clothing color tags and photo thumbnails) in the visitor's personal center on the mini-program or large screen. At this point, the pre-registration process for park entry is complete, and the temporary identifier can be retrieved for assistance in case the child gets lost later.

[0023] S2. Based on the surveillance cameras deployed in the park, real-time video streams of visitors are acquired. The background AI analysis system detects whether unaccompanied children have entered the preset electronic fence area. If so, an early warning message containing the scene is sent to the administrator's handheld terminal for manual review and intervention.

[0024] Specifically, park managers use a polygon drawing tool on the map interface of the smart management backend to delineate high-risk areas within the park, such as the edges of waterways, entrance and exit passages, steep slopes, and secluded forest areas, as electronic fence zones. Each electronic fence zone is stored in the backend database as a set of latitude and longitude coordinates, represented as: ; in, and These are the longitude and latitude coordinates of the i-th boundary point, respectively, and n is the number of polygon vertices constituting the region. Each electronic fence region is also associated with a region type label, such as "water area" or "entrance / exit," for the subsequent classification and display of early warning information.

[0025] The system, based on a pre-defined electronic fence area, selects cameras from surveillance cameras deployed within the park whose field of view covers the area and its surrounding buffer zone. The buffer zone is typically set to extend 5 to 10 meters outward from the electronic fence boundary to ensure early detection of children about to enter the danger zone. The system continuously acquires real-time video streams from these cameras at a frame rate of 25 frames per second via the RTSP real-time streaming protocol or the GB / T 28181 national standard protocol. Each frame of video image has a resolution of no less than 1080P and carries the camera's device ID and timestamp. The background AI analysis system performs frame-by-frame analysis of the acquired real-time video stream, specifically performing the following processing: Object detection: Using a deep learning-based convolutional neural network object detection model, such as YOLOv8 or Faster R-CNN, object detection is performed on each frame to identify pedestrian targets in the image. The detection model outputs the bounding box coordinates, object category, and confidence score for each target. The bounding box coordinates are represented as follows: ,in and The pixel coordinates of the top-left corner of the bounding box in the image; and The coordinates are shown in the lower right corner. Target classification: For each detected pedestrian target, attribute analysis is further performed using an age estimation model and a gender recognition model to classify it as a child target or an adult target. The age estimation model infers based on facial features or body proportions; if the estimated age is less than 12 years old, it is marked as a child target; otherwise, it is marked as an adult target. Each target is assigned a unique target identifier (TargetID). Target location mapping: For each detected target, based on the camera device identifier (DeviceID) that captured the frame and its preset monitoring area geographic coordinate range, the target's pixel coordinates in the image are converted into actual geographic location coordinates. The conversion formula is: ; in, , where is the pixel coordinate of the target center point; F is the mapping function from pixel coordinates to geographic coordinates established based on camera calibration parameters, and the specific parameters include the camera installation position, lens focal length, pitch angle and horizontal angle.

[0026] The system determines whether a child is accompanied by an adult based on the spatial relationship between the child and adult targets within the frame. The specific determination logic is as follows: Spatial distance calculation: For each child target, calculate the Euclidean distance between it and all adult targets in the frame. The distance calculation formula is: ; in, and The pixel coordinates of the center point of the child's target. and t represents the pixel coordinates of the center point of the adult target.

[0027] Association Determination: A spatial distance threshold D is set. In this embodiment, D is 15% of the image height or 100 pixels. When the distance d between a child target and an adult target is less than the threshold D, the adult target is determined to be an accompanying adult associated with the child. Simultaneously, the system also considers temporal information. If the same child and the same adult maintain a similar spatial distance in multiple consecutive frames, their companionship relationship is further confirmed. Unaccompanied Adult Determination: When there are no associated adult targets within a preset range around the child target, i.e., for all adult targets d ≥ D, the child is determined to be unaccompanied.

[0028] When the system detects an unaccompanied child entering the electronic fence area, it immediately generates an alert and pushes it to the administrator's handheld terminal. The alert includes the child's real-time location, a screenshot of the scene, a timestamp, the area type, and camera identification. The screenshot is marked with a red border around the child. The system sends the alert in real-time to the administrator's handheld terminal, which is currently online and on duty, via a WebSocket connection. Upon receiving the alert, the administrator's handheld terminal displays a pop-up window and vibration notification, and shows the child's location on a map. The administrator then determines whether intervention is necessary: ​​if it's a false alarm (e.g., the adult is obstructed), they can click "ignore" to close the alert; if it's a genuine danger, they can click "go to handle," and the system automatically plans the optimal route from the administrator's current location to the child's location and notifies nearby security guards for assistance.

[0029] S3. In response to a lost person request triggered by a tourist through a mini-program or smart guide screen, retrieve the temporary feature code bound to the tourist's account and delineate the search area starting from the location where the request was triggered.

[0030] Specifically, when a visitor discovers a child is missing, they can initiate a request for help through the "Missing Child Help" button on the park's official mini-program homepage or the "Emergency Help - Missing Child" touch area on the smart guide screen. The system responds to the missing child help command by parsing the visitor's account information and the location coordinates that triggered the help request. When initiating a help request through the mini-program, the system obtains the currently logged-in user's account identifier from the session context and calls the WeChat mini-program API to obtain the visitor's phone's GPS location data to obtain the longitude coordinates. and latitude coordinates When initiating a request for assistance through the smart navigation screen, visitors must first scan the screen's QR code to complete identity authentication. The system then obtains the visitor's account identifier temporarily bound to the QR code and reads the pre-recorded coordinates of the screen's fixed installation location as the location for requesting assistance. To prevent GPS location drift, the system filters the acquired coordinates. When the standard deviation of the distance between three consecutive location coordinates is less than 5 meters, the average value is taken as the final location coordinates.

[0031] The system performs a related query in the background database based on the acquired visitor account information. The database uses the visitor account as the primary key and stores the temporary feature code generated when the account enters the park, along with the corresponding temporary feature information. After a successful query, the system loads the temporary feature code into memory for use in subsequent search steps. The temporary feature code contains at least the child's clothing color information, stored in a standardized color label format, such as "red," "blue," and "yellow." The system delineates the search area centered on the location coordinates that triggered the request for help, according to a preset strategy. This embodiment provides two optional area delineation methods, which can be configured according to the actual layout and management needs of the park. Method one is: circular search area delineation: the system uses the location coordinates that triggered the request for help (… , Using a circle as the center, a circular search area is defined with a preset radius R. The default value for radius R is 500 meters, but park management personnel can adjust it in the backend management system according to the park area and visitor density, with a range of 200 to 1000 meters. The mathematical representation of the circular search area is: ; The distance function is the Haversine formula for calculating the geographic distance between two points: ; in, The average radius of the Earth is taken as 6371 kilometers. ; lon All are radian values.

[0032] Method 2: Area Search Delineation: For parks that have been divided into areas, the system obtains the AreaID, the park area identifier located at the coordinates of the triggered request location. Park areas are administrative divisions pre-defined on the map by the park management, such as "Children's Play Area," "Lakeside Sightseeing Area," and "Forest Recreation Area." Based on the area topology, the system queries all area identifiers adjacent to the current area, forming a search area set. ; The search area is the entire geographical area covered by the set of regions. The region boundaries are also stored in the backend database as a set of polygon vertex coordinates. The system encapsulates the defined search area information into a search area object, containing parameters such as region type identifiers (e.g., "circle" or "district"), region boundary coordinate sets, center point coordinates, and radius, for use in subsequent step S4. After completing the above steps, the system returns a confirmation message to the visitor's interface that the request for help has been accepted, prompting the visitor to "Searching for your child; please keep your phone accessible." Simultaneously, the system packages the visitor's account information, temporary feature code, search area object, and request time into a search task, placing it in the task queue to await execution of the intelligent search process in step S4.

[0033] S4. Retrieve real-time video streams from surveillance cameras within the search area, perform structured human body search based on temporary feature codes, generate a list of suspected targets, and push it to the administrator's handheld terminal.

[0034] Specifically, based on a defined search area, the system performs a spatial query in the background database to retrieve the device identifiers of all surveillance cameras within that area. The camera device table `cameras` contains fields... Longitude (London), Latitude (Lattor), Installation Location Description RTSP stream address and monitoring area type The system retrieves the corresponding RTSP real-time stream address from the camera management service based on the device identifier of each camera in CameraList. For each camera, the system establishes an independent streaming session using FFmpeg or a similar multimedia processing framework, continuously acquiring real-time video streams at a frame rate of 25 frames per second. Each video stream session is assigned a unique session identifier (SessionID) and is associated with the camera device identifier. The current timestamp is bound and stored in the video stream session table in memory.

[0035] The system extracts the color information of the child's clothing from the temporary feature code. (Read) The values ​​of the field serve as target features for structured human body search. In this embodiment, clothing color information is represented using standardized color labels, including 10 common colors such as red, blue, yellow, green, white, black, pink, purple, orange, and gray. The background AI analysis system processes the real-time video stream of each camera session frame by frame, performing the following analysis steps: Human target detection: A deep learning-based YOLOv8 target detection model is used to detect each frame, outputting the bounding box coordinates and confidence scores of all human targets. The bounding box coordinates are represented as follows: , , , All values ​​are pixel coordinates. The detection model sets a confidence threshold of 0.7, retaining only targets with a confidence score greater than 0.7. Age estimation and child target screening: For each detected humanoid target, the facial proportions and body contour features are analyzed using an age estimation model to output an estimated age value. When the estimated age is less than 12 years old, the target is marked as a child target and assigned a temporary target identifier TempTargetID. Clothing color attribute extraction: For each marked child target, the system extracts an image patch of the clothing portion from its bounding box region. The clothing portion is usually located in the lower half of the bounding box, i.e., from... +0.3×( - )arrive -0.1×( - The image patch is then analyzed using a color histogram. The dominant hue is extracted and matched with standardized color labels to obtain the child's clothing color attribute, `child_color`. Color matching uses the HSV color space, with the H hue component ranging from 0 to 180. Preset hue intervals are mapped to standardized color labels; for example, H values ​​in the 0-10 or 156-180 interval are mapped to red.

[0036] The system will assign each child's clothing color attribute With target features Perform a comparison. The comparison rules are as follows: When and If the colors match perfectly, it's considered a color match; when a photo of the clothing worn that day is uploaded, the system performs additional image feature comparison. A ResNet50 convolutional neural network is used to extract image feature vectors from the children's clothing in the photos. Simultaneously, image feature vectors of the target clothing area of ​​children in the real-time video stream are extracted. Calculate the cosine similarity between two feature vectors: ; When the similarity is greater than the preset threshold of 0.85, a match is considered to be made even if the color labels do not match perfectly. For children's targets that are determined to be matches, the system adds them to the SuspectedSet.

[0037] For each suspected target in the SuspectedSet, the system obtains the following detailed information: Real-time location calculation: based on the camera device identifier that captured the target. The system uses the preset geographic coordinate range of the monitoring area to convert the pixel coordinates of the target center point into actual geographic location coordinates. , The conversion formula is: ; in, The Map function establishes a mapping relationship between pixel coordinates and geographic coordinates based on camera calibration parameters.

[0038] On-site screenshot generation: Capture the current frame image, use the OpenCV library to mark the suspected target in the image with a green bounding box, and add a label above the bounding box to display the matching degree or color information. Save the screenshot as a JPEG, maintaining the original frame size at the resolution, and setting the encoding quality factor to 90 to ensure clarity. Timestamp recording: Record the current system time, accurate to the second, in the format YYYY-MM-DD HH:MM:SS.

[0039] The system encapsulates all acquired suspected target information into a SuspectedList, with each record containing the following fields: Temporary identifier for suspected targets; : Storage path or Base64 encoded data of the on-site screenshot; position: Real-time geographic location coordinates ; Screenshot timestamp; Match score, calculated based on color matching or image similarity; The system identifies the camera device used to capture the target. The system pushes the list of suspected targets to the administrator's handheld terminal via a WebSocket long-lived connection. Upon receiving the list, the administrator's handheld terminal displays a screenshot of the live feed and location information for each suspected target in a card-based list format, sorted from highest to lowest match. Each card has "Confirm" and "Ignore" buttons for manual confirmation by the administrator. The system continuously analyzes the video stream within the search area, updating the list of suspected targets every 5 seconds. When a new suspected target is discovered, it is incrementally pushed to the administrator's handheld terminal; when a target leaves the search area or is not detected for 10 consecutive seconds, it is removed from the list. This update mechanism ensures that the administrator always sees the latest search results.

[0040] S5. In response to the administrator's confirmation command for a suspected target on the handheld terminal, obtain the real-time location of the target.

[0041] Specifically, when an administrator views the list of suspected targets on a handheld terminal, each suspected target is displayed as a card showing a screenshot of the scene, a location overview, and a match score. When the administrator selects a target by clicking the confirmation button on a card, the handheld terminal generates a confirmation command. This command carries the target identifier corresponding to the suspected target, denoted as [target identifier]. . A unique temporary identifier, represented by a 32-bit UUID string, is assigned to each suspected target by the system in step S4. After receiving a confirmation click command, the system parses the target identifier from the command. Based on this target identifier, the system performs a query operation in the in-memory database of the background AI analysis system. The in-memory database contains... The primary key stores real-time video stream analysis data associated with the target, including: the identifier of the camera device that captured the target. The bounding box coordinate sequence of the target in the most recent consecutive frames, and the timestamps corresponding to each frame. The data includes the image storage path or memory buffer index for each frame. To ensure query efficiency, real-time video stream analysis data is only temporarily stored in memory for 30 seconds while the target is on the monitoring screen.

[0042] The system extracts the latest timestamped frame from the retrieved real-time video stream analysis data as the current frame. From this frame, it obtains the bounding box coordinates of the target within that frame. And calculate the image coordinates of the target center point ( , The calculation formula is: ; ; in, and The pixel coordinates of the top-left corner of the bounding box. and These are the pixel coordinates of the bottom right corner of the bounding box.

[0043] The system obtains the device identifier of the surveillance camera that captured the current frame. It then reads the camera's preset spatial location parameters from the camera parameter configuration table. These spatial location parameters include: the geographic coordinates of the camera's installation location (…). , ), Installation height Lens focal length f, horizontal field of view Vertical field of view The system measures the elevation angle P, horizontal angle Y, and rotation angle R. These parameters are obtained through on-site measurement and calibration during camera installation and are pre-stored in the system database. Based on these parameters, the system calculates the image coordinates of the target center point (P, Y, R). , Convert ) to actual geographical coordinates in the park map ( , The conversion process uses a pinhole camera model combined with a coordinate mapping algorithm. The specific implementation is as follows: First, the image coordinates are converted to normalized planar coordinates (…). , ): ; ; in, and The coordinates of the principal point of the image are taken as half of the image width and height. and Normalized focal length; , and This represents the physical size of a single pixel. Then, considering the camera's pose parameters, the normalized planar coordinates are transformed into a three-dimensional spatial orientation vector centered on the camera. The conversion process involves rotation matrix operations on the pitch angle P and the horizontal angle Y, which are mathematically expressed as follows: ; in, and These are the rotation matrices about the vertical and horizontal axes, respectively. Finally, based on the camera's mounting position coordinates ( , ) and installation height The target's geographical coordinates are obtained by combining the intersection of the direction vector and the ground. Assuming the park ground is horizontal, the formula for calculating the target's location is: ; ; in, The average radius of the Earth is taken as 6,371,000 meters. This is the conversion factor for radians to degrees. The system will calculate ( , This serves as the real-time location of the target, along with the target identifier. The current frame screenshot and timestamp are encapsulated together for use in step S6. Simultaneously, the system writes this real-time location to the missing person task data table, updates the task status to "target located," and records the location time. At this point, the system has completed the entire process from administrator confirmation to obtaining the target's real-time location, providing accurate destination coordinates for subsequent navigation guidance.

[0044] S6. Based on real-time location, plan and display dynamic navigation routes for tourists' mini-programs, and send assistance instructions to the handheld terminals of administrators near the real-time location.

[0045] Specifically, in step S5, the system obtains the real-time position of the target and records it as ( , Simultaneously, the system reads the visitor account that triggered the missing person assistance task from the missing person assistance task data table and records it as follows. Based on guest account The system calls the location interface of the mini-program service to obtain the real-time location coordinates of the tourist's current location. The tourist's mini-program continuously uploads location data in the background, with a location frequency of once every 5 seconds and a location accuracy of GPS / BeiDou dual-mode positioning. (The coordinates are marked as follows...) , If real-time location cannot be obtained due to network signal issues, the system will use the coordinates of the location where the distress call was triggered, as recorded in step S3. , The system will use the starting point coordinates () as the tourist's current location. , ) and endpoint coordinates ( , Input the path planning engine into the backend. The path planning engine calculates the optimal path based on the park's internal road network data model. The park's road network model is stored in a graph structure, where nodes represent road intersections or key landmarks, edges represent road segments, and each edge's associated attributes include the segment length. Road type, such as main road, secondary road, stepped road, and average travel time. Accessibility signage. The path planning engine uses the A algorithm to search for the optimal path, and the heuristic function uses the great circle distance between two points, calculated as follows: ; in, The Earth's average radius is 6,371,000 meters; , ) represents the coordinates of the current node; , Let f(n) represent the coordinates of the target node. Algorithm A starts from the starting node and gradually expands to the ending node, selecting the path that minimizes f(n) = g(n) + h(n), where g(n) is the actual path cost from the starting node to the current node, measured by path length or estimated travel time. The engine outputs the optimal travel route, represented as a sequence of path point coordinates P = ... and include the total distance and estimated walking time The system encapsulates the calculated optimal walking route into dynamic navigation instructions. These instructions use JSON data format and include the following fields: route identifier. Path point coordinate sequence P, total distance Expected time The system includes voice prompts for key turning points along the route. Dynamic navigation instructions are pushed to the visitor's mini-program via a WebSocket long connection. Upon receiving the instructions, the visitor's mini-program renders and displays them on the map interface. The map uses either Gaode Maps or Tencent Maps SDK, drawing the path point coordinate sequence P as a highlighted polyline with a blue color and a width of 6 pixels. Simultaneously, arrow animations indicate the direction of travel on the polyline, displaying a directional arrow every 50 meters. The mini-program enables real-time location tracking, dynamically adjusting the map view based on the visitor's current location and displaying real-time updates of "remaining distance to destination" and "estimated arrival time" on the map. When the visitor deviates from the planned route by more than 15 meters, the mini-program automatically triggers a replanning request, and the system repeats the above route planning process and pushes a new route.

[0046] The system synchronously generates and pushes administrator assistance commands. Based on the target's real-time location ( , The system performs a spatial query in the administrator's handheld terminal management table. This table stores information about currently online and active terminals; each terminal has a device identifier. Current real-time location ( , ), duty status indicator (Status) and last heartbeat time The system sets the query radius. The default value is 100 meters. Spatial query conditions are: terminal status (Status: on duty) and last heartbeat time. Within 30 seconds and the terminal's current location ( , ) and the target's real-time location ( , The distance d between the two is less than or equal to the query radius. The formula for calculating distance d is the Haversine formula: ; in, Given that the Earth's average radius is 6,371,000 meters, angle values ​​need to be converted to radians for calculation.

[0047] The system generates assistance instructions based on the query results. Assistance instructions include the following fields: Instruction Identifier. Real-time location of the target ( , ), screenshots of the scene, and text describing the assistance task, such as "Please go to this location to take care of the lost child and guide the parents to meet up," and the current location of the parents / tourists. , The system sends assistance instructions to all administrator handheld terminals via a WebSocket long connection, along with the parents' contact number. Upon receiving the instruction, the administrator's handheld terminal alerts the administrator via pop-up and vibration. The terminal's map interface automatically switches to the target location view, marking the target's real-time location with a red marker. , The parent's current location is marked with a blue dot. , The system also plans for the administrator to start from the administrator's current location. , ) to the target's real-time location ( , The recommended route is the same as the path planning algorithm mentioned above (A). The algorithm is consistent. The recommended route is drawn on the map as a green broken line, displaying distance and estimated time. Administrators can click the "Navigate" button to invoke the terminal's built-in map application for real-time voice navigation. The system continuously tracks the administrator's terminal location changes; when the administrator reaches within 10 meters of the target location, the terminal automatically displays a prompt: "You have arrived near the target; please be aware of children nearby." At this point, the system completes the two-way collaborative process of navigation guidance from the visitor's end and the sending of assistance commands from the administrator's end.

[0048] S7. In response to the end-of-search instruction triggered by tourists through the mini-program or smart guide screen, or after the preset park visit period ends, automatically delete the temporary feature information related to this park visit.

[0049] Specifically, the system provides two mechanisms for triggering the deletion of temporary feature information: deletion initiated by the visitor and automatic deletion by the system at set intervals. For scenarios where deletion is initiated by the visitor, after the parent and lost child are successfully reunited, the parent can click the "End Search" button on the missing person task page of the park's official mini-program, or scan the QR code on the smart guide screen and then click the "Task End" touch area. The system responds to the visitor's command to end the search and extracts the visitor's account information from the command. and instruction trigger time Visitor account information This serves as a unique identifier for visitors upon entry to the park, using a mobile phone number or WeChat OpenID as the account primary key. Command trigger time. This is the server time at which the system received the termination command, accurate to the second, in the format YYYYMMDDHHMMSS. The system is based on guest account information. The query operation is performed in the temporary feature information table of the background database. The temporary feature information table structure includes the following field: Account Identifier. Temporary signature Temporary feature information data block Photo storage path Creation time And the binding status. After the system finds the record bound to this account, it retrieves the photo storage path. The photo storage path points to the specific location on the file server where the photo of the child's outfit on that day is stored. The system sends a data deletion request to the database management system. The deletion request performs the following operations in a transactional manner: First, it deletes the photo storage path from the file server. For the corresponding photo file, call the file system's `unlink` command or the object storage's delete interface to ensure physical deletion, not just marking it as deleted; secondly, delete the record from the temporary feature information table by executing the SQL statement "DELETE FROM temp_feature_table WHERE user_id = ' Finally, the temporary feature information is unbound from the tourist account, and the cached data in the system memory is simultaneously invalidated, ensuring that no temporary feature information can be retrieved through this account in the future. After the deletion operation is successfully executed, the system returns a confirmation message to the tourist's mini-program or smart guide screen: "The search task has ended, and your information has been safely deleted."

[0050] For scenarios where the system automatically deletes entries at set intervals, the park management presets the end time of the visit cycle in the system configuration, denoted as... The closing time is usually set to the day's closing time, such as 22:00 in summer and 20:00 in winter, or uniformly set to 24:00 every day. The system starts a scheduled task daemon process, which runs every natural day. The system automatically performs batch deletion operations at all times. The scheduled task execution process is as follows: The system scans the temporary feature information table for all creation times. Earlier The record. Creation time. The entry time recorded in step S1 is in the format YYYYMMDDHHMMSS. The scanning conditions are... < The system iterates through the scan result set, and for each record that meets the criteria, performs the following operations: retrieves the photo storage path. The system deletes the corresponding photo file from the file server and removes the record from the temporary feature information table. To improve deletion efficiency, the system employs a batch deletion strategy, committing 100 records as a transaction batch to avoid the impact of long transactions on database performance. After the deletion operation is complete, the system executes a VACUUM or similar storage space reclamation command to release the physical storage space occupied by the deleted data. Simultaneously, the system records a task log for this scheduled deletion, including the total number of records deleted, execution time, and any exceptions, for administrator review the following day.

[0051] Before the scheduled deletion task is executed, the system sends a friendly reminder to all visitors currently in the park's mini-program: "Dear visitor, your visit to the park today is coming to an end, and the system will..." "Your child safety information for this visit will be automatically cleared. Thank you for using our service." This message is solely for user experience optimization and does not affect the automatic execution of the scheduled deletion task.

[0052] Through the combined use of these two deletion mechanisms, the system ensures that all temporary characteristic information is completely deleted after the visit ends or is actively terminated, preventing visitor privacy data from being stored on the server for extended periods, thus complying with the principle of data security minimization and relevant regulatory requirements. At this point, the system completes a full closed-loop process from pre-registration upon entry, visitor monitoring, lost person assistance, intelligent search, target confirmation, navigation guidance to data deletion.

[0053] Example 2: Existing smart wayfinding systems for urban parks suffer from low efficiency in finding lost children during peak holiday periods. Current methods primarily rely on manual announcements, broadcasting the missing child's characteristics through the park's public address system. However, this method has significant technical drawbacks: lost children are typically young and have limited recognition abilities, making it difficult for them to understand the announcements or find their parents in crowded and noisy outdoor environments, resulting in poor search effectiveness. Furthermore, parents, in their panic, often verbally describe their child's features to staff, leading to unclear descriptions and omissions of crucial information, further complicating the search. In addition, existing smart wayfinding systems mainly focus on basic services such as attraction navigation, information dissemination, and visitor statistics, lacking proactive protection and precise search functions for emergency scenarios such as lost children, failing to meet the urgent need for child safety protection during peak holiday periods. To address these issues, this invention provides a smart wayfinding system for urban parks, the structure of which is as follows... Figure 2 As shown. The specific implementation process of this system is as follows: The park entry pre-registration module is used to respond to the guardian activation command triggered by tourists through the park's official mini-program or smart guide screen, collect and bind the child's temporary characteristic information, and generate a temporary characteristic code associated with the tourist's account. The park visitor monitoring module is used to acquire real-time video streams of visitors based on surveillance cameras deployed in the park. The back-end AI analysis system detects whether unaccompanied children have entered the preset electronic fence area. If so, it sends an early warning message containing the scene to the administrator's handheld terminal for manual review and intervention. The lost and found assistance module is used to respond to lost and found assistance commands triggered by tourists through mini-programs or smart guide screens, retrieve the temporary feature code bound to the tourist's account, and delineate the search area starting from the location where the assistance was triggered. The intelligent search module is used to retrieve real-time video streams from surveillance cameras within the search area, perform structured human body searches based on temporary feature codes, generate a list of suspected targets, and push it to the administrator's handheld terminal. The target confirmation module is used to respond to the administrator's confirmation command for a suspected target on a handheld terminal and obtain the real-time location of the target; The navigation guidance module is used to plan and display dynamic navigation routes for tourists' mini-programs based on real-time location, while sending assistance instructions to the handheld terminals of administrators near that real-time location. The data clearing module is used to respond to the search end command triggered by tourists through the mini-program or smart guide screen, or to automatically delete temporary feature information related to the current visit after the preset park visit period ends.

[0054] Specifically, the park pre-registration module responds to the visitor's activation command by collecting the child's clothing color information and a photo of their attire for the day as temporary feature information, generating a temporary feature code associated with the visitor's account; the park monitoring module acquires real-time video streams from surveillance cameras and uses an AI analysis system to send an alert containing the scene to the administrator's handheld terminal when an unaccompanied child enters the electronic fence area; the lost and found assistance module responds to the visitor's lost and found assistance command by retrieving the temporary feature code bound to the account and defining a search area starting from the location of the request for help; the intelligent search module retrieves real-time video streams from cameras within the search area, performs a structured human body search based on the clothing color information in the temporary feature code, generates a list of suspected targets, and pushes it to the administrator's handheld terminal; the target confirmation module responds to the administrator's confirmation command for suspected targets by converting the target image coordinates into actual geographical coordinates on the park map; the navigation guidance module plans a dynamic navigation route for the visitor's mini-program based on the real-time location, and simultaneously sends an assistance command containing the target location and scene to the administrator's handheld terminal near the location; the data clearing module responds to the visitor's end-of-search command or the end of the preset park visit period by automatically deleting all temporary feature information related to this visit.

[0055] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art can still modify the technical solutions described in the foregoing embodiments or make equivalent substitutions for some of the technical features. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. A smart wayfinding method for urban parks, characterized in that, Includes the following steps: S1. In response to the guardian activation command triggered by tourists through the park's official mini-program or smart guide screen, collect and bind the child's temporary characteristic information, and generate a temporary characteristic code associated with the tourist's account. S2. Based on the surveillance cameras deployed in the park, the system acquires real-time video streams of visitors and uses the background AI analysis system to detect whether unaccompanied children have entered the preset electronic fence area. If so, the system sends an early warning message containing the scene to the administrator's handheld terminal for manual review and intervention. S3. In response to a lost person request triggered by a tourist through a mini-program or smart guide screen, retrieve the temporary feature code bound to the tourist's account and delineate a search area starting from the location where the request was triggered. S4. Retrieve the real-time video stream from the surveillance cameras within the search area, perform a structured human body search based on the temporary feature code, generate a list of suspected targets, and push it to the administrator's handheld terminal. S5. In response to the administrator's confirmation command for the suspected target on the handheld terminal, obtain the real-time location of the target; S6. Based on the real-time location, plan and display a dynamic navigation route for the tourist's mini-program, and at the same time send an assistance instruction to the administrator's handheld terminal near the real-time location. S7. In response to a visitor's command to end the search for a person triggered by a mini-program or smart guide screen, or after the preset park visit period ends, automatically delete the temporary feature information related to this park visit.

2. The smart wayfinding method for urban parks according to claim 1, characterized in that, In step S1, the steps of collecting and binding the child's temporary characteristic information and generating a temporary feature code associated with the guest account specifically include: In response to the guardian activation command triggered by tourists through the park's official mini-program or smart guide screen, a feature collection menu is displayed on the interactive interface, which includes at least one clothing color option. In response to the visitor's instruction to select the clothing color option, obtain the child's clothing color information; In response to a visitor's photo upload command triggered through the interactive interface, a photo of the child's outfit for the day is obtained; The clothing color information and the photo of the outfit worn on the day are used as temporary feature information, which are then bound to the current visitor's account to generate a temporary feature code, and the current time is recorded as the entry time.

3. The smart wayfinding method for urban parks according to claim 1, characterized in that, In step S2, the step of detecting whether an unaccompanied child has entered the preset electronic fence area through the background AI analysis system specifically includes: Based on the surveillance cameras deployed in the park, real-time video streams of park visitors are obtained from the preset electronic fence area and its surroundings; The background AI analysis system analyzes the park visit video stream in real time to identify child and adult targets in the video footage. Based on the relative positions of the child target and the adult target in the image, determine whether there is an associated adult target within a preset range around the child target; When there is no associated adult target within a preset range around the child target, the child is determined to be unaccompanied by an adult, and an early warning message containing the child's location and the scene is generated.

4. The smart wayfinding method for urban parks according to claim 1, characterized in that, In step S3, the step of defining the search area starting from the location where the distress call is triggered specifically includes: In response to lost and need-help requests triggered by visitors through the park's official mini-program or smart guide screen, the system obtains the current visitor's account information and the location coordinates of the location that triggered the request. Based on the visitor account information, a temporary feature code bound to the account is retrieved from the backend database; the temporary feature code contains the color information of the children's clothing collected during the pre-registration for park entry. Using the location coordinates of the triggered distress call as the center, a circular search area is defined according to a preset radius distance, or the park area where the location coordinates are located and adjacent areas are defined as the search area.

5. A smart wayfinding method for urban parks according to claim 4, characterized in that, In step S4, the step of generating a list of suspected targets and pushing it to the administrator's handheld terminal specifically includes: Based on the defined search area, obtain the device identifiers of all surveillance cameras within that area and retrieve their real-time video streams; The color information of the child's clothing is extracted from the temporary feature code and used as the target feature for human body structured search; The background AI analysis system performs human detection and attribute analysis on the real-time video stream to identify all child targets in the video and their corresponding clothing color attributes. The clothing color attribute of the child target is compared with the target features, and child targets with matching clothing color attributes are selected as suspected targets. Obtain the real-time location and on-site screenshot of each suspected target, and generate a list of suspected targets containing target images, location information and timestamps; The list of suspected targets is pushed to the administrator's handheld terminal for manual confirmation.

6. The smart wayfinding method for urban parks according to claim 5, characterized in that, In step S5, the step of obtaining the real-time location of the target specifically includes: In response to the administrator's confirmation click command on a suspected target image in the list of suspected targets on the handheld terminal, the target identifier corresponding to the suspected target is obtained; Based on the target identifier, real-time video stream analysis data associated with the target is retrieved from the background AI analysis system; Extract the current frame of the target from the real-time video stream analysis data, and parse the image coordinates of the target in the frame; Based on the device identifier of the surveillance camera that captured the current frame and its preset spatial location parameters, the image coordinates are converted into actual geographical location coordinates on the park map, which serve as the real-time location of the target.

7. A smart wayfinding method for urban parks according to claim 6, characterized in that, In step S6, the steps of planning and displaying dynamic navigation routes for tourists' mini-program terminals, and simultaneously sending assistance instructions to the handheld terminals of administrators near the real-time location, specifically include: Based on the real-time location of the target and the current location of the tourist account that triggered the missing person assistance request, the optimal walking route from the current location to the real-time location is calculated by the background path planning engine. The optimal walking route is pushed to the tourist's mini-program in the form of dynamic navigation instructions, and the navigation path and direction guidance are rendered and displayed in real time on the map interface of the mini-program. Based on the real-time location, query the handheld terminals of administrators who are on duty within a preset range around the location, and generate an assistance instruction that includes the real-time location of the target, a screenshot of the scene, and a description of the assistance task. The assistance command is pushed to the queried administrator's handheld terminal, and the real-time location and recommended route are marked on the terminal's map interface.

8. The smart wayfinding method for urban parks according to claim 1, characterized in that, In step S7, the step of automatically deleting the temporary feature information related to this visit specifically includes: In response to a visitor's command to end the search for a person triggered by the park's official mini-program or smart guide screen, the system obtains the visitor's account information and the time the command was triggered. Based on the visitor account information, locate the temporary feature information bound to the account and its storage path from the backend database; Send a data deletion request to the database management system to clear the temporary feature information under the storage path and unbind the information from the visitor account; or when the preset park visit period ends, the system automatically scans all temporary feature information in the database whose creation time is earlier than the end time, performs batch deletion operations, and releases storage space.

9. A smart wayfinding system for urban parks, characterized in that, A smart wayfinding method for urban parks according to any one of claims 1-8, the system comprising: The park entry pre-registration module is used to respond to the guardian activation command triggered by tourists through the park's official mini-program or smart guide screen, collect and bind the child's temporary characteristic information, and generate a temporary characteristic code associated with the tourist's account. The park visitor monitoring module is used to acquire real-time video streams of visitors based on surveillance cameras deployed in the park. The back-end AI analysis system detects whether unaccompanied children have entered the preset electronic fence area. If so, it sends an early warning message containing the scene to the administrator's handheld terminal for manual review and intervention. The lost and found assistance module is used to respond to lost and found assistance commands triggered by tourists through mini-programs or smart guide screens, retrieve the temporary feature code bound to the tourist's account, and delineate the search area starting from the location where the assistance was triggered. The intelligent search module is used to retrieve real-time video streams from surveillance cameras within the search area, perform structured human body searches based on the temporary feature codes, generate a list of suspected targets, and push it to the administrator's handheld terminal. The target confirmation module is used to respond to the administrator's confirmation command for a suspected target on a handheld terminal and obtain the real-time location of the target; The navigation guidance module is used to plan and display dynamic navigation routes for tourists' mini-program terminals based on the real-time location, and at the same time send assistance instructions to the handheld terminals of administrators near the real-time location. The data clearing module is used to automatically delete the temporary feature information related to the current visit in response to a visitor's command to end the search via a mini-program or smart guide screen, or after the preset visit period has ended.

Images