Intelligent method and system for searching for lost children in urban rail transit scene

Abstract

This application relates to the field of urban traffic management technology, and in particular to a smart method and system for finding lost children in urban rail transit scenarios. This application employs automatic locking of the monitoring range based on urban rail spatiotemporal indexing, combined with multimodal feature extraction and cross-system matching techniques. Compared to existing technologies, it effectively solves the problems of traditional manual screening methods, such as strong time delays, difficulty in handling the uncertain movement paths of children within the urban rail network, and the disconnect between information acquisition and personnel scheduling caused by the independence of multiple urban rail systems. It achieves rapid and accurate location of lost children across stations and systems, significantly shortening search time and reducing the risk to children's safety.

Description

Technical Field

[0001] This application relates to the field of urban traffic management technology, and in particular to a smart method and system for finding lost children in urban rail transit scenarios. Background Technology

[0002] As a core component of urban public transportation, urban rail transit is characterized by high passenger density, numerous stations, and high mobility. During daily operations, factors such as overcrowding and guardian negligence lead to frequent cases of children going missing, highlighting the increasing risks to child safety and operational management. Finding missing children as quickly as possible and mitigating the safety risks they face while abducted are crucial requirements for ensuring the safe operation of urban rail transit and improving service quality.

[0003] Currently, responding to missing children incidents in urban rail transit primarily relies on manual operation. When guardians discover their child is missing, they typically need to seek help from on-site staff, who then contact and retrieve surveillance video for frame-by-frame review. Due to the long length of urban rail lines and the numerous stations, manual review requires a significant amount of manpower and is time-consuming, resulting in a substantial time lag between the incident and the child's location. Furthermore, during the review process, the child may move with the train to other stations or transfer to other lines, further increasing the difficulty of tracking them.

[0004] In addition, although the urban rail transit system has deployed a relatively complete security monitoring system and operation and maintenance management system, when a child goes missing, staff need to log into different systems to operate. During this process, the retrieval of monitoring images, the query of train operation information, and the notification of station staff need to be completed step by step, and multiple manual communications and information transmissions take up a lot of time.

[0005] Therefore, how to overcome the limitations of traditional manual operation mode and achieve rapid location of lost children in the complex scenarios of urban rail transit has become an urgent problem to be solved. Summary of the Invention

[0006] (a) Technical problems to be solved

[0007] In view of the above-mentioned shortcomings and deficiencies of the prior art, this application provides a smart search method and system for lost children in urban rail transit scenarios. It solves the technical problems in the related technologies, such as the strong time delay in search caused by the reliance on manual investigation mode, the difficulty in dealing with the uncertainty of the child's movement path in the urban rail network, and the lengthy handling process and low overall efficiency caused by the independent operation of multiple urban rail systems and the need for step-by-step operation of information acquisition and personnel scheduling.

[0008] (II) Technical Solution

[0009] To achieve the above objectives, the main technical solutions adopted in this application include:

[0010] In a first aspect, embodiments of this application provide a smart method for finding lost children in urban rail transit scenarios, including:

[0011] The urban rail transit cloud receives a missing child search request sent from a mobile device. The search request includes: the child's basic information, the child's multimedia information, the current location information of the person searching, and the time range for the entry.

[0012] The cloud generates index query information based on the search request to find monitoring equipment from the pre-built urban rail transit spatiotemporal index tree. Based on the index query information, it searches the urban rail transit spatiotemporal index tree for the identifiers of all monitoring equipment associated with the child in the search request and their corresponding monitoring time periods. The urban rail transit spatiotemporal index tree is an index tree constructed by the cloud based on the train timetables and road network topology reported by various operation and maintenance systems in urban rail transit, including route codes, station codes, platform codes, standardized time period nodes, and train numbers.

[0013] The cloud generates a missing child identification code for multimedia search based on the search request, and determines the corresponding control system and filtering strategy based on the monitoring device identifier and monitoring time period. Based on the filtering strategy, the child identification code, monitoring device identifier and monitoring time period are sent to the corresponding control system, so that the control system can match the monitoring video in the monitoring time period based on the child identification code and obtain the matching monitoring frame and matching result to be returned to the cloud.

[0014] The cloud platform provides feedback to the mobile device based on the monitoring frames and matching results.

[0015] Specifically, the cloud generates index query information based on the search request to locate monitoring equipment from a pre-built urban rail transit spatiotemporal index tree, including:

[0016] Based on road network topology data, the cloud constructs urban rail routes, station nodes, and platform nodes to form a static structure of urban rail spatiotemporal index tree. Based on train timetable data, at least one time period sub-node is constructed for each platform node, and each time period sub-node is labeled with the corresponding Unix time start stamp and Unix time end stamp.

[0017] Under each time period sub-node, the cloud constructs the train number sub-node corresponding to that time period based on the train timetable data, and pre-configures the corresponding monitoring equipment identification information for each train number sub-node to build a complete urban rail transit spatiotemporal index tree;

[0018] Based on the current location information, the cloud determines the corresponding route code L, station code S, and platform code P, and generates a standardized spatial index I. space Standardized Spatial Index I space Satisfy the following formula:

[0019] ;

[0020] Based on the time range entered in the search request, the cloud generates corresponding start and end Unix timestamps as a standardized time index range;

[0021] Use the range of standardized spatial index and standardized temporal index as index query information.

[0022] Specifically, based on the index query information, the system searches the urban rail transit spatiotemporal index tree for the identifiers of all monitoring devices associated with the child in the search request and their corresponding monitoring time periods, including:

[0023] Based on a standardized spatial index, the cloud locates the corresponding target station node in the urban rail transit spatiotemporal index tree;

[0024] Based on the standardized time index range, the cloud traverses the time period sub-nodes under the target station node and determines the target time period node as the sub-node whose Unix time interval intersects with the standardized time index range.

[0025] The cloud traverses the train sub-nodes under the target time period node, and determines the train node that covers the station node corresponding to the current location information based on the train operation interval obtained from the train operation timetable data.

[0026] The cloud extracts the pre-configured monitoring device identification information of the target station node or matching train node, and uses the Unix start time stamp and Unix end time stamp of the target time period node as the monitoring time period.

[0027] Optionally, the cloud generates a missing child identification code for multimedia retrieval based on the search request, including:

[0028] Based on children's multimedia information, the cloud acquires children's image samples, performs preprocessing and feature encoding on the children's image samples, and obtains children's image feature codes;

[0029] The cloud-based system performs structured processing and encoding on the text describing a child's appearance from the child's basic information to generate a child text feature code that is consistent with the dimension of the child's image feature code.

[0030] The cloud platform fuses and calculates the child's text feature code and the child's image feature code to obtain the child's feature code.

[0031] Optionally, before obtaining children's image samples based on children's multimedia information in the cloud, the following steps are also included:

[0032] The search request also includes entry card swipe information, which includes the entry site, gate number, and card swipe time information in Unix timestamp format;

[0033] The cloud generates a standardized spatial index based on the station and gate number, and matches the corresponding gate monitoring equipment identification information in the urban rail transit spatiotemporal index tree based on the standardized spatial index and card swiping time information to determine the corresponding control system and monitoring time period of the gate.

[0034] Based on the identification information of the gate monitoring equipment, the cloud retrieves the monitoring video of the corresponding monitoring period from the control system, filters the key frames containing clear images of people in the monitoring video, and extracts the clothing color distribution features and contour features of the people in the key frames to generate a comparison sample.

[0035] Based on the child's basic information, such as age, gender, and clothing characteristics, as well as the child's photos and videos in the child's multimedia information, the cloud extracts child feature templates for initial matching.

[0036] The cloud calculates the feature similarity between the child feature template and the sample to be compared, and the set of sample to be compared with similarity higher than the first preset threshold is the image sample library of initial successful matching;

[0037] The cloud-based image sample library is further filtered to extract multi-angle image frames containing at least two different angles of the child's front, side, and oblique sides, as well as at least two different postures of standing and walking. The extracted multi-angle image frames are then added to the child image sample library.

[0038] Specifically, the cloud performs preprocessing and feature encoding on children's image samples to obtain children's image feature codes, including:

[0039] The cloud-based system preprocesses children's image samples and uses a lightweight convolutional neural network to extract features from the preprocessed image samples.

[0040] The cloud-based system extracts shallow, mid, and deep features from a lightweight convolutional neural network based on multi-scale feature fusion and performs weighted integration. The shallow features include edge and color features, the mid-level features include local texture features, and the deep features include global semantic features.

[0041] The cloud platform quantizes and encodes the weighted and integrated features to generate image feature codes for children.

[0042] Specifically, the cloud-based system performs structured processing and encoding on the text describing a child's appearance from the child's basic information to generate a child text feature code that is consistent with the dimensions of the child's image feature code, including:

[0043] The cloud uses a combination of tag encoding and one-hot encoding to process the discrete text in the child's basic information, mapping the discrete text into a binary vector;

[0044] The cloud performs normalization processing on the continuous text in the child's basic information, mapping the continuous text into a numerical vector;

[0045] The cloud integrates binary vectors and numerical vectors to generate a unified-dimensional text feature code for children.

[0046] Optionally, the cloud determines the corresponding control system and filtering strategy based on the monitoring device identifier and the monitoring time period. Based on the filtering strategy, it sends the child's feature code, monitoring device identifier, and monitoring time period to the corresponding control system. This allows the control system to match the child's feature code against the monitoring video within the monitoring time period, obtaining the matched monitoring frames and matching results to be returned to the cloud, including:

[0047] The cloud sends the child's ID code, monitoring device identifier, monitoring time period, and filtering strategy to the monitoring information matching module.

[0048] The monitoring information matching module calls the corresponding control system to obtain the monitoring video within the monitoring time period based on the received monitoring device identifier;

[0049] The monitoring information matching module analyzes the monitoring video frame by frame and extracts monitoring frames containing clear images of people based on the moving target detection algorithm;

[0050] The monitoring information matching module preprocesses the extracted monitoring frames, extracts the image feature code of each person in the monitoring frame, calculates the similarity value between the child's feature code and the person's image feature code using the cosine similarity algorithm, selects monitoring frames with similarity values ​​higher than the second preset threshold as matching monitoring frames, and records the corresponding similarity values ​​and adds them to the matching results.

[0051] The monitoring information matching module returns the location information, monitoring time period, and similarity value corresponding to the monitoring device identifier to the cloud as the matching result and the matched monitoring frame.

[0052] Optionally, it also includes:

[0053] The cloud creates search cases based on the initial information package and determines the status information of the search cases as the case initiation status;

[0054] Based on the received trigger events, the cloud updates the status information of the search cases and synchronizes it to the mobile device;

[0055] When an interruption request is received, the cloud updates the status information to the case interrupted early, verifies the identity of the user who initiated the interruption request, and enters the termination state after successful verification.

[0056] When a matching result is received, the cloud will update the status information to "case confirmation in progress";

[0057] If the cloud receives a successful confirmation instruction during the case confirmation process, the status information will be updated to the "Completed" status; if the cloud receives a failed confirmation instruction, the status information will be updated to the "Case Initiated" status.

[0058] When the matching time exceeds the third preset threshold and no matching result is received, the cloud will update the status information to a case abnormality alert status;

[0059] When a case is in an abnormal alert state, if the cloud receives a recovery command, the status information will be updated to the case initiation status; if the cloud receives a termination command, the status information will be updated to the termination status.

[0060] Secondly, embodiments of this application provide a smart system for finding lost children in urban rail transit scenarios, comprising:

[0061] The mobile module is used to upload the initial information package of a missing child and receive updated status information;

[0062] A cloud-based backend processing module for urban rail transit, used to execute the method as described in any one of claims 1-9;

[0063] The monitoring information matching module is used to receive the child feature code distributed by the urban rail transit cloud back-end processing module, call the external control system to obtain the corresponding carriage monitoring video, perform image matching on the monitoring video, generate matching results, and return the matching results to the urban rail transit cloud back-end processing module.

[0064] The task state machine case process management module is used to create search cases based on the initial information package, update the status information of the search cases according to the received trigger events, and synchronize them to the mobile terminal.

[0065] (III) Beneficial Effects

[0066] The beneficial effects of this application are as follows: The intelligent search method and system for lost children in urban rail transit scenarios proposed in this application, by adopting the automatic locking of the monitoring range based on the spatiotemporal index of urban rail transit and combining multimodal feature extraction and cross-system matching techniques, can effectively solve the problems of strong time delay and difficulty in tracking children whose movement paths are uncertain in the urban rail transit network, as well as the problem of information acquisition and personnel scheduling being disconnected due to the independence of multiple urban rail transit systems. It achieves the technical effect of rapid and accurate location of lost children across stations and systems, significantly shortening the search time and reducing the safety risks to children. Attached Figure Description

[0067] Figure 1 This is a schematic diagram illustrating the steps of a smart search method for lost children in an urban rail transit scenario according to an embodiment of this application;

[0068] Figure 2 This is a schematic diagram illustrating the functions of a cloud according to an embodiment of this application;

[0069] Figure 3 This is a schematic diagram of a task state machine example process management according to an embodiment of this application;

[0070] Figure 4 This is a schematic diagram illustrating the logical function division of a smart search system for lost children in an urban rail transit scenario according to an embodiment of this application.

[0071] Figure 5 This is a schematic diagram illustrating another logical function division of the intelligent search system for lost children in urban rail transit scenarios according to an embodiment of this application. Detailed Implementation

[0072] To better understand the above technical solutions, exemplary embodiments of this application will be described in more detail below with reference to the accompanying drawings. Although exemplary embodiments of this application are shown in the drawings, it should be understood that this application can be implemented in various forms and should not be limited to the embodiments set forth herein. Rather, these embodiments are provided so that this application can be understood more clearly and thoroughly, and that the scope of this application can be fully conveyed to those skilled in the art.

[0073] Example 1

[0074] like Figure 1 As shown in the embodiment of this application, a smart method for finding lost children in urban rail transit scenarios includes:

[0075] S100 The urban rail transit cloud receives a missing child search request sent by a mobile terminal. The search request includes: the child's basic information, the child's multimedia information, the current location information of the searcher, and the time range for data entry.

[0076] S200 and the cloud generate index query information based on the search request to find monitoring equipment from the pre-built urban rail transit spatiotemporal index tree. Based on the index query information, the cloud searches the urban rail transit spatiotemporal index tree for the identifiers of all monitoring equipment associated with the child in the search request and their corresponding monitoring time periods. The urban rail transit spatiotemporal index tree is an index tree constructed by the cloud based on the train timetables and road network topology reported by various operation and maintenance systems in urban rail transit, including route codes, station codes, platform codes, standardized time period nodes, and train numbers.

[0077] The S300 and cloud-based systems generate a missing child identification code for multimedia search based on the search request. They also determine the corresponding control system and filtering strategy based on the monitoring device identifier and monitoring time period. Based on the filtering strategy, the system sends the child identification code, monitoring device identifier, and monitoring time period to the corresponding control system. This allows the control system to match the child identification code against the monitoring video within the monitoring time period and obtain the matching monitoring frames and matching results to be returned to the cloud.

[0078] The S400 and cloud platforms provide feedback to the mobile device based on monitoring frames and matching results.

[0079] The cloud-based backend processing module of urban rail transit serves as the core data processing carrier of the system. It receives the initial information packets uploaded by the mobile terminal, and completes multi-dimensional feature association and binding through information classification and sorting, standardized matching of platform and train information, multimedia preprocessing and image encoding. This provides standardized feature data support for downstream train integrated control systems and operation and maintenance integrated control systems.

[0080] This embodiment, by employing automatic locking of the monitoring range based on urban rail transit spatiotemporal index and combining multimodal feature extraction and cross-system matching techniques, can effectively solve the problems of strong time lag and difficulty in tracking children whose movement paths are uncertain within the urban rail transit network, as well as the problem of information acquisition and personnel scheduling being disconnected due to the independence of multiple urban rail transit systems, compared with existing technologies. It achieves rapid and accurate location of lost children across stations and systems, significantly shortens search time, and reduces the technical effect of children's safety risks.

[0081] Example 2

[0082] like Figure 2 As shown in this embodiment, a smart method for finding lost children in urban rail transit scenarios includes:

[0083] The urban rail transit cloud platform receives a missing child search request sent from a mobile device. The search request includes: the child's basic information, the child's multimedia information, the current location information of the person searching, and the time range for the entry.

[0084] The cloud generates index query information based on the search request to find monitoring equipment from the pre-built urban rail transit spatiotemporal index tree. Based on the index query information, it searches the urban rail transit spatiotemporal index tree for the identifiers of all monitoring equipment associated with the child in the search request and their corresponding monitoring time periods. The urban rail transit spatiotemporal index tree is an index tree constructed by the cloud based on the train timetables and road network topology reported by various operation and maintenance systems in urban rail transit, including route codes, station codes, platform codes, standardized time period nodes, and train numbers.

[0085] The cloud generates a missing child identification code for multimedia search based on the search request, and determines the corresponding control system and filtering strategy based on the monitoring device identifier and monitoring time period. Based on the filtering strategy, the child identification code, monitoring device identifier and monitoring time period are sent to the corresponding control system, so that the control system can match the monitoring video in the monitoring time period based on the child identification code, and obtain the matching monitoring frame and matching result to be returned to the cloud.

[0086] The cloud platform provides feedback to the mobile device based on the monitoring frames and matching results.

[0087] Specifically, the cloud generates index query information based on the search request to locate monitoring equipment from a pre-built urban rail transit spatiotemporal index tree, including:

[0088] Based on road network topology data, the cloud constructs urban rail routes, station nodes, and platform nodes to form a static structure of urban rail spatiotemporal index tree. Based on train timetable data, at least one time period sub-node is constructed for each platform node, and each time period sub-node is labeled with the corresponding Unix time start stamp and Unix time end stamp.

[0089] Under each time period sub-node, the cloud constructs the train number sub-node corresponding to that time period based on the train timetable data, and pre-configures the corresponding monitoring equipment identification information for each train number sub-node to build a complete urban rail transit spatiotemporal index tree.

[0090] Based on the current location information, the cloud determines the corresponding route code L, station code S, and platform code P, and generates a standardized spatial index I. space Standardized Spatial Index I space Satisfy the following formula: .

[0091] Based on the time range entered in the search request, the cloud generates corresponding start and end Unix timestamps as a standardized time index range.

[0092] Use the range of standardized spatial index and standardized temporal index as index query information.

[0093] Before matching platform and train information in the cloud, the platform labeling information and entry time uploaded by the mobile terminal are synchronized through the data access layer. At the same time, data information from the urban rail operation and maintenance system is accessed to obtain data such as the road network topology and train timetable. Furthermore, based on standardized time and location information, it is mapped into a standardized urban rail spatial index that is accurate to the platform and carriage section.

[0094] The standardized spatial index is generated by synchronizing mobile terminal location and urban rail operation and maintenance data in the cloud, mapping the location to a standardized index. For example, the index corresponding to platform 3 of station 15 on line 3 is "3015003". The standardized time index is generated by traversing all time period sub-nodes under the locked platform node in the cloud. Each time period node is marked with a Unix time start stamp and end stamp as standardized index information. The time range entered in the search request refers to the time range during which the child went missing, which makes it easier to locate time segments such as monitoring data.

[0095] This embodiment constructs a static structure of urban rail routes, stations, and platform nodes based on road network topology data. It then builds time-period and train-number sub-nodes for each platform node based on train timetables and pre-configures monitoring equipment identification information, thus establishing a hierarchical index tree covering spatiotemporal dimensions. Simultaneously, it generates a standardized spatial index based on current location information and a standardized temporal index range based on the entered time range. Using both as index query information provides a precise and efficient query basis for subsequent locating monitoring equipment, helping to narrow the search scope and improve retrieval speed.

[0096] Specifically, the cloud uses the index query information to search the urban rail transit spatiotemporal index tree for the identifiers of all monitoring devices associated with the child in the search request and their corresponding monitoring time periods, including:

[0097] Cloud-based standardized spatial index I space The corresponding target station node is located in the urban rail transit spatiotemporal index tree; the cloud traverses the time period sub-nodes under the target station node based on the standardized time index range, and determines the target time period node as the sub-node whose Unix time interval intersects with the standardized time index range.

[0098] The system iterates through the train sub-nodes under the target time period node in the cloud, and determines the train node that covers the station node corresponding to the current location information based on the train operation timetable data.

[0099] The cloud extracts the pre-configured monitoring device identification information of the target station node or matching train node, and uses the Unix start time stamp and Unix end time stamp of the target time period node as the monitoring time period.

[0100] The cloud-based system uses a pre-built multi-branch spatiotemporal index tree of "route-station-platform-time period-train number" (time period nodes are labeled with Unix timestamps) to match layer by layer starting from the root node. It matches route, station, and platform nodes sequentially, automatically pruning irrelevant nodes to quickly locate the target platform node and narrow down the subsequent matching range. Then, it directly compares the standardized time index to filter out target time period nodes containing the Unix timestamp, using these as the target range for train number matching. Finally, it traverses all train number child nodes under the target time period node and, combined with the train timetable of the urban rail transit system, verifies whether the train's operating section covers the current station. If it does, the precise binding of the platform and train number is completed, and the set of monitoring device identifiers pre-bound to the train number node is extracted.

[0101] This embodiment utilizes a standardized spatial index from the index query information to locate the target station node. Then, it traverses the time period sub-nodes using a standardized time index range, filtering out time period nodes that intersect with the time interval. Next, it traverses the train number sub-nodes and combines this with operational interval verification to determine the matching train number node. Finally, it extracts the pre-configured monitoring device identifier and corresponding monitoring time period for the target station node or matching train number node. This hierarchical matching method can quickly locate monitoring devices potentially related to a missing child, reduce invalid data interference, and improve the accuracy and efficiency of monitoring device location.

[0102] Optionally, the cloud generates a missing child identification code for multimedia retrieval based on the search request, including:

[0103] Based on children's multimedia information, the cloud acquires children's image samples, preprocesses and encodes the children's image samples to obtain children's image feature codes; based on the text describing the child's appearance in the child's basic information, the cloud performs structured processing and encoding to generate children's text feature codes with the same dimensions as the children's image feature codes.

[0104] The cloud platform fuses and calculates the child's text feature code and the child's image feature code to obtain the child's feature code.

[0105] The children's multimedia information includes photos and videos of children uploaded by the searchers. The photos are stored directly in the children's photo library. Keyframe images are extracted from the videos and added to the photo library to enrich the sample of children's images from different angles and poses. Finally, the cloud processes the data for feature point extraction and quantization encoding.

[0106] The child's basic information includes text tags submitted by the searcher regarding the child's appearance characteristics, such as age, clothing style and color. These unstructured text information are standardized to complete text structuring and encoding generation. The resulting child text feature code is unified with the child image feature code in 32 dimensions, ensuring that the child text feature code can be mathematically fused and calculated with the child image feature code.

[0107] This embodiment generates a child image feature code based on the child's multimedia information and a child text feature code based on the child's basic information. The two are then fused to calculate the final child feature code. This multimodal feature generation method integrates image information and text description, resulting in a more comprehensive and accurate child feature code that enhances the overall quality and accuracy of subsequent matching.

[0108] Optionally, before obtaining children's image samples based on children's multimedia information in the cloud, the following steps are also included:

[0109] The search request also includes entry card swipe information, which includes the entry site, gate number, and card swipe time information in Unix timestamp format.

[0110] The cloud platform generates a standardized spatial index based on the station and gate number, and matches the corresponding gate monitoring equipment identification information in the urban rail transit spatiotemporal index tree based on the standardized spatial index and card swiping time information to determine the corresponding control system and monitoring time period of the gate.

[0111] Based on the identification information of the gate monitoring equipment, the cloud retrieves the monitoring video of the corresponding monitoring period from the control system, filters the key frames containing clear images of people in the monitoring video, and extracts the clothing color distribution features and contour features of the people in the key frames to generate a comparison sample.

[0112] Based on the child's basic information, such as age, gender, and clothing characteristics, as well as the child's photos and videos in the multimedia information, the cloud extracts child feature templates for initial matching.

[0113] The cloud calculates the feature similarity between the child feature template and the sample to be compared, and the set of sample to be compared with the similarity higher than the first preset threshold is the image sample library of the initial successful match.

[0114] The cloud-based image sample library is further filtered to extract multi-angle image frames containing at least two different angles of the child's front, side, and oblique sides, as well as at least two different postures of standing and walking. The extracted multi-angle image frames are then added to the child image sample library.

[0115] To address the core pain points of insufficient multi-angle and clarity in urban rail surveillance images, and the limited availability of children's photos and videos provided by parents, which makes it difficult to meet the multi-angle sample requirements for feature encoding, this paper combines the characteristics of urban rail operation scenarios with information input by parents to obtain gate entry information. By retrieving entry surveillance footage and performing initial matching, multi-angle images of children near the gate are obtained, supplementing high-quality samples for the image preprocessing and feature encoding stages described above. The specific implementation is as follows:

[0116] By retrieving the entry data of the guardian swiping the gate at the data access layer, the core data includes the card swipe time (Unix timestamp), the gate number, and key information about the entry station. Combining the urban rail transit spatiotemporal index matching logic mentioned earlier, the entry station and gate number are mapped to a standardized spatial index, which is then matched with the corresponding gate and surrounding area monitoring device IDs to quickly pinpoint the monitoring coverage area when the child enters the station.

[0117] Monitoring retrieval and initial matching: Based on the locked monitoring device ID, monitoring videos for the corresponding time period (e.g., 1 minute before and after the card swipe time, adapted to the time taken for children to enter the station) are retrieved. A lightweight image initial matching algorithm is used, combined with a small amount of footage of children provided by parents, to complete the initial screening. The initial matching algorithm uses a combination of color histograms and HOG contour features, which does not require complex calculations and balances real-time performance with basic matching accuracy. The core logic is to extract the clothing color distribution and approximate body contour features from the footage provided by parents and the monitoring frames, calculate the feature similarity, and filter out monitoring frames with a similarity higher than the first preset threshold (70%).

[0118] Multi-angle sample extraction and supplementation: The effective monitoring frames selected in the initial matching are further filtered, prioritizing the extraction of clear frames from different angles such as the front, side, and oblique side of the child, as well as different postures such as standing and walking. The filtered multi-angle monitoring frames are supplemented into the child image sample library and integrated with the materials and video keyframes provided by the parents to form a comprehensive multi-angle image sample set. This provides supplementary coding data support for lightweight CNN + multi-scale feature fusion coding, effectively solving the problems of incomplete feature coding and insufficient matching accuracy caused by single samples.

[0119] This embodiment supplements the child image samples with frame samples from the gate monitoring, which can make full use of the monitoring resources for quickly locating children when they enter the station. By obtaining children's images from multiple angles and poses through gate monitoring, the sample library is effectively enriched, providing a better data foundation for subsequent feature encoding and enhancing the robustness of feature expression.

[0120] Specifically, the cloud performs preprocessing and feature encoding on children's image samples to obtain children's image feature codes, including:

[0121] The cloud-based system preprocesses children's image samples and uses a lightweight convolutional neural network to extract features from the preprocessed image samples.

[0122] The cloud-based system extracts shallow, mid, and deep features from a lightweight convolutional neural network based on multi-scale feature fusion and performs weighted integration. The shallow features include edge and color features, the mid-level features include local texture features, and the deep features include global semantic features.

[0123] The cloud platform quantizes and encodes the weighted and integrated features to generate image feature codes for children.

[0124] First, the cloud-based system enhances uploaded children's photos and videos by deblurring, brightening, and noise reduction to ensure clear image features. Then, the image feature encoding module of the feature processing unit extracts and quantizes the feature points from the preprocessed image samples. Feature encoding employs a lightweight CNN + multi-scale feature fusion approach, balancing encoding efficiency and feature robustness to meet the real-time processing needs of urban rail transit monitoring. The image feature encoding module of the feature processing unit uses MobileNetV3 as its base model, relying on depthwise separable convolutions for efficient feature extraction. An attention mechanism is used to strengthen the weights of angle-insensitive features such as clothing and personal belongings. Simultaneously, multi-scale feature fusion theory is combined to extract and weighted integrate shallow, medium, and deep CNN features. The shallow layer retains stable features such as edges and colors, the medium layer extracts local texture features, and the deep layer captures global semantic features, effectively compensating for feature distortion caused by blurriness and angle deviation in the monitoring images. This completes feature point extraction and quantization encoding, generating standardized image feature codes for children.

[0125] This embodiment preprocesses children's image samples by deblurring, brightening, and denoising. It then uses a lightweight convolutional neural network to extract features and integrates shallow edge color features, mid-level local texture features, and deep global semantic features through multi-scale feature fusion. Finally, it quantizes and encodes the children's image feature code, which can integrate information from different levels of the image and improve the feature's adaptability to problems such as blurriness and angle deviation in monitoring images. At the same time, the lightweight network design helps to balance computational efficiency and feature expression capabilities.

[0126] Specifically, the cloud-based system performs structured processing and encoding on the text describing a child's appearance from the child's basic information to generate a child text feature code that is consistent with the dimensions of the child's image feature code, including:

[0127] The cloud-based system processes discrete text in children's basic information using a combination of tag encoding and one-hot encoding, mapping the discrete text into binary vectors.

[0128] The cloud platform normalizes the continuous text in the child's basic information, mapping the continuous text into numerical vectors.

[0129] The cloud integrates binary vectors and numerical vectors to generate a unified-dimensional text feature code for children.

[0130] When generating textual feature codes for children, the cloud first analyzes and classifies the text content describing the child's appearance in the child's basic information. The child's basic information usually contains two types of textual features: one is discrete text (such as clothing color "red" and style "school uniform"), and the other is continuous text (such as age "8 years old").

[0131] For discrete text, the cloud employs a combination of tag encoding and one-hot encoding for digitization. Specifically, for each discrete feature (such as clothing color), the cloud first establishes an enumeration set of possible values ​​for that feature. Then, based on the child's actual information about the feature's value, it maps it to a corresponding index value through tag encoding. This index value is then converted into a binary vector through one-hot encoding. The dimension of this vector is the same as the number of elements in the enumeration set, with only the position corresponding to the actual value being 1 and the rest being 0. For example, if the clothing color enumeration set contains four colors: {red, blue, yellow, green}, and the child's actual clothing color is "red," then the index is 0 obtained through tag encoding, and a binary vector [1,0,0,0] is generated through one-hot encoding. If the child's clothing color is "blue," then a binary vector [0,1,0,0] is generated. After performing the above processing on multiple discrete features, the cloud can concatenate the binary vectors of each feature to form a complete discrete text feature vector.

[0132] For continuous text (such as age), the cloud uses normalization to map it into a unified numerical vector. Specifically, the cloud pre-defines reasonable value ranges for each continuous feature (e.g., age range of 0-16 years old), and then maps it to the [0,1] interval using min-max normalization based on the values ​​in the child's actual information. The normalized values ​​of each feature are then combined to form a numerical vector. Finally, the cloud fuses the generated discrete text binary vector with the continuous text numerical vector to generate a unified-dimensional child text feature code, ensuring that the child text feature code can be mathematically fused with the child image feature code.

[0133] This embodiment transforms unstructured text descriptions into structured vectors, enabling them to be fused with image features for computation. This enriches the feature dimensions and helps improve the comprehensiveness of multimodal matching.

[0134] Optionally, the cloud determines the corresponding control system and filtering strategy based on the monitoring device identifier and the monitoring time period. Based on the filtering strategy, it sends the child's feature code, monitoring device identifier, and monitoring time period to the corresponding control system. This allows the control system to match the child's feature code against the monitoring video within the monitoring time period, obtaining the matched monitoring frames and matching results to be returned to the cloud, including:

[0135] The cloud sends the child's ID, monitoring device identifier, monitoring time period, and filtering strategy to the monitoring information matching module.

[0136] The monitoring information matching module retrieves the monitoring video within the monitoring time period by calling the corresponding control system based on the received monitoring device identifier.

[0137] The monitoring information matching module analyzes the monitoring video frame by frame and extracts monitoring frames containing clear images of people based on the moving target detection algorithm.

[0138] The monitoring information matching module preprocesses the extracted monitoring frames, extracts the image feature code of each person in the monitoring frame, calculates the similarity value between the child's feature code and the person's image feature code using the cosine similarity algorithm, selects monitoring frames with similarity values ​​higher than the second preset threshold as matching monitoring frames, and records the corresponding similarity values ​​and adds them to the matching results.

[0139] The monitoring information matching module returns the location information, monitoring time period, and similarity value corresponding to the monitoring device identifier to the cloud as the matching result and the matched monitoring frame.

[0140] In one optional implementation, the monitoring information matching module corresponds to both the train integrated control system and the platform operation and maintenance integrated control system. The filtering strategy includes two aspects: First, the selection of the matching object, determining which control system the matching task should be distributed to based on the monitoring equipment identifier. Since monitoring equipment may be deployed in train carriages or platform areas, the cloud implicitly includes its type when generating the monitoring equipment identifier. The filtering strategy explicitly specifies that if the monitoring equipment identifier corresponds to a train carriage, the matching module needs to call the train integrated control system to obtain the monitoring video of the corresponding carriage; if it corresponds to a platform area, it calls the platform operation and maintenance integrated control system to obtain the monitoring video of the corresponding platform. This selection ensures the correct correspondence between the matching task and the video source. Second, the control of the matching process, including the starting time of matching, the matching time window, and the similarity threshold. For example, the filtering strategy can specify to start matching from the beginning of the monitoring time period, or prioritize matching the most recent video frames; a second preset similarity threshold (such as 0.85) can be set, and only when the cosine similarity between the child's feature code and the person's image feature code exceeds this threshold will the frame be considered as a candidate result; an upper limit on the number of returned results or a matching time limit can also be set, etc. These strategy parameters can be dynamically adjusted according to the actual scenario to balance the accuracy and efficiency of matching.

[0141] The monitoring information matching module uses the child's feature code, monitoring device identifier, monitoring time period, and filtering strategy sent from the cloud to lock the monitoring range and carry out intelligent image matching.

[0142] In the specific implementation process, firstly, the monitoring information matching module narrows down the monitoring scope. Based on the current location information (urban rail station, platform) and the input time range (multiple Unix timestamps) in the initial information packet, it calls the previously constructed "line-station-platform-time period-monitoring equipment" urban rail spatiotemporal index tree, and matches the monitoring equipment identification information of the corresponding urban rail station and platform through spatial index. At the same time, combined with the time dimension information, it locks down the monitoring equipment that is in working status within this time period, covering platform monitoring and train carriage monitoring, excluding irrelevant monitoring, and narrowing down the matching scope.

[0143] Next, the monitoring information matching module analyzes the locked monitoring video frame by frame, extracts key frames containing people based on the moving target detection algorithm, and removes frames with no people or unclear personnel features, thereby reducing the amount of comparison data and improving matching efficiency.

[0144] Image retrieval and comparison are performed by comparing the child image feature codes generated in the cloud with the personnel image features in the monitoring keyframes one by one. A similarity calculation algorithm is used to obtain the similarity value of each comparison result, and candidate results with similarity higher than a threshold are selected. First, the extracted monitoring keyframes are uniformly preprocessed (deblurring, brightening, and noise reduction). Using the lightweight CNN + multi-scale feature fusion method mentioned above, the image feature codes of each person in the keyframes are quickly extracted to ensure consistency with the multi-dimensional fused feature codes of the children generated in the cloud. Further, a cosine similarity calculation algorithm is used to calculate the similarity value between the child feature code and the feature code of each monitoring person. The core theory of cosine similarity is to measure the similarity between two feature vectors by the cosine value of the angle between the vectors. The value range is [0,1]. The closer the value is to 1, the higher the feature matching degree. The cosine similarity calculation formula is:

[0145] ,in This represents a multi-dimensional fused feature vector of a child generated in the cloud. The image feature vector representing the person in the keyframe of the surveillance video. The dot product of two vectors. , These are the magnitudes of the two vectors, respectively.

[0146] Based on the characteristics of urban rail monitoring scenarios, a reasonable similarity threshold is set to filter out candidate results with similarity higher than the threshold, while invalid results with low similarity are removed, balancing the accuracy and recall of matching, and avoiding missed or false matches.

[0147] Finally, the results are fed back to the cloud, and the candidate comparison results, key frames to be monitored, monitoring device location, and time information are uploaded to the cloud in real time.

[0148] This embodiment distributes the matching task to the monitoring information matching module for local processing, reducing the bandwidth pressure caused by video data transmission to the cloud. At the same time, the cosine similarity quantification combined with threshold screening helps to balance the accuracy and recall of the matching, ensuring the reliability of the results.

[0149] Optionally, it also includes:

[0150] The cloud creates a search case based on the initial information package and determines the status information of the search case as the case initiation status; the cloud updates the status information of the search case and synchronizes it to the mobile terminal according to the received trigger event.

[0151] When an interruption request is received, the cloud updates the status information to the case's early interruption status, verifies the identity of the user who initiated the interruption request, and enters the termination status after successful verification.

[0152] When a matching result is received, the cloud will update the status information to "case confirmation in progress".

[0153] If the cloud receives a successful confirmation instruction during the case confirmation process, it will update the status information to "Completed"; if the cloud receives a failed confirmation instruction, it will update the status information to "Case Initiated".

[0154] When the matching time exceeds the third preset threshold and no matching result is received, the cloud will update the status information to a case exception alert status.

[0155] When a case is in an abnormal alert state, if the cloud receives a recovery command, the status information will be updated to the case initiation status; if the cloud receives a termination command, the status information will be updated to the termination status.

[0156] like Figure 3 As shown, the above process manages the process through a task state machine case, involving the transition relationships and triggering conditions of each state.

[0157] Case Initiation Status: After the guardian completes the information entry and uploads it, the system automatically creates a case search and enters the case initiation status. At this time, the cloud and monitoring matching modules start working simultaneously, updating the processing progress in real time and feeding back to the mobile terminal and the urban rail operation backend.

[0158] Case interruption status: If the guardian finds the child during the search, they can initiate an interruption request via mobile device. After the system verifies the identity, it will terminate the matching process and case progress, record the reason and time of the interruption, and generate case archive data. The urban rail operation backend will receive the interruption notification simultaneously and stop the relevant resource scheduling.

[0159] Match successful, case confirmation in progress: When the guardian confirms the monitoring match result is correct, or when on-site staff find the child based on the match result, the system automatically enters this state, recording the time, location, and personnel involved in the successful match, and simultaneously synchronizing the relevant information to the urban rail operation backend. This information is then pushed to the APP and platform maintenance system via the cloud.

[0160] Case anomaly alert status: When the case process stalls (e.g., no matching results or matching time exceeds the third preset threshold), an alert will be automatically sent to the urban rail operation backend and the guardian's mobile terminal, prompting manual intervention.

[0161] Case process completion status: The case process completion status is the final closed loop status. It will automatically enter this status when the guardian / staff terminates the search process in advance, or when the system completes normal matching confirmation.

[0162] This embodiment covers the entire "initiation-running-interruption / termination" chain, taking into account both automatic state switching and manual intervention, making the changes in the search process status clear and controllable, allowing guardians to know the progress of the case in real time. At the same time, the preset abnormal handling process facilitates timely intervention when matching stalls, improving the standardization and flexibility of process management.

[0163] Example 3

[0164] like Figure 4 As shown, this embodiment provides a smart system for finding lost children in urban rail transit scenarios, including:

[0165] The mobile module is used to upload the initial information package of a lost child and receive updated status information.

[0166] The urban rail transit cloud-based backend processing module is used to execute the method as described in any one of claims 1-9.

[0167] The monitoring information matching module is used to receive the child feature code distributed by the urban rail transit cloud back-end processing module, call the external control system to obtain the corresponding carriage monitoring video, perform image matching on the monitoring video, generate matching results, and return the matching results to the urban rail transit cloud back-end processing module.

[0168] The task state machine case process management module is used to create search cases based on the initial information package, update the status information of the search cases according to the received trigger events, and synchronize them to the mobile terminal.

[0169] The mobile module is integrated into a WeChat mini-program or standalone app, providing guardians with a convenient entry point for inputting information about lost children. It allows guardians to manually input basic child information and upload multimedia information. Basic information includes the child's name, age, gender, clothing characteristics, and contact information, while multimedia information includes photos and videos of the child. Simultaneously, it acquires the guardian's current location (corresponding to the urban rail station and platform area) and input time information, as well as station entry card swipe information, forming an initial information package which is then uploaded to the urban rail transit cloud-based backend processing module.

[0170] For a detailed description of the above methods, please refer to the aforementioned method embodiments, which will not be repeated here. Furthermore, the explanations and descriptions of the beneficial effects of any of the smart child search systems provided above for urban rail transit scenarios can be found in the corresponding method embodiments, which will not be repeated here.

[0171] It should be noted that, Figure 4 The module division shown in the text is illustrative and represents only one logical functional division; in actual implementation, there may be other division methods. For example... Figure 5 As shown, in another optional logical function division, the function of the monitoring information matching module can also be distributed to multiple modules in the train integrated control system and the platform operation and maintenance integrated control system.

[0172] In the description of this application, it should be understood that the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0173] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0174] In this application, unless otherwise expressly specified and limited, "above" or "below" the second feature can mean that the first and second features are in direct contact, or that they are in indirect contact through an intermediate medium. Furthermore, "above," "on top of," and "over" the second feature can mean that the first feature is directly above or diagonally above the second feature, or simply that the first feature is at a higher horizontal level than the second feature. "Below," "below," and "under" the second feature can mean that the first feature is directly below or diagonally below the second feature, or simply that the first feature is at a lower horizontal level than the second feature.

[0175] In the description of this specification, the terms "one embodiment," "some embodiments," "embodiment," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of this application. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0176] Although embodiments of this application have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting this application. Those skilled in the art can make modifications, alterations, substitutions and variations to the above embodiments within the scope of this application.

Claims

1. A smart method for finding lost children in urban rail transit scenarios, characterized in that, include: The urban rail transit cloud receives a missing child search request sent from a mobile terminal. The search request includes: the child's basic information, the child's multimedia information, the current location information of the searcher, and the time range for the data entry. The cloud platform generates index query information based on the search request to locate monitoring devices from a pre-built urban rail transit spatiotemporal index tree. Based on this index query information, it searches the urban rail transit spatiotemporal index tree for the identifiers and corresponding monitoring time periods of all monitoring devices associated with the child in the search request. The urban rail transit spatiotemporal index tree is an index tree constructed by the cloud platform based on train timetables and network topology reported by various operation and maintenance systems in the urban rail transit system, including route codes, station codes, platform codes, standardized time period nodes, and train numbers. The cloud generates a missing child feature code for multimedia search based on the search request, and determines the corresponding control system and filtering strategy based on the monitoring device identifier and monitoring time period. Based on the filtering strategy, the cloud sends the child feature code, monitoring device identifier and monitoring time period to the corresponding control system, so that the control system matches the monitoring video in the monitoring time period based on the child feature code, and obtains the matching monitoring frame and matching result to be returned to the cloud. The cloud platform provides feedback to the mobile device based on the monitoring frames and matching results.

2. The method as described in claim 1, characterized in that, The cloud platform generates index query information based on the search request to locate monitoring equipment from a pre-built urban rail transit spatiotemporal index tree, including: The cloud platform constructs urban rail routes, station nodes, and platform nodes based on road network topology data, forming a static structure of urban rail spatiotemporal index tree. Based on train timetable data, it constructs at least one time period sub-node for each platform node and marks each time period sub-node with a corresponding Unix time start stamp and Unix time end stamp. Under each time period sub-node, the cloud constructs a train number sub-node corresponding to that time period based on the train timetable data, and pre-configures the corresponding monitoring equipment identification information for each train number sub-node to construct a complete urban rail transit spatiotemporal index tree; Based on the current location information, the cloud determines the corresponding route code L, station code S, and platform code P, and generates a standardized spatial index I. space Standardized Spatial Index I space Satisfy the following formula: ； Based on the input time range in the search request, the cloud generates a corresponding start Unix timestamp and end Unix timestamp as a standardized time index range; The standardized spatial index and the standardized time index range are used as the index query information.

3. The method as described in claim 2, characterized in that, The step of searching the urban rail transit spatiotemporal index tree for the identifiers and corresponding monitoring time periods of all monitoring devices associated with the child in the search request based on the index query information includes: Based on the standardized spatial index, the cloud platform locates the corresponding target station node in the urban rail transit spatiotemporal index tree. Based on the standardized time index range, the cloud traverses the time period sub-nodes under the target station node and determines the time period sub-nodes whose Unix time intervals intersect with the standardized time index range as the target time period nodes. The cloud traverses the train sub-nodes under the target time period node, and determines the train node whose operating section covers the station node corresponding to the current location information as the matching train node based on the train operation timetable data. The cloud extracts the pre-configured monitoring device identification information of the target station node or matching train node, and uses the Unix start time stamp and Unix end time stamp of the target time period node as the monitoring time period.

4. The method as described in claim 1, characterized in that, The cloud platform generates a missing child identification code for multimedia retrieval based on the search request, including: The cloud platform acquires children's image samples based on children's multimedia information, performs preprocessing and feature encoding on the children's image samples, and obtains children's image feature codes. The cloud platform performs structured processing and encoding on the text describing the child's appearance in the child's basic information to generate a child text feature code that is consistent with the dimension of the child's image feature code. The cloud platform fuses and calculates the child's text feature code and the child's image feature code to obtain the child's feature code.

5. The method as described in claim 4, characterized in that, Before acquiring children's image samples based on children's multimedia information, the cloud platform also includes: The search request also includes entry card swiping information, which includes the entry site, gate number, and card swiping time information in Unix timestamp format; The cloud platform generates a corresponding standardized spatial index based on the station and gate number, and matches the corresponding gate monitoring equipment identification information in the urban rail transit spatiotemporal index tree based on the standardized spatial index and card swiping time information to determine the corresponding control system and monitoring time period of the gate. The cloud-based system retrieves monitoring videos from the corresponding control system based on the gate monitoring equipment identification information, filters key frames containing clear images of people in the monitoring videos, and extracts the clothing color distribution features and contour features of the people in the key frames to generate a comparison sample. The cloud platform extracts child feature templates for initial matching based on the child's age, gender, and clothing characteristics in the child's basic information, as well as the child's photos and videos in the child's multimedia information. The cloud-based calculation of the feature similarity between the child feature template and the sample to be compared determines the set of sample to be compared with a similarity higher than a first preset threshold as an image sample library with initial successful matching. The cloud-based system further filters the image sample library, extracting multi-angle image frames containing at least two different angles from the front, side, and oblique sides of the child, as well as at least two different postures from standing and walking. The extracted multi-angle image frames are then added to the child image sample library.

6. The method as described in claim 4, characterized in that, The cloud platform preprocesses and encodes the child image samples to obtain child image feature codes, including: The cloud platform preprocesses the children's image samples and uses a lightweight convolutional neural network to extract features from the preprocessed image samples. The cloud-based system extracts shallow, mid, and deep features from a lightweight convolutional neural network based on multi-scale feature fusion and performs weighted integration. The shallow features include edge and color features, the mid-level features include local texture features, and the deep features include global semantic features. The cloud platform quantizes and encodes the weighted and integrated features to generate a child image feature code.

7. The method as described in claim 4, characterized in that, The cloud-based system performs structured processing and encoding on the text describing the child's appearance from the child's basic information to generate a child text feature code that is consistent with the dimension of the child's image feature code, including: The cloud-based system processes the discrete text in the child's basic information using a combination of tag encoding and one-hot encoding, mapping the discrete text into binary vectors. The cloud platform normalizes the continuous text in the child's basic information, mapping the continuous text into a numerical vector. The cloud-based system integrates binary vectors and numerical vectors to generate a unified-dimensional text feature code for children.

8. The method as described in claim 1, characterized in that, The cloud platform determines the corresponding control system and filtering strategy based on the monitoring device identifier and the monitoring time period. Based on the filtering strategy, it sends the child's feature code, monitoring device identifier, and monitoring time period to the corresponding control system. This allows the control system to match the child's feature code against the monitoring video within the monitoring time period, obtaining the matched monitoring frames and matching results to be returned to the cloud, including: The cloud platform sends the child's feature code, monitoring device identifier, monitoring time period, and filtering strategy to the monitoring information matching module. The monitoring information matching module calls the corresponding control system to obtain the monitoring video within the monitoring time period based on the received monitoring device identifier; The monitoring information matching module analyzes the monitoring video frame by frame and extracts monitoring frames containing clear images of people based on the moving target detection algorithm; The monitoring information matching module preprocesses the extracted monitoring frames, extracts the image feature code of each person in the monitoring frame, calculates the similarity value between the child's feature code and the person's image feature code using the cosine similarity algorithm, selects monitoring frames with similarity values ​​higher than the second preset threshold as matching monitoring frames, and records the corresponding similarity values ​​and adds them to the matching result. The monitoring information matching module returns the location information corresponding to the monitoring device identifier, the monitoring time period, and the corresponding similarity value as the matching result and the matched monitoring frame to the cloud.

9. The method as described in claim 1, characterized in that, Also includes: The cloud platform creates a search case based on the initial information package and determines the status information of the search case as the case initiation status. The cloud platform updates the status information of the searched cases and synchronizes it to the mobile device based on the received trigger events. When an interruption request is received, the cloud updates the status information to the case early interruption status, verifies the identity of the user who initiated the interruption request, and enters the termination status after successful verification. When a matching result is received, the cloud will update the status information to "case confirmation in progress"; In the case confirmation state, if the cloud receives a confirmation success instruction, it will update the status information to the end state; if the cloud receives a confirmation failure instruction, it will update the status information to the case initiation state. When the matching time exceeds the third preset threshold and no matching result is received, the cloud will update the status information to a case abnormality reminder status; When a case is in an abnormal alert state, if the cloud receives a recovery command, it will update the status information to the case initiation status; if the cloud receives a termination command, it will update the status information to the termination status.

10. A smart system for finding lost children in urban rail transit scenarios, characterized in that, include: The mobile module is used to upload the initial information package of a missing child and receive updated status information; The urban rail transit cloud-based backend processing module is used to execute the method as described in any one of claims 1-9; The monitoring information matching module is used to receive the child feature code distributed by the urban rail transit cloud back-end processing module, call the external control system to obtain the corresponding carriage monitoring video, perform image matching on the monitoring video, generate matching results, and return the matching results to the urban rail transit cloud back-end processing module. The task state machine case process management module is used to create a search case based on the initial information package, update the status information of the search case according to the received trigger event, and synchronize it to the mobile terminal.

Images