An end-cloud collaborative passenger flow hot area dynamic line splicing method based on ReID deduplication

Abstract

The application provides a terminal-cloud collaborative passenger flow hot area motion line splicing method based on ReID deduplication. The method comprises the following steps: receiving structured data after ReID deduplication of each camera terminal side by hour. The cloud side completes cross-camera pedestrian identity association and daily full data correction, and constructs a global pedestrian trajectory database. The scheme relies on a site plan to realize regional binding configuration without camera calibration, combines the deduplicated number of pedestrians and the stay time to quantify the regional heat and perform hierarchical rendering, simultaneously constructs a passenger flow topology network, splices the cross-regional passenger flow motion line, and generates a visual result. The application is simple to deploy, has less data redundancy, and is accurate in passenger flow statistics, and can stably realize fine analysis of the heat and flow trajectory of the site passenger flow.

Description

Technical Field

[0001] This application relates to the field of video analytics technology, and in particular to a method for edge-cloud collaborative passenger flow hotspot stitching based on ReID deduplication. Background Technology

[0002] Existing commercial venue customer flow analysis solutions mostly adopt independent processing modes, either purely on the edge or purely in the cloud. Traditional edge devices can only complete local pedestrian detection and simple counting based on a single camera image, making it difficult to achieve cross-camera pedestrian identity association. This easily leads to problems such as duplicate counting of the same pedestrian on multiple devices, broken trajectories across time periods, and redundant customer flow data. At the same time, traditional customer flow statistics only judge popularity based on the number of pedestrians, without considering behavioral characteristics such as pedestrian dwell time. This results in a single dimension of regional popularity evaluation, which cannot truly reflect the actual customer flow activity status and user preferences in different areas of the venue, leading to poor accuracy and objectivity in the analysis.

[0003] Furthermore, existing passenger flow analysis technologies have high deployment barriers and poor adaptability. They generally rely on precise camera coordinate calibration and complex matrix transformations to achieve area matching, resulting in high on-site debugging costs and long cycles, making it difficult to quickly adapt to commercial scenarios with different structures and layouts. Moreover, traditional solutions lack a layered verification mechanism for edge-cloud collaboration. Short-term window statistics on the edge are prone to pedestrian ID splitting and data distortion, while the cloud lacks the ability to perform secondary correction of all data. Ultimately, this leads to significant deviations in the analysis results of passenger flow hotspot distribution and passenger flow patterns, failing to meet the actual needs of refined and intelligent passenger flow operation analysis in commercial venues. Summary of the Invention

[0004] In view of this, this application provides a cloud-edge collaborative passenger flow hotspot stitching method based on ReID deduplication. By combining lightweight ReID deduplication on the edge, full data correction in the cloud, uncalibrated area matching, and multi-dimensional heat quantification with topology flow analysis, this method effectively solves the shortcomings of existing technologies such as redundant and distorted passenger flow statistics, broken trajectories, complex deployment, single evaluation dimensions, and large deviations in analysis results. It significantly reduces deployment costs and equipment computing power and bandwidth consumption, and significantly improves the accuracy and stability of passenger flow statistics, heat assessment, and flow analysis. It can be efficiently adapted to various commercial scenarios and meet the actual needs of refined and intelligent operation analysis of commercial passenger flow.

[0005] The technical solution of this application embodiment is implemented as follows: In a first aspect, embodiments of this application provide a method for edge-cloud collaborative passenger flow hotspot stitching based on ReID deduplication, comprising the following steps: The system receives pedestrian structured data uploaded by each camera end with a time window of natural hours and after local ReID deduplication. The structured data is obtained by the camera end performing pedestrian detection, tracking and ReID feature extraction on the real-time video stream and completing local deduplication. The local feature cache is cleared after the end uploads the data. Perform cross-camera ReID identity association on the received multi-camera structured data, build and maintain a global pedestrian trajectory database, and perform secondary comparison, ID merging and data correction on the full historical data of the day at each specific period; Based on the floor plan of the target location, multiple monitoring areas are divided, and a binding relationship is established between the cameras and the monitoring areas. Using each monitored area as a statistical unit, the number of unique pedestrians N_k and the average stay time T_k are counted. Combined with the preset time weighting coefficient λ, the regional heat H_k of each monitored area is calculated through the preset weighting formula. Based on the heat value, the regional hierarchical rendering display is completed on the plan. Based on a global pedestrian trajectory database, continuous pedestrian movement lines across regions are stitched together, the probability of passenger flow conversion between regions is statistically analyzed, and a spatial flow topology network is constructed to generate corresponding passenger flow line visualization results.

[0006] Secondly, embodiments of this application also provide a cloud-edge collaborative passenger flow hotspot splicing device based on ReID deduplication, the device comprising: The data receiving module is used to receive pedestrian structured data uploaded by each camera end with natural hours as the time window and after local ReID deduplication. The structured data is obtained by the camera end performing pedestrian detection, tracking and ReID feature extraction on the real-time video stream and completing local deduplication. After the end uploads the data, the local feature cache is cleared. The ReID verification module is used to perform cross-camera ReID identity association on the received multi-camera structured data, build and maintain a global pedestrian trajectory database, and perform secondary comparison, ID merging and data correction on the full historical data of the day at each specific period. The area configuration module is used to divide multiple monitoring areas based on the floor plan of the target location and establish a binding relationship between cameras and monitoring areas. The statistical calculation module is used to count the number of unique pedestrians N_k and the average stay time T_k per person, taking each monitored area as a statistical unit. Combined with the preset time weighting coefficient λ, the module calculates the regional heat H_k of each monitored area through a preset weighting formula. Based on the heat value, the module completes the regional hierarchical rendering display on the plan. The pedestrian flow visualization module is used to stitch together continuous pedestrian flow lines across regions based on a global pedestrian trajectory database, calculate the probability of passenger flow conversion between regions, construct a spatial flow topology network, and generate corresponding passenger flow visualization results.

[0007] Thirdly, embodiments of this application also provide an electronic device, including: a processor, a storage medium, and a bus. The storage medium stores machine-readable instructions executable by the processor. When the electronic device is running, the processor communicates with the storage medium via the bus, and the processor executes the machine-readable instructions to perform the ReID deduplication-based end-to-cloud collaborative passenger flow hotspot splicing method described in any of the first aspects.

[0008] Fourthly, embodiments of this application also provide a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, performs the end-to-cloud collaborative passenger flow hot zone splicing method based on ReID deduplication as described in any one of the first aspects.

[0009] The embodiments of this application have the following beneficial effects: By using a lightweight ReID deduplication preprocessing method with a natural hour window on the device before uploading data, redundant data captured repeatedly within a short period of time by a single device can be effectively filtered out, significantly reducing network transmission pressure and the real-time data processing load on the cloud. At the same time, the device-side cache is cleared periodically to avoid data accumulation and pollution, ensuring the standardization and efficiency of data flow. Relying on the cloud-based cross-camera ReID identity association and end-of-day full data secondary correction mechanism, it effectively compensates for pedestrian ID splitting, cross-device and cross-time period trajectory breaks, and passenger flow mismatch defects caused by the limitations of the single-device perspective, achieving complete stitching of global pedestrian trajectories and accurate correction of passenger flow data. At the same time, it abandons the traditional camera precision The complex deployment process of quasi-coordinate calibration is adapted to the scene by relying on the logical area binding method of the floor plan, which greatly simplifies the on-site deployment process, reduces the implementation cost and adaptation threshold. Moreover, by integrating the number of unique pedestrians and the average stay time per person to build a standardized regional heat quantification system, and combining the probability of passenger flow conversion to build a spatial flow topology network, it can accurately and comprehensively quantify the activity level and dynamic flow patterns of passenger flow in various areas of the venue. While ensuring the real-time and stability of passenger flow analysis, it realizes the integrated and accurate analysis of static heat distribution and dynamic flow of passenger flow, providing reliable technical support for the refined monitoring of passenger flow and basic operational data analysis of commercial venues. Attached Figure Description

[0010] To more clearly illustrate the technical solutions of the embodiments of this application, the accompanying drawings used in the embodiments will be briefly introduced below. It should be understood that the following drawings only show some embodiments of this application and should not be regarded as a limitation of the scope. For those skilled in the art, other related drawings can be obtained based on these drawings without creative effort.

[0011] Figure 1 This is a flowchart illustrating steps S101-S105 provided in the embodiments of this application; Figure 2 This is a flowchart illustrating steps S201-S202 provided in the embodiments of this application; Figure 3 This is a flowchart illustrating steps S301-S303 provided in the embodiments of this application; Figure 4 This is a flowchart illustrating steps S401-S402 provided in the embodiments of this application; Figure 5 This is a flowchart illustrating steps S501-S502 provided in the embodiments of this application; Figure 6 This is a flowchart illustrating steps S601-S602 provided in the embodiments of this application; Figure 7 This is a schematic diagram of the structure of the end-to-cloud collaborative passenger flow hot zone splicing device based on ReID deduplication provided in the embodiments of this application; Figure 8 This is a schematic diagram of the composition structure of the electronic device provided in the embodiments of this application. Detailed Implementation

[0012] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the accompanying drawings in this application are for illustrative and descriptive purposes only and are not intended to limit the scope of protection of this application. Furthermore, it should be understood that the schematic drawings are not drawn to scale. The flowcharts used in this application illustrate operations implemented according to some embodiments of this application. It should be understood that the operations in the flowcharts may not be implemented in sequence, and steps without logical contextual relationships may be reversed or implemented simultaneously. In addition, those skilled in the art, guided by the content of this application, may add one or more other operations to the flowcharts, or remove one or more operations from the flowcharts.

[0013] In the following description, references are made to “some embodiments,” which describe a subset of all possible embodiments. However, it is understood that “some embodiments” may be the same subset or different subsets of all possible embodiments and may be combined with each other without conflict.

[0014] Furthermore, the described embodiments are merely some, not all, of the embodiments of this application. The components of the embodiments of this application described and illustrated herein can typically be arranged and designed in various different configurations. Therefore, the following detailed description of the embodiments of this application provided in the accompanying drawings is not intended to limit the scope of the claimed application, but merely to illustrate selected embodiments of the application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without inventive effort are within the scope of protection of this application.

[0015] In the following description, the terms "first, second, third" are used merely to distinguish similar objects and do not represent a specific ordering of objects. It is understood that "first, second, third" may be interchanged in a specific order or sequence where permitted, so that the embodiments of this application described herein can be implemented in an order other than that illustrated or described herein.

[0016] It should be noted that the term "comprising" will be used in the embodiments of this application to indicate the presence of the features declared thereafter, but does not exclude the addition of other features.

[0017] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing embodiments of this application and is not intended to limit this application.

[0018] See Figure 1 , Figure 1 This is a flowchart illustrating steps S101-S105 of the end-to-cloud collaborative passenger flow hotspot splicing method based on ReID deduplication provided in this application embodiment. Figure 1 Steps S101-S105 are explained below.

[0019] In step S101, the system receives pedestrian structured data uploaded by each camera end with natural hours as the time window and local ReID deduplication. The structured data is obtained by the camera end performing pedestrian detection, tracking and ReID feature extraction on the real-time video stream and completing local deduplication. The local feature cache is cleared after the end uploads the data. The cloud receives structured data uploaded by multiple front-end cameras after they have completed pedestrian intelligent analysis and hourly deduplication on their respective local edge devices. Specifically, the front-end camera edge devices continuously perform pedestrian detection, multi-target tracking, ReID feature extraction, and temporal information recording on the real-time video stream. Pedestrian detection can be achieved using a lightweight YOLOv8n model for efficient edge-side inference, multi-target tracking can be achieved using the ByteTrack algorithm for continuous cross-frame ID association, and ReID features can be output as fixed-dimensional feature vectors through the OSNet network. The edge devices use natural hours as independent time windows, and perform local ReID deduplication within the window. For the same pedestrian, only the highest quality single feature data is retained, eliminating a large number of redundant features caused by pedestrians repeatedly entering and exiting in a short period of time, slight image shaking, and repeated captures in consecutive frames. The edge devices upload structured data in batches at the end of each hourly window or when the feature cache reaches its limit. The uploaded content only includes the tracking ID, the best ReID feature, the location coordinates, and the dwell time, without uploading the original video and captured images, thus balancing data privacy and transmission efficiency. After receiving data in the cloud, it serves as the data source for global analysis. Once the data is successfully uploaded to the client, the cache of the current window is immediately cleared to ensure that the data in each hourly window is independent and free from superimposed pollution.

[0020] In step S102, cross-camera ReID identity association is performed on the received multi-camera structured data, a global pedestrian trajectory database is constructed and maintained, and secondary comparison, ID merging and data correction are performed on the full historical data of the day in each specific period; Here, the cloud uses the structured data reported by all cameras at the same target location for the day, with the entire camera data as a unified comparison database, and employs a cosine similarity algorithm to complete the cross-camera pedestrian identity association. The cloud calculates similarity through feature vector dot product and modulus operation, and determines that different camera features with similarity higher than the threshold belong to the same pedestrian. The tracking IDs of multiple cameras are associated with the same global pedestrian ID, and the global pedestrian trajectory database is continuously built and dynamically updated, recording the full trajectory information of each pedestrian, such as the time of appearance, location, duration of stay, and movement order. To address issues such as truncation of hourly windows on the end-of-day side, splitting of pedestrian IDs across hours, and missed matching across cameras, the cloud automatically starts pairwise comparison of all features of the day during the off-peak business hours at the end of each calendar day, covering all feature data across cameras and hourly windows. Multiple temporary IDs of the same pedestrian that were mistakenly split are merged and unified, correcting the global ID, trajectory fragments, statistical data, and visualization results.

[0021] In step S103, multiple monitoring areas are divided based on the target location floor plan, and a binding correspondence between the camera and the monitoring area is established; Unlike traditional solutions that require complex deployment methods involving homography and similarity transformation matrix coordinate registration for each camera, this embodiment directly configures areas in the cloud based on the target location's floor plan. The cloud supports users uploading floor plans or CAD drawings of locations such as shopping malls, stores, and exhibition halls, dividing them into multiple independent monitoring sub-areas using polygon drawing, corresponding to functional zones such as shelf areas, aisle areas, cashier areas, and entrance areas. The cloud establishes a binding mapping relationship between cameras and monitoring areas, with the logical rule that a single camera is uniquely bound to one area, and a single area can be bound to one or more cameras. It supports multi-camera joint coverage for scenarios such as large aisles and atriums. This mechanism does not require mapping calibration between camera pixel coordinates and physical coordinates, nor does it rely on camera installation angles and heights; it relies solely on business logic binding to determine area affiliation, significantly simplifying the on-site deployment process. For the same area covered by multiple cameras, the cloud can subsequently perform cross-camera deduplication statistics within the area based on the ReID global ID, eliminating duplicate counting errors in overlapping areas.

[0022] In step S104, each monitoring area is used as a statistical unit to count the number of unique pedestrians N_k and the average stay time T_k. Combined with the preset time weight coefficient λ, the regional heat H_k of each monitoring area is calculated by the preset weight formula. Based on the heat value, the regional hierarchical rendering display is completed on the plan. Here, the cloud platform treats each monitored area as an independent statistical unit, performing quantitative calculations of heat volume based on deduplicated real passenger flow data. Core statistical variables include the deduplicated number of pedestrians N_k, the average dwell time per person in the area T_k, and the time weighting coefficient λ. N_k represents the scale of passenger flow to the area, T_k represents customer dwell preferences and activity levels, and λ adjusts the contribution weight of dwell time to heat volume. The cloud platform calculates the quantitative heat value H_k for each area using the fixed formula H_k=N_k×(1+λ×T_k), achieving a fusion of passenger flow quantity and dwell behavior in a quantitative evaluation. Based on the heat value of each area, the cloud platform divides it into high, medium, and low heat levels, and uses gradient or graded color schemes to fill and render each area on the plan view, providing an intuitive and visual display of the area heat distribution.

[0023] In step S105, pedestrian cross-regional continuous movement lines are spliced ​​based on the global pedestrian trajectory database, the probability of passenger flow conversion between regions is statistically analyzed, and a spatial flow topology network is constructed to generate corresponding passenger flow line visualization results.

[0024] Finally, based on a global pedestrian trajectory database, the cloud-based system stitches together discrete trajectory fragments of the same pedestrian scattered across different cameras and at different times, chronologically, to form a complete and continuous cross-regional passenger flow sequence, fully reconstructing the customer's entire movement path. The cloud-based system aggregates the entire venue's flow data according to a preset statistical period, counting the number of passenger flow transfers between every two adjacent monitored areas and calculating the directed conversion probability between areas. This constructs a spatial flow topology network with monitored areas as network nodes, passenger flow transfers as directed edges, and conversion probabilities as edge weights, accurately depicting the spatial flow structure and transfer patterns of passenger flow within the venue. Based on the topology network and heat map data, the cloud-based system performs comprehensive intelligent analysis and generates overlay-style visualization results, achieving an integrated presentation of static heat map distribution and dynamic passenger flow direction.

[0025] In some embodiments, see Figure 2 , Figure 2 This is a flowchart illustrating steps S201-S202 provided in the embodiments of this application. The method further includes steps S201-S202, which will be explained in conjunction with each step.

[0026] In step S201, a filtered personnel feature library containing a food delivery worker feature library and a courier feature library is pre-built and stored, and the filtered personnel feature library is sent to each end-side camera so that the camera end-side prioritizes matching the filtered personnel feature library when performing local ReID comparison. In step S202, when the receiving end uploads data, invalid passenger flow data marked by the end is removed. The invalid passenger flow refers to pedestrians whose features match the filter personnel feature database. Invalid passenger flow does not participate in the area heat statistics or the probability statistics of movement route conversion.

[0027] Here, a general invalid personnel feature library is pre-trained and solidified in the cloud, mainly including feature libraries for food delivery workers and couriers. These feature libraries are trained based on common appearance features of delivery personnel such as clothing, helmets, delivery boxes, and body posture, and can be adapted to various commercial venues for general filtering scenarios. The cloud distributes the pre-set feature library in batches to all edge cameras within the venue, enabling each device to hold an offline filtering model locally. Before performing regular ReID comparison and deduplication, the device prioritizes matching the real-time detected pedestrian features with the invalid personnel feature library. If the similarity matches a preset threshold, the pedestrian is directly marked as invalid. For invalid pedestrians, the device directly abandons the upload of their feature vectors and structured data, preventing invalid data from entering the cloud database and eliminating the consumption of storage and computing resources from the source. Simultaneously, the cloud completely removes invalid pedestrian data throughout the entire process of subsequent popularity statistics, dwell time calculation, movement probability statistics, and topology network construction, ensuring that all statistical results only reflect the behavioral characteristics of real customers.

[0028] In some embodiments, see Figure 3 , Figure 3 This is a flowchart illustrating steps S301-S303 provided in the embodiments of this application. The filtered personnel feature library also includes a store employee uniform feature library, which can be constructed through steps S301-S303. The explanation will be based on each step.

[0029] In step S301, the designated camera at the target location is configured to enable the pedestrian cropping image reporting function, and the pedestrian cropping images uploaded by the receiving end are manually labeled with the store clerk's identity. In step S302, the ReID feature vectors corresponding to the labeled images of the same store clerk from multiple perspectives and in multiple scenes are aggregated to generate a store clerk uniform feature library adapted to the corresponding target location. In step S303, the employee uniform feature database is sent to all end-side cameras in the target location so that the end-side cameras can perform local filtering of employee and customer traffic.

[0030] Here, the cloud first designates specific cameras for the target location, enabling pedestrian cropped image reporting permissions. This allows the client to upload detected pedestrian cropped images to the cloud, uploading only the images and not the feature data, ensuring lightweight data and privacy. Secondly, cloud administrators manually classify and label the reported batch of images, accurately marking employee sample images. The labeling process prioritizes multi-view samples of the same employee from the front, side, back, under different lighting conditions, and in different poses, ensuring sample diversity. Thirdly, the cloud aggregates the ReID feature vectors of all labeled employee samples, generating stable and robust general-purpose employee uniform feature vectors through mean aggregation or cluster center extraction, integrating them into a unique employee feature library for that store. Finally, the cloud distributes the updated employee uniform feature library to all camera clients with a single click. During local matching, the client prioritizes matching the employee feature library; if a match is found, it is filtered out without uploading, achieving normalized and precise filtering of employee and customer traffic. When store uniforms change or personnel shifts occur, the above process can be repeated to iteratively update the feature library.

[0031] In some embodiments, see Figure 4 , Figure 4 This is a flowchart illustrating steps S401-S402 provided in the embodiments of this application. The method further includes steps S401-S402, which will be explained in conjunction with each step.

[0032] In step S401, the feature reporting rules of the end-side camera are configured so that the camera end-side reduces the reporting frequency for a continuously stable tracking ID; In step S402, high-frequency reporting is restored for scenarios involving newly added tracking IDs, ID switching, or significant changes in human characteristics.

[0033] This embodiment features two modes: steady-state frequency reduction and high-frequency reporting for anomalies. For pedestrians who remain stationary in the image for an extended period and whose identities are stable, the device continuously monitors the ReID feature similarity corresponding to the same tracking ID. If the feature similarity remains higher than the confidence threshold configured in the cloud within a preset duration T, the pedestrian is determined to be in a stable stationary state, eliminating the need for high-frequency data collection. The device automatically switches from the default 1-frame-per-second high-frequency reporting mode to a low-frequency reporting mode of N-second single-frames, reducing repetitive and redundant feature data output. Conversely, when the device detects any anomaly, such as the generation of a new tracking ID, a switch of an existing tracking ID, or significant changes in pedestrian posture, clothing, or backpack status leading to abrupt feature mutations, it immediately reverts to the highest-frequency reporting mode of 1-frame-per-second. This ensures that new pedestrian states, identities, and behaviors are captured promptly, avoiding missed or incorrect identity matching due to delayed feature updates. This approach balances data integrity with the need for lightweight system operation.

[0034] In some embodiments, see Figure 5 , Figure 5 This is a flowchart illustrating steps S501-S502 provided in the embodiments of this application. Local ReID deduplication on the end side of the natural hour window can be achieved through steps S501-S502, which will be explained in conjunction with each step.

[0035] In step S501, the deduplication rules for the end-side natural hour window are configured so that the camera end-side can complete the ReID feature cosine similarity comparison, optimal feature retention, and new feature cache within a single natural hour window; In step S502, after the hourly window ends or the number of features exceeds the limit, structured data is uploaded in batches and the local cache is cleared.

[0036] Here, the client-side uses natural hours as strictly independent statistical time windows. Each hourly window is independently cached, deduplicated, uploaded, and cleared, with data between windows remaining independent. During the operation of a single hourly window, after detecting a pedestrian and extracting the ReID feature vector, the client immediately performs a cosine similarity comparison with all historical feature vectors already cached in the current window. If the similarity between the new feature and existing cached features exceeds a preset threshold, it is determined that the same pedestrian has been detected repeatedly. The client then compares the quality scores of the two features, taking into account multiple dimensions such as human bounding box integrity, image clarity, detection confidence, and pose positivity. Only the best feature vector with the higher score is retained, while low-quality redundant features are discarded, ensuring that the best features for a single pedestrian are always retained in the cache. If the similarity does not reach the threshold, it is determined to be a completely new and independent pedestrian, and the new feature is added to the local cache. When the current natural hour ends, or the number of deduplicated features cached in the window reaches the preset cloud cache limit, the client stops caching for the current window and batch packages all the best features and structured data of this window for uploading to the cloud. After successful upload verification, the client-side forcibly clears all feature caches for the current window and starts caching again from zero in the next natural hour to avoid data residue and pollution across windows.

[0037] In some embodiments, see Figure 6 , Figure 6 This is a flowchart illustrating steps S601-S602 provided in the embodiments of this application. The process of performing a second comparison, ID merging, and data correction on all historical data of the day in each specific period can be achieved through steps S601-S602, which will be explained in conjunction with each step.

[0038] In step S601, all deduplicated ReID feature vectors uploaded by all end-side devices on the same day are traversed, and full pairwise cosine similarity comparisons are performed across cameras and across natural hour windows. Global ID merging is performed on pedestrians with the same origin but different IDs. In step S602, the complete pedestrian trajectory is spliced ​​based on the merged unique global ID, the number of pedestrians, dwell time, heat value and area conversion probability of the corresponding area are recalculated, and the global trajectory database and visualization data are updated.

[0039] This embodiment automatically initiates a full-data correction task after the end of each natural day. It iterates through all deduplicated ReID feature vectors reported by all cameras and all natural hour windows within the target location for the day, performing a full-coverage, pairwise cosine similarity comparison without omission. The comparison scope covers both cross-camera and cross-time window scenarios. For multiple different temporary ID features with similarity exceeding a preset threshold, the cloud determines that they correspond to the same real pedestrian in the physical world and automatically merges multiple scattered temporary IDs into a unique global ID. After ID merging, the cloud reassembles and merges the discrete trajectory fragments of the pedestrian across all time periods, all cameras, and all areas throughout the day according to the time axis, generating a complete, continuous, and unbroken global pedestrian trajectory. Based on the corrected complete trajectory, the cloud iteratively calculates the visit records and cumulative dwell time of the pedestrian in each area, thereby updating the deduplicated number of people N_k, average dwell time T_k, area popularity H_k, and inter-area passenger flow conversion probability in each area. Finally, it synchronously updates the global trajectory database, area popularity distribution map, and traffic topology visualization results to ensure that the daily statistical data is completely accurate and reliable.

[0040] In some embodiments, the rules for regional passenger flow statistics and popularity calculation include: When performing regional data statistics, N_k is the number of unique pedestrians in the corresponding monitoring area within the preset statistical period. If the same pedestrian enters the same area multiple times within the same statistical period, only one count is made. For monitoring areas with multiple cameras, the cloud performs unified deduplication and counting of the same pedestrian data collected by multiple overlapping cameras. T_k represents the average dwell time of all unique pedestrians within the corresponding monitoring area. If the same pedestrian enters the same area multiple times, the dwell time of each time in the area is accumulated and then included in the average calculation. The λ is a preset time weighting coefficient, and the dimension of λ matches the time unit of the stay duration T_k, which is used to adjust the contribution weight of the stay duration to the regional heat intensity.

[0041] Specifically, for the deduplicated number of pedestrians in a region, N_k represents the number of unique, real customers in the corresponding monitored area within the statistical period, strictly based on globally unique ReID IDs for deduplication. Regardless of whether the same pedestrian enters or exits the area multiple times within the statistical period, or is simultaneously captured by multiple cameras in the area, the cloud only counts it once, completely eliminating duplicate counting and overlapping capture errors, and truly reflecting the breadth of customer visits to the area. For the average dwell time per person in the region, T_k, the cloud accumulates the single dwell time of the same pedestrian entering the same area multiple times to obtain the total dwell time of the pedestrian in the area. Then, the total dwell time of all deduplicated pedestrians in the area is summed and divided by N_k to obtain the average dwell time per person in the area, objectively reflecting the overall customer dwell preference in the area. For the time weighting coefficient λ, the dimension of this parameter matches the unit of time and is used to linearly adjust the contribution ratio of dwell time to popularity, supporting scenario adaptation. It can be adjusted to emphasize the number of pedestrians or the popularity of dwell time according to business needs. Meanwhile, for large areas with multiple cameras, the cloud performs cross-camera ReID deduplication across the entire area, unifies the statistical standards for the area, and ensures that the statistical results for large areas are accurate.

[0042] In some embodiments, the method further includes: When the heat value H_k of the target monitoring area exceeds the preset heat threshold, the passenger flow source area is traced back along the directed edge of the spatial flow topology network. The passenger flow source tracing results are output in descending order of region conversion probability. Alternatively, based on the real-time heat values ​​of each monitored area and the probability of passenger flow conversion between areas, the heat distribution of each area in the next statistical period can be predicted. Alternatively, when the probability of passenger flow conversion between any two monitored areas exceeds a preset multiple of the historical average, or when the heat of the target area exceeds the limit and continues for a preset duration, an abnormal alarm will be automatically triggered.

[0043] This embodiment also provides three types of intelligent functions. First, congestion source tracing: The cloud monitors the heat value of each area in real time. When a target area H_k continuously or instantaneously exceeds a preset heat threshold, it is determined that the area is experiencing congestion or abnormal customer flow. Relying on the constructed directed spatial topology network, the cloud traces back all directed edges pointing to the upstream areas of the congested area, extracts all customer flow source areas, and prioritizes the source areas based on historical and real-time conversion probabilities to accurately locate the main source of congested customer flow, providing a basis for on-site customer flow management, personnel diversion, and business format adjustment. Second, heat prediction: Based on the current real-time heat base of each area, the stable customer flow conversion probability between areas, and the customer flow time-series continuity decay coefficient, the cloud establishes a heat prediction model to predict the heat distribution changes of each area in the next statistical period, realizing advance prediction of customer flow trends and supporting merchants to adjust their display, scheduling, and customer attraction strategies in advance. Third, the abnormal alarm function: The cloud monitors the sudden changes in the conversion probability between areas and the status of area popularity exceeding the limit in real time. When the conversion probability of any pair of areas far exceeds the historical average, or the area popularity exceeds the limit and continues for a preset time, the cloud automatically identifies abnormal passenger flow and triggers the abnormal alarm mechanism to promptly report risks such as abnormal passenger flow, instantaneous gathering, and sudden changes in movement.

[0044] In some embodiments, the step of stitching together continuous pedestrian movement lines across regions based on a global pedestrian trajectory database, statistically analyzing the probability of passenger flow conversion between regions, constructing a spatial flow topology network, and generating corresponding passenger flow line visualization results includes: The system sorts, categorizes, and colors the passenger flow data between regions. The top N passenger flow groups are displayed with different colors, while the remaining groups are displayed together. The regional heat map layer, the topology network layer, and the floor plan layer are overlaid to achieve an integrated visualization.

[0045] Here, the cloud platform first sorts the passenger flow conversion traffic between all areas within the statistical period, selecting the top N core flow paths and rendering them with distinct color schemes to highlight the mainstream passenger flow lines. All secondary and niche flow paths ranked below N are uniformly merged and categorized as "Others," using a unified color scheme to avoid cluttered visuals and obscuring core trends. The cloud platform employs an integrated layered display method, overlaying the site plan, area heat hierarchy rendering layer, and directed passenger flow topology network layer onto the same visualization canvas. This achieves a fusion display of static area heat distribution and dynamic passenger flow direction and intensity. The direction of the movement arrows represents the direction of passenger movement, the arrow width corresponds to the conversion probability, and the heat color scheme represents the area's congestion and activity level, allowing staff to intuitively and comprehensively grasp the spatial distribution and flow patterns of passenger flow throughout the venue.

[0046] In summary, the embodiments of this application have the following beneficial effects: By employing a lightweight, hourly ReID deduplication preprocessing method on the edge, dynamic adaptive feature reporting frequency adjustment, and a multi-level mechanism for precise filtering of invalid passenger flow, the computational power and network bandwidth consumption of edge devices are significantly reduced. Invalid interference data such as that from store clerks and delivery personnel are eliminated from the data source, effectively solving the problems of duplicate counting, data redundancy, and statistical distortion inherent in traditional passenger flow statistics methods. Simultaneously, relying on a cloud-based full-feature secondary comparison and correction mechanism across cameras and time windows, the defects of short-time window data fragmentation, pedestrian ID splitting, and cross-scene mismatches on the edge are compensated for. This significantly improves the accuracy of global pedestrian trajectory stitching, passenger flow statistics, dwell time calculation, and regional heat quantification results, without requiring complex coordinate calibration of cameras; it only requires logic based on the planar map area. The binding process allows for rapid scenario deployment, significantly reducing on-site implementation costs and deployment barriers. Building upon this foundation, this application constructs a standardized regional heat quantification model and spatial passenger flow topology network. This enables integrated intelligent analysis capabilities, including visualized display of venue area heat levels, passenger flow topology presentation, congestion tracing, heat prediction, and anomaly alarms. It can comprehensively and meticulously reconstruct the spatial distribution and dynamic flow patterns of passenger flow in commercial venues, ensuring the real-time nature, stability, and accuracy of passenger flow data analysis while enhancing the intelligence and visualization of passenger flow operation analysis. This provides precise and reliable data support for business model optimization, passenger flow management, staff scheduling, and scenario control in commercial scenarios, possessing strong scenario adaptability, engineering feasibility, and practical promotion value.

[0047] Based on the same inventive concept, this application also provides a device for splicing passenger flow hot zones based on ReID deduplication, which corresponds to the method for splicing passenger flow hot zones based on ReID deduplication in the first embodiment. Since the principle of the device in this application is similar to the above-mentioned method for splicing passenger flow hot zones based on ReID deduplication, the implementation of the device can refer to the implementation of the method, and the repeated parts will not be described again.

[0048] like Figure 7 As shown, Figure 7 This is a schematic diagram of the structure of the ReID-based deduplication-based end-to-end cloud collaborative passenger flow hotspot splicing device 700 provided in this application embodiment. The ReID-based deduplication-based end-to-end cloud collaborative passenger flow hotspot splicing device 700 includes: The data receiving module 701 is used to receive pedestrian structured data uploaded by each camera end with natural hours as the time window and after local ReID deduplication. The structured data is obtained by the camera end performing pedestrian detection, tracking and ReID feature extraction on the real-time video stream and completing local deduplication. After the end uploads the data, the local feature cache is cleared. The ReID verification module 702 is used to perform cross-camera ReID identity association on the received multi-camera structured data, build and maintain a global pedestrian trajectory database, and perform secondary comparison, ID merging and data correction on the full historical data of the day at each specific period; The area configuration module 703 is used to divide multiple monitoring areas based on the floor plan of the target location and establish a binding relationship between the camera and the monitoring area. The statistical calculation module 704 is used to count the number of unique pedestrians N_k and the average stay time T_k per person, taking each monitored area as a statistical unit. Combined with the preset time weighting coefficient λ, the regional heat H_k of each monitored area is calculated through the preset weighting formula. Based on the heat value, the regional hierarchical rendering display is completed on the plan. The pedestrian flow visualization module 705 is used to stitch together continuous pedestrian flow lines across regions based on a global pedestrian trajectory database, statistically analyze the probability of passenger flow conversion between regions, construct a spatial flow topology network, and generate corresponding passenger flow visualization results.

[0049] Those skilled in the art should understand that Figure 7 The functions of each unit in the ReID-based deduplication end-to-end cloud collaborative passenger flow hot zone splicing device 700 can be understood by referring to the relevant description of the ReID-based deduplication end-to-end cloud collaborative passenger flow hot zone splicing method. Figure 7 The functions of each unit in the ReID-based deduplication-based end-to-cloud collaborative passenger flow hot zone splicing device 700 shown can be implemented by a program running on a processor or by specific logic circuits.

[0050] In one possible implementation, the data receiving module 701 is further configured to: A filtered personnel feature library containing the feature libraries of food delivery workers and couriers is pre-built and stored, and the filtered personnel feature library is distributed to each end-side camera so that the camera end-side prioritizes matching the filtered personnel feature library when performing local ReID comparison. When uploading data at the receiving end, invalid passenger flow data marked by the receiving end is removed. The invalid passenger flow refers to pedestrians whose features match the filter personnel feature database. Invalid passenger flow is not included in the regional heat statistics or the probability statistics of movement route conversion.

[0051] In one possible implementation, the personnel feature library further includes a store employee uniform feature library, which is constructed in the following manner: Configure the designated cameras in the target location to enable the pedestrian cropped image reporting function, receive the pedestrian cropped images uploaded by the receiving end and complete the manual labeling of the store clerk's identity; Aggregate ReID feature vectors corresponding to labeled images of the same store clerk from multiple perspectives and in multiple scenes to generate a store clerk uniform feature library adapted to the corresponding target location; The store employee uniform feature database is distributed to all edge cameras in the target location to enable local filtering of store employee and customer traffic at the edge.

[0052] In one possible implementation, the data receiving module 701 is further configured to: Configure the feature reporting rules for the end-side camera to reduce the reporting frequency for a continuously stable tracking ID; Restore high-frequency reporting for scenarios involving new tracking IDs, ID switching, or significant changes in human characteristics.

[0053] In one possible implementation, local ReID deduplication within the end-side natural hour window is achieved in the following way: Configure the deduplication rules for the natural hour window on the camera side so that the camera side can complete the ReID feature cosine similarity comparison, optimal feature retention, and new feature caching within a single natural hour window; After the hourly window ends or the number of features exceeds the limit, upload structured data in batches and clear the local cache.

[0054] In one possible implementation, the ReID verification module 702 is specifically used for: Iterate through all the deduplicated ReID feature vectors uploaded by all devices on the same day, perform full pairwise cosine similarity comparisons across cameras and natural hour windows, and perform global ID merging for pedestrians with the same origin but different IDs. Based on the merged unique global ID, the complete pedestrian trajectory is pieced together, the number of pedestrians, dwell time, popularity value and area conversion probability of the corresponding area are recalculated, and the global trajectory database and visualization data are updated.

[0055] In one possible implementation, the rules for regional passenger flow statistics and popularity calculation include: When performing regional data statistics, N_k is the number of unique pedestrians in the corresponding monitoring area within the preset statistical period. If the same pedestrian enters the same area multiple times within the same statistical period, only one count is made. For monitoring areas with multiple cameras, the cloud performs unified deduplication and counting of the same pedestrian data collected by multiple overlapping cameras. T_k represents the average dwell time of all unique pedestrians within the corresponding monitoring area. If the same pedestrian enters the same area multiple times, the dwell time of each time in the area is accumulated and then included in the average calculation. The λ is a preset time weighting coefficient, and the dimension of λ matches the time unit of the stay duration T_k, which is used to adjust the contribution weight of the stay duration to the regional heat intensity.

[0056] In one possible implementation, the traffic flow visualization module 705 is further used for: When the heat value H_k of the target monitoring area exceeds the preset heat threshold, the passenger flow source area is traced back along the directed edge of the spatial flow topology network. The passenger flow source tracing results are output in descending order of region conversion probability. Alternatively, based on the real-time heat values ​​of each monitored area and the probability of passenger flow conversion between areas, the heat distribution of each area in the next statistical period can be predicted. Alternatively, when the probability of passenger flow conversion between any two monitored areas exceeds a preset multiple of the historical average, or when the heat of the target area exceeds the limit and continues for a preset duration, an abnormal alarm will be automatically triggered.

[0057] In one possible implementation, the movement path visualization module 705 is specifically used for: The system sorts, categorizes, and colors the passenger flow data between regions. The top N passenger flow groups are displayed with different colors, while the remaining groups are displayed together. The regional heat map layer, the topology network layer, and the floor plan layer are overlaid to achieve an integrated visualization.

[0058] The aforementioned edge-cloud collaborative passenger flow hotspot stitching device based on ReID deduplication significantly reduces the computing power and network bandwidth consumption of edge devices through hourly lightweight ReID deduplication preprocessing on the edge side, dynamic adaptive feature reporting frequency adjustment, and a multi-level invalid passenger flow precision filtering mechanism. It removes invalid interference data such as store clerks and delivery personnel from the data source, effectively solving the problems of duplicate counting, data redundancy, and statistical distortion in traditional passenger flow statistics methods. At the same time, relying on the cloud-based cross-camera and cross-time window full feature secondary comparison and correction mechanism, it makes up for the defects of short-time window data fragmentation, pedestrian ID splitting, and cross-scene missed matching on the edge side, greatly improving the accuracy of global pedestrian trajectory stitching, passenger flow statistics, dwell time calculation, and regional heat quantification results, without the need for complex coordinate calibration of cameras. This application establishes a standardized regional heat quantification model and spatial passenger flow topology network. This enables rapid deployment of scenarios simply by logically binding areas on a floor plan, significantly reducing on-site implementation costs and deployment barriers. Furthermore, it constructs an integrated intelligent analysis capability encompassing visualized regional heat levels, passenger flow topology presentation, congestion tracing, heat prediction, and anomaly alerts. This capability comprehensively and meticulously recreates the spatial distribution and dynamic flow patterns of passenger flow in commercial venues. It ensures the real-time performance, stability, and accuracy of passenger flow data analysis while enhancing the intelligence and visualization of passenger flow operation analysis. This provides precise and reliable data support for business optimization, passenger flow management, staff scheduling, and scenario control in commercial scenarios, demonstrating strong scenario adaptability, engineering feasibility, and practical promotion value.

[0059] like Figure 8 As shown, Figure 8This is a schematic diagram of the composition structure of the electronic device 800 provided in the embodiments of this application. The electronic device 800 includes: The device includes a processor 801, a storage medium 802, and a bus 803. The storage medium 802 stores machine-readable instructions that can be executed by the processor 801. When the electronic device 800 is running, the processor 801 communicates with the storage medium 802 via the bus 803. The processor 801 executes the machine-readable instructions to perform the steps of the end-to-cloud collaborative passenger flow hot zone splicing method based on ReID deduplication described in the embodiments of this application.

[0060] In practical applications, the various components in the electronic device 800 are coupled together via bus 803. It can be understood that bus 803 is used to achieve communication between these components. In addition to a data bus, bus 803 also includes a power bus, a control bus, and a status signal bus. However, for clarity, in... Figure 8 The general labeled all buses as Bus 803.

[0061] The aforementioned electronic devices significantly reduce the computing power and network bandwidth consumption of edge devices through hourly lightweight ReID deduplication preprocessing on the edge, dynamic adaptive feature reporting frequency adjustment, and a multi-level invalid passenger flow precise filtering mechanism. This eliminates invalid interference data such as that from store clerks and delivery personnel at the data source, effectively solving the problems of duplicate counting, data redundancy, and statistical distortion in traditional passenger flow statistics methods. Simultaneously, relying on a cloud-based full-feature secondary comparison and correction mechanism across cameras and time windows, it compensates for the shortcomings of short-time window data fragmentation, pedestrian ID splitting, and cross-scene mismatch on the edge, greatly improving the accuracy of global pedestrian trajectory stitching, passenger flow statistics, dwell time calculation, and regional heat quantification results. Furthermore, it eliminates the need for complex coordinate calibration of cameras, relying solely on a planar map area... Domain logic binding enables rapid scenario deployment, significantly reducing on-site implementation costs and deployment barriers. Building upon this, this application constructs a standardized regional heat quantification model and spatial passenger flow topology network, achieving integrated intelligent analysis capabilities including visualized display of venue regional heat levels, passenger flow topology presentation, congestion tracing, heat prediction, and anomaly alarms. This allows for a comprehensive and detailed reconstruction of the spatial distribution and dynamic flow patterns of passenger flow in commercial venues, ensuring the real-time nature, stability, and accuracy of passenger flow data analysis while enhancing the intelligence and visualization of passenger flow operation analysis. It provides precise and reliable data support for business format optimization, passenger flow management, staff scheduling, and scenario control in commercial scenarios, possessing strong scenario adaptability, engineering feasibility, and practical promotion value.

[0062] This application also provides a computer-readable storage medium storing executable instructions. When the executable instructions are executed by at least one processor 801, the end-to-cloud collaborative passenger flow hot zone splicing method based on ReID deduplication described in this application is implemented.

[0063] In some embodiments, the storage medium may be a magnetic random access memory (FRAM), a read-only memory (ROM), or a programmable read-only memory (PROM). Erasable Programmable Read-Only Memory (EPROM) Electrically Erasable Programmable Read-Only Memory (EEPROM) Read-only memory, flash memory, magnetic surface storage, optical disc, or CD-ROM ROM, Compact Disc Read It can be a memory such as a memory only; or it can be a device that includes one or any combination of the above-mentioned memories.

[0064] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0065] As an example, executable instructions may, but do not necessarily, correspond to files in the file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a HyperText Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborating files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0066] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0067] The aforementioned computer-readable storage medium significantly reduces the computing power and network bandwidth consumption of edge devices through hourly lightweight ReID deduplication preprocessing on the edge, dynamic adaptive feature reporting frequency adjustment, and a multi-level invalid passenger flow precise filtering mechanism. It removes invalid interference data such as that from store clerks and delivery personnel at the data source, effectively solving the problems of duplicate counting, data redundancy, and statistical distortion in traditional passenger flow statistics methods. Simultaneously, relying on a cloud-based cross-camera, cross-time window full feature secondary comparison and correction mechanism, it compensates for the shortcomings of short-time window data fragmentation, pedestrian ID splitting, and cross-scene missed matching on the edge, greatly improving the accuracy of global pedestrian trajectory stitching, passenger flow statistics, dwell time calculation, and regional heat quantification results. Furthermore, it eliminates the need for complex coordinate calibration of cameras, relying solely on planar... Logical binding of graph regions allows for rapid scenario deployment, significantly reducing on-site implementation costs and deployment barriers. Building upon this, this application constructs a standardized regional heat quantification model and spatial passenger flow topology network. This enables integrated intelligent analysis capabilities, including visualized display of venue regional heat levels, passenger flow topology presentation, congestion tracing, heat prediction, and anomaly alarms. It can comprehensively and meticulously reconstruct the spatial distribution and dynamic flow patterns of passenger flow in commercial venues, ensuring the real-time nature, stability, and accuracy of passenger flow data analysis while enhancing the intelligence and visualization of passenger flow operation analysis. It provides precise and reliable data support for business format optimization, passenger flow management, staff scheduling, and scenario control in commercial scenarios, possessing strong scenario adaptability, engineering feasibility, and practical promotion value.

[0068] In the several embodiments provided in this application, it should be understood that the disclosed methods and electronic devices can be implemented in other ways. The device embodiments described above are merely illustrative. For example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods, such as: multiple units or components may be combined, or integrated into another system, or some features may be ignored or not executed. In addition, the coupling, direct coupling, or communication connection between the various components shown or discussed may be through some interfaces, and the indirect coupling or communication connection between devices or units may be electrical, mechanical, or other forms.

[0069] The modules described as separate components may or may not be physically separate. The components shown as modules may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0070] In addition, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit.

[0071] If the aforementioned functions are implemented as software functional units and sold or used as independent products, they can be stored in a processor-executable, non-volatile, computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, a platform server, or a network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this application. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, ROM, RAM, magnetic disks, or optical disks.

[0072] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

Claims

1. A method for edge-cloud collaborative passenger flow hotspot splicing based on ReID deduplication, characterized in that, Includes the following steps: The system receives pedestrian structured data uploaded by each camera end with a time window of natural hours and after local ReID deduplication. The structured data is obtained by the camera end performing pedestrian detection, tracking and ReID feature extraction on the real-time video stream and completing local deduplication. The local feature cache is cleared after the end uploads the data. Perform cross-camera ReID identity association on the received multi-camera structured data, build and maintain a global pedestrian trajectory database, and perform secondary comparison, ID merging and data correction on the full historical data of the day at each specific period; Based on the floor plan of the target location, multiple monitoring areas are divided, and a binding relationship is established between the cameras and the monitoring areas. Using each monitored area as a statistical unit, the number of unique pedestrians N_k and the average stay time T_k are counted. Combined with the preset time weighting coefficient λ, the regional heat H_k of each monitored area is calculated through the preset weighting formula. Based on the heat value, the regional hierarchical rendering display is completed on the plan. Based on a global pedestrian trajectory database, continuous pedestrian movement lines across regions are stitched together, the probability of passenger flow conversion between regions is statistically analyzed, and a spatial flow topology network is constructed to generate corresponding passenger flow line visualization results.

2. The method according to claim 1, characterized in that, The method further includes: A filtered personnel feature library containing the feature libraries of food delivery workers and couriers is pre-built and stored, and the filtered personnel feature library is distributed to each end-side camera so that the camera end-side prioritizes matching the filtered personnel feature library when performing local ReID comparison. When uploading data at the receiving end, invalid passenger flow data marked by the receiving end is removed. The invalid passenger flow refers to pedestrians whose features match the filter personnel feature database. Invalid passenger flow is not included in the regional heat statistics or the probability statistics of movement route conversion.

3. The method according to claim 2, characterized in that, The filtered personnel feature database also includes a store employee uniform feature database, which is constructed in the following way: Configure the designated cameras in the target location to enable the pedestrian cropped image reporting function, receive the pedestrian cropped images uploaded by the receiving end and complete the manual labeling of the store clerk's identity; Aggregate ReID feature vectors corresponding to labeled images of the same store clerk from multiple perspectives and in multiple scenes to generate a store clerk uniform feature library adapted to the corresponding target location; The store employee uniform feature database is distributed to all edge cameras in the target location to enable local filtering of store employee and customer traffic at the edge.

4. The method according to claim 1, characterized in that, The method further includes: Configure the feature reporting rules for the end-side camera to reduce the reporting frequency for a continuously stable tracking ID; Restore high-frequency reporting for scenarios involving new tracking IDs, ID switching, or significant changes in human characteristics.

5. The method according to claim 1, characterized in that, Local ReID deduplication within the natural hour window on the client side is achieved through the following method: Configure the deduplication rules for the natural hour window on the camera side so that the camera side can complete the ReID feature cosine similarity comparison, optimal feature retention, and new feature caching within a single natural hour window; After the hourly window ends or the number of features exceeds the limit, upload structured data in batches and clear the local cache.

6. The method according to claim 1, characterized in that, The process of performing a second comparison, ID merging, and data correction on all historical data for each specific period includes: Iterate through all the deduplicated ReID feature vectors uploaded by all devices on the same day, perform full pairwise cosine similarity comparisons across cameras and natural hour windows, and perform global ID merging for pedestrians with the same origin but different IDs. Based on the merged unique global ID, the complete pedestrian trajectory is pieced together, the number of pedestrians, dwell time, popularity value and area conversion probability of the corresponding area are recalculated, and the global trajectory database and visualization data are updated.

7. The method according to claim 1, characterized in that, The rules for regional passenger flow statistics and popularity calculation include: When performing regional data statistics, N_k is the number of unique pedestrians in the corresponding monitoring area within the preset statistical period. If the same pedestrian enters the same area multiple times within the same statistical period, only one count is made. For monitoring areas with multiple cameras, the cloud performs unified deduplication and counting of the same pedestrian data collected by multiple overlapping cameras. T_k represents the average dwell time of all unique pedestrians within the corresponding monitoring area. If the same pedestrian enters the same area multiple times, the dwell time of each time in the area is accumulated and then included in the average calculation. The λ is a preset time weighting coefficient, and the dimension of λ matches the time unit of the stay duration T_k, which is used to adjust the contribution weight of the stay duration to the regional heat intensity.

8. The method according to claim 1, characterized in that, The method further includes: When the heat value H_k of the target monitoring area exceeds the preset heat threshold, the passenger flow source area is traced back along the directed edge of the spatial flow topology network. The passenger flow source tracing results are output in descending order of region conversion probability. Alternatively, based on the real-time heat values ​​of each monitored area and the probability of passenger flow conversion between areas, the heat distribution of each area in the next statistical period can be predicted. Alternatively, when the probability of passenger flow conversion between any two monitored areas exceeds a preset multiple of the historical average, or when the heat of the target area exceeds the limit and continues for a preset duration, an abnormal alarm will be automatically triggered.

9. The method according to claim 1, characterized in that, The process involves stitching together continuous pedestrian movement lines across regions based on a global pedestrian trajectory database, statistically analyzing the probability of passenger flow transitions between regions, constructing a spatial flow topology network, and generating corresponding passenger flow line visualization results, including: The system sorts, categorizes, and colors the passenger flow data between regions. The top N passenger flow groups are displayed with different colors, while the remaining groups are displayed together. The regional heat map layer, the topology network layer, and the floor plan layer are overlaid to achieve an integrated visualization.

10. A cloud-edge collaborative passenger flow hotspot splicing device based on ReID deduplication, characterized in that, The device includes: The data receiving module is used to receive pedestrian structured data uploaded by each camera end with natural hours as the time window and after local ReID deduplication. The structured data is obtained by the camera end performing pedestrian detection, tracking and ReID feature extraction on the real-time video stream and completing local deduplication. After the end uploads the data, the local feature cache is cleared. The ReID verification module is used to perform cross-camera ReID identity association on the received multi-camera structured data, build and maintain a global pedestrian trajectory database, and perform secondary comparison, ID merging and data correction on the full historical data of the day at each specific period. The area configuration module is used to divide multiple monitoring areas based on the floor plan of the target location and establish a binding relationship between cameras and monitoring areas. The statistical calculation module is used to count the number of unique pedestrians N_k and the average stay time T_k per person, taking each monitored area as a statistical unit. Combined with the preset time weighting coefficient λ, the module calculates the regional heat H_k of each monitored area through a preset weighting formula. Based on the heat value, the module completes the regional hierarchical rendering display on the plan. The pedestrian flow visualization module is used to stitch together continuous pedestrian flow lines across regions based on a global pedestrian trajectory database, calculate the probability of passenger flow conversion between regions, construct a spatial flow topology network, and generate corresponding passenger flow visualization results.

Images