Mobile stall tracking method and device, electronic equipment and storage medium

Abstract

The invention is suitable for the technical field of artificial intelligence, and provides a mobile booth tracking method, which comprises the following steps: when a fixed camera in a current area detects a mobile booth to be tracked, determining event persistence and a moving direction of the mobile booth to be tracked; according to the event persistence and the moving direction, determining a next area to which the to-be-tracked flowing stall goes; if the untrackable distance between the next area and the current area is greater than the preset distance, searching a patrol robot in a predetermined area; wherein the preset distance is dynamically set according to the event occurrence probability between the current area and the next area; and when the event discovery probability of the patrol robot is smaller than a preset event discovery probability, scheduling the patrol robot to execute a tracking task for the to-be-tracked event between the current area and the next area. According to the invention, through cooperation of the fixed camera and the patrol robot, cross-region seamless continuous tracking of the mobile stalls is realized, and the resource utilization efficiency is improved.

Description

Technical Field

[0001] This application belongs to the field of artificial intelligence technology, and in particular relates to a method, device, electronic device and storage medium for tracking mobile stalls. Background Technology

[0002] With the acceleration of urbanization, mobile stalls, such as food carts, street vendors, and temporary sales points, have become a major challenge in urban governance due to their high mobility, random distribution, and difficulty in continuous monitoring. Traditional fixed camera monitoring methods have blind spots and cannot achieve continuous tracking of mobile stalls across different areas. Patrol robots are mobile, but since they mostly follow preset routes, continuous tracking of even a single mobile stall requires significant resources. Therefore, there is an urgent need for a new method to track mobile stalls, improving the continuity and efficiency of tracking. Summary of the Invention

[0003] This application provides a method for tracking mobile stalls, which solves the problems of blind spots in existing fixed camera monitoring methods, making it difficult to continuously track mobile stalls across areas, and the high resource investment required for continuous tracking of mobile stalls across areas using patrol robots. By coordinating fixed cameras and patrol robots, and dynamically adjusting the tracking strategy based on the behavioral characteristics of mobile stalls and the probability of events occurring between areas, seamless and continuous tracking of mobile stalls across areas can be achieved, and the scheduling mechanism of patrol robots can be optimized to improve resource utilization efficiency.

[0004] In a first aspect, embodiments of this application provide a method for tracking mobile stalls, the method comprising the following steps:

[0005] When a fixed camera in the current area detects a mobile stall to be tracked, determine the duration of the event and the direction of movement of the mobile stall to be tracked;

[0006] Based on the duration of the event and the direction of movement, determine the next area to which the mobile stall to be tracked is going;

[0007] If the untrackable distance between the next area and the current area is greater than the preset distance, then the patrol robot in the predetermined area is searched; wherein, the preset distance is dynamically set according to the probability of events occurring between the current area and the next area;

[0008] When the probability of an event being detected by the patrol robot is less than the preset probability of an event being detected, the patrol robot is scheduled to perform a tracking task for the event to be tracked between the current area and the next area.

[0009] Optionally, determining the event persistence of the mobile stall to be tracked includes:

[0010] The system obtains the dwell time of the mobile stall to be tracked in the current area, the time interval since the last move, and the crowd density around the stall.

[0011] The duration of the event is calculated based on the dwell time, the time interval, and the crowd density.

[0012] Optionally, determining the next area to which the mobile stall to be tracked is going based on the duration of the event and the direction of movement includes:

[0013] Based on the duration of the event, predict the time it will take for the tracked mobile stall to move to the next area;

[0014] Based on the direction of movement, determine the direction in which the mobile stall to be tracked will move to the next area.

[0015] Optionally, the preset distance is dynamically set based on the probability of events occurring between the current area and the next area, including:

[0016] Obtain the historical frequency of mobile stalls along the path between the current area and the next area, and determine the probability of events occurring between the current area and the next area based on the historical frequency of mobile stalls.

[0017] Based on the preset mapping relationship between the probability of an event and the distance value, the probability of an event is converted into a corresponding preset distance value. The probability of an event and the preset distance value are negatively correlated, so that the preset distance between the paths of regions with higher event probabilities is smaller.

[0018] Optionally, the step of locating patrol robots within a predetermined area if the untrackable distance between the next area and the current area is greater than a preset distance includes:

[0019] Obtain the first coverage boundary of the fixed camera in the current area and the second coverage boundary of the fixed camera in the next area;

[0020] Calculate the spatial straight-line distance or path distance between the first coverage boundary and the second coverage boundary to obtain the untrackable distance;

[0021] Compare the untrackable distance with the preset distance;

[0022] When the untraceable distance is greater than the preset distance, the planned area is searched with the connection path between the current area and the next area as the center line and the preset search radius as the predetermined area.

[0023] Locate the patrol robots within the designated area.

[0024] Optionally, the process of locating patrol robots within a predetermined area includes:

[0025] Obtain the current location, current task status, and remaining battery power of all patrol robots within the predetermined area;

[0026] Based on the current location, current task status, and remaining battery power, at least one robot is selected.

[0027] Optionally, before scheduling the patrol robot to perform a tracking task for the event to be tracked between the current area and the next area when the event detection probability of the patrol robot is less than a preset event detection probability, the method further includes:

[0028] Obtain the type of task currently being performed by the patrol robot;

[0029] Based on the task type, search for historical event data of the task type within the target time period in the area where the patrol robot is located;

[0030] Based on the historical event discovery data, the event discovery probability of the patrol robot is determined.

[0031] Secondly, embodiments of this application provide a tracking device for mobile stalls, the tracking device for mobile stalls comprising:

[0032] The event detection module is used to determine the duration of the event and the direction of movement of the mobile stall being tracked when the fixed camera in the current area detects the mobile stall being tracked.

[0033] The first processing module is used to determine the next area to which the mobile stall to be tracked is going based on the duration of the event and the direction of movement.

[0034] The second processing module is used to locate the patrol robot within the predetermined area if the untrackable distance between the next area and the current area is greater than a preset distance; wherein, the preset distance is dynamically set according to the probability of events occurring between the current area and the next area.

[0035] The scheduling and tracking module is used to schedule the patrol robot to perform a tracking task for the event to be tracked between the current area and the next area when the event detection probability of the patrol robot is less than the preset event detection probability.

[0036] Thirdly, embodiments of the present invention provide an electronic device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps in the mobile stall tracking method provided in embodiments of the present invention.

[0037] Fourthly, embodiments of the present invention provide a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps in the mobile stall tracking method provided in the embodiments of the present invention.

[0038] The above-described solution of this application has the following beneficial effects: When a fixed camera in the current area detects a mobile stall to be tracked, the persistence of the event and the direction of movement of the mobile stall are determined; based on the persistence of the event and the direction of movement, the next area to which the mobile stall is going is determined; if the untrackable distance between the next area and the current area is greater than a preset distance, a patrol robot in the predetermined area is searched; wherein, the preset distance is dynamically set according to the probability of event occurrence between the current area and the next area; when the probability of event detection by the patrol robot is less than the preset probability of event detection, the patrol robot is scheduled to perform the tracking task for the event to be tracked between the current area and the next area. This invention achieves seamless and continuous tracking of mobile stalls across areas by coordinating the fixed camera and the patrol robot, combining the behavioral characteristics of the mobile stall and the probability of event occurrence between areas to dynamically adjust the tracking strategy, and optimizes the scheduling mechanism of the patrol robot to improve resource utilization efficiency.

[0039] Other beneficial effects of this application will be described in detail in the following detailed description section. Attached Figure Description

[0040] To more clearly illustrate the technical solutions in the embodiments of this application, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0041] Figure 1 A flowchart illustrating a method for tracking mobile stalls according to an embodiment of this application;

[0042] Figure 2 This is a schematic diagram of the structure of a tracking device for a mobile stall provided in one embodiment of this application;

[0043] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0044] In the following description, specific details such as particular system architectures and techniques are set forth for illustrative purposes and not for limitation, in order to provide a thorough understanding of the embodiments of this application. However, those skilled in the art will understand that this application may also be implemented in other embodiments without these specific details. In other instances, detailed descriptions of well-known systems, apparatuses, circuits, and methods have been omitted so as not to obscure the description of this application with unnecessary detail.

[0045] It should be understood that, when used in this application specification and the appended claims, the term "comprising" indicates the presence of the described features, integrals, steps, operations, elements and / or components, but does not exclude the presence or addition of one or more other features, integrals, steps, operations, elements, components and / or a collection thereof.

[0046] It should also be understood that the term “and / or” as used in this application specification and the appended claims means any combination of one or more of the associated listed items and all possible combinations, and includes such combinations.

[0047] As used in this application specification and the appended claims, the term "if" may be interpreted, depending on the context, as "when," "once," "in response to determination," or "in response to detection." Similarly, the phrase "if determined" or "if detected [the described condition or event]" may be interpreted, depending on the context, as meaning "once determined," "in response to determination," "once detected [the described condition or event]," or "in response to detection [the described condition or event]."

[0048] Furthermore, in the description of this application and the appended claims, the terms "first," "second," "third," etc., are used only to distinguish descriptions and should not be construed as indicating or implying relative importance.

[0049] References to "one embodiment" or "some embodiments" as described in this specification mean that one or more embodiments of this application include a specific feature, structure, or characteristic described in connection with that embodiment. Therefore, the phrases "in one embodiment," "in some embodiments," "in other embodiments," "in still other embodiments," etc., appearing in different parts of this specification do not necessarily refer to the same embodiment, but rather mean "one or more, but not all, embodiments," unless otherwise specifically emphasized. The terms "comprising," "including," "having," and variations thereof mean "including but not limited to," unless otherwise specifically emphasized.

[0050] like Figure 1 As shown, Figure 1This is a flowchart of a mobile stall tracking method provided by an embodiment of the present invention. The mobile stall tracking method includes the following steps:

[0051] 101. When a fixed camera in the current area detects a mobile stall to be tracked, determine the duration of the event and the direction of movement of the mobile stall to be tracked.

[0052] In this embodiment of the invention, a fixed camera refers to a fixed monitoring device deployed in areas such as city streets, parks, and communities, which has a stable monitoring field of view and coverage. Each fixed camera corresponds to a monitoring area, and the monitoring areas of one or more fixed cameras can be considered as a target area.

[0053] Event detection algorithms can be used to detect events in video streams within a monitored area. The target of event detection is mobile stalls. Event detection algorithms can be object detection models based on deep learning, such as YOLO and SSD.

[0054] Mobile stalls to be tracked refer to mobile stalls detected by fixed cameras that require continuous tracking. These can include food carts, street vendors, temporary sales points, etc. It should be noted that mobile stalls are characterized by mobility, temporariness, and clustering. Fixed stalls, on the other hand, will be marked or skipped and will not be tracked further.

[0055] Event persistence is used to characterize the stability or management importance of a tracked mobile stall within its current area. Higher event persistence indicates that the stall is more worthy of resource investment for continuous tracking. Event persistence can be determined based on factors such as the stall's dwell time in the current area, its movement frequency, and the density of people in the surrounding area. Specifically, longer dwell time, lower movement frequency (i.e., longer intervals between movements), and higher surrounding population density result in higher event persistence.

[0056] The direction of movement refers to the movement trend of the mobile stall being tracked within the current field of view of the fixed camera, and is usually represented by a direction vector. The direction of movement can be determined by analyzing the changes in the stall's position in consecutive video frames, and is used to indicate the stall's tendency to move away from the current area.

[0057] Specifically, when any fixed camera detects a mobile stall to be tracked, the tracking process is initiated. The fixed camera analyzes the real-time video stream using a built-in target detection algorithm, identifies the mobile stall target, and marks it as the mobile stall to be tracked.

[0058] After detecting the mobile stalls to be tracked, the system further extracts the stall's dwell time, the time interval since its last move, the crowd density, and the direction of movement.

[0059] The duration of continuous appearance of a mobile stall within the field of view of a fixed camera is recorded using a target tracking algorithm; this duration is taken as the dwell time. A longer dwell time indicates more stable business operations for the stall in the current area.

[0060] The system records the movement history of the mobile stalls to be tracked and calculates the time difference between the current moment and the last time a stall movement was detected. This time difference is used as the interval between the last movement and the current movement. The shorter the interval, the more frequently the stall moves and the higher its mobility.

[0061] The number of people within a preset radius (e.g., 3 meters) around the stall is counted using a personnel detection algorithm, and then normalized to obtain the crowd density. A higher crowd density indicates a greater impact of the stall on its surrounding environment and thus higher management importance.

[0062] The event persistence of the mobile stalls to be tracked can be calculated based on the dwell time, the time interval since the last move, and the crowd density.

[0063] The duration of an event can be calculated using a weighted summation method: normalize the dwell time, the reciprocal of the time interval, and the crowd density, multiply each by its corresponding weighting coefficient, and then sum the products to obtain the event duration value. The weights for dwell time, the reciprocal of the time interval, and crowd density can be configured according to actual management needs; for example, they can be set to 0.4, 0.3, and 0.3, respectively.

[0064] The event persistence value can range from 0 to 1. The higher the value, the more stable the stall is in the current area, the higher its management importance, and the more worthwhile it is to invest resources in continuous tracking.

[0065] The movement trend of the stall to be tracked is determined by optical flow or trajectory analysis, generating a movement direction vector. This movement direction vector can predict the stall's path after leaving the current area.

[0066] 102. Based on the duration of the event and the direction of movement, determine the next area to which the mobile stall to be tracked is going.

[0067] In this embodiment of the invention, the next area is the area of ​​coverage of an adjacent fixed camera that the mobile stall to be tracked may enter after leaving the coverage area of ​​the current fixed camera, based on its direction of movement. The determination of the next area depends on a comprehensive judgment of the event duration and the direction of movement, where the event duration affects the prediction of the time it takes for the stall to leave the current area, and the direction of movement determines the location of the area the stall is heading to.

[0068] The time it takes for a tracked mobile stall to move to the next area can be predicted based on the event's persistence. Specifically, the historical average dwell time of the tracked mobile stall in the current area can be obtained (e.g., calculated based on the stall's historical behavior data), and then the event persistence can be used to adjust this historical average dwell time. The higher the event persistence, the longer the adjusted predicted dwell time, meaning the tracked mobile stall will stay in the current area for a longer period. Adding the current time to the predicted dwell time gives the predicted time for the stall to leave the current area.

[0069] The direction of the mobile stall to be tracked to the next area can be determined based on its movement direction. Specifically, the entry direction and movement trajectory of the mobile stall to be tracked in the current area are obtained. Prediction algorithms such as Kalman filtering or particle filtering are used to smooth and predict the movement direction vector of the mobile stall to be tracked, resulting in a stable movement direction vector. Then, based on the area pointed to by this vector, combined with the area division information in the geographic information system, the corresponding coverage area of ​​the adjacent fixed camera is determined as the next area.

[0070] 103. If the untrackable distance between the next area and the current area is greater than the preset distance, then search for patrol robots within the predetermined area.

[0071] In this embodiment of the invention, the untrackable distance can be understood as the spatial gap between the coverage boundary of the current area's fixed camera and the coverage boundary of the next area's fixed camera. This distance can be the straight-line distance between the two coverage boundaries or the path distance along an actual road. The untrackable distance represents the size of the blind spot in the fixed camera monitoring network. In this embodiment, there is no coverage of fixed cameras within this blind spot, and patrol robots can be used for continuous tracking.

[0072] The preset distance is dynamically set based on the probability of events occurring between the current area and the next area. It is used to determine whether to dispatch patrol robots for continued tracking. It should be noted that a higher event probability indicates a higher frequency of mobile stalls appearing along the path between the current and next areas, and a greater risk of these stalls operating in blind spots between fixed cameras. Therefore, more aggressive robot tracking is required, resulting in a smaller preset distance. Conversely, a lower event probability results in a larger preset distance, meaning robots are only dispatched when the blind spot distance is significant. This mapping relationship is designed to be negatively correlated with the event probability and the preset distance value; that is, the higher the event probability of a path between areas, the smaller the corresponding preset distance value; and the lower the event probability of a path, the larger the corresponding preset distance value.

[0073] The probability of an event can be calculated by comparing the number of mobile stalls detected along the paths within the region during a historical period with the total number of days counted.

[0074] Alternatively, pedestrian density data along the path between the current area and the next area can be obtained, and the probability of an event can be calculated based on the mapping relationship between pedestrian density and the probability of stalls appearing. This method reflects real-time dynamic risk. Specifically, pedestrian density data can be obtained by statistically analyzing real-time video streams collected by fixed cameras using personnel detection algorithms; alternatively, it can be obtained by counting the number of mobile devices passing through the path based on mobile phone signaling data within the area; or it can be obtained through other sensor data that can acquire pedestrian flow data. The obtained pedestrian density values ​​are matched with a preset pedestrian-stall probability mapping table to obtain the corresponding event probability. The higher the pedestrian density, the more frequent the personnel activity along the path between areas, the greater the possibility of mobile stalls staying and operating, and therefore the higher the probability of the event of mobile stalls staying and operating.

[0075] Alternatively, the probability value based on historical stall frequency can be weighted and fused with the probability value based on real-time pedestrian flow statistics to obtain the overall probability of the event. For example, the overall probability = α × historical probability + β × real-time probability, where α and β are preset weighting coefficients that can be adjusted according to actual management needs.

[0076] The aforementioned predetermined area refers to the spatial range used to search for dispatchable patrol robots. The predetermined area is typically centered on the path connecting the current area and the next area, and determined according to a preset search radius. The size of the predetermined area can be configured according to actual scenario requirements, for example, as a circular area with a radius of 300 meters, or a strip-shaped area extending along the connecting path. This predetermined area constitutes the patrol robot's mission execution area.

[0077] 104. When the event detection probability of the patrol robot is less than the preset event detection probability, the patrol robot is scheduled to perform the tracking task for the event to be tracked between the current area and the next area.

[0078] In this embodiment of the invention, the event detection probability can be understood as the success rate of event detection within the area to which the patrol robot belongs during a target time period. The event detection probability describes the impact of the patrol robot being highly targeted for tracking a particular stall within a target time period on its original task in that area. A lower event detection probability results in a smaller impact on the original task, while a higher probability results in a greater impact. For example, if, based on the event detection probability, the patrol robot's original task in its area was to detect littering incidents and it could detect 10 incidents, then after being reassigned to track mobile stalls within the target time period, 10 littering incidents would not be detected within that time period. Conversely, if the patrol robot's original task was to detect littering incidents and it could detect 1 incident, then after being reassigned to track mobile stalls within the target time period, 1 littering incident would not be detected within that time period. The target time period is the time between the point when the mobile stall leaves the current area and the predicted point when it enters the next area.

[0079] The probability of an event being detected is calculated by: statistically analyzing the historical event detection data of the patrol robot within its patrol area during the target time period, calculating the average number of events detected per time unit (e.g., hourly, daily), or calculating the expected number of events to be detected within the target time period. The probability of an event being detected can be an absolute value (e.g., expected to detect 10 events) or a relative value (e.g., the proportion relative to the historical average).

[0080] The lower the probability of event detection, the fewer events the patrol robot could have detected within the target time period. The less negative impact dispatching the robot to perform tracking tasks would have on its original task, making the robot more suitable for dispatch.

[0081] The higher the event detection probability, the more events the patrol robot could have detected within the target time period. Dispatching the robot to perform tracking tasks will have a greater negative impact on its original task, making the robot less suitable for dispatching.

[0082] The acquired event discovery probability is compared with a preset event discovery probability threshold. The event discovery probability threshold is set so that when the event discovery probability is below the threshold, it indicates that the robot has low value in the original task, scheduling it will have little impact on the original task, and it can be scheduled; when the event discovery probability is above the threshold, it indicates that the robot has high value in the original task, scheduling it will have a significant impact on the original task, and it is not advisable to schedule it.

[0083] In one specific implementation, the event detection probability threshold can be dynamically adjusted based on historical data. For example, when the overall demand for event detection is high within a target time period, the threshold can be increased to schedule only robots with a lower event detection probability; when the demand is low, the threshold can be decreased to allow more robots to be scheduled.

[0084] If the event detection probability of a candidate robot is less than the preset event detection probability threshold, then the robot can be scheduled to perform the tracking task; if the event detection probability is not less than the preset threshold, then other candidate robots can be selected or the exception handling process can be entered.

[0085] After determining the scheduling, the predicted movement path of the mobile stall to be tracked from the current area to the next area can be determined. This predicted movement path is generated based on the stall's movement direction vector, event persistence (affecting movement speed), and geographic road information. Then, the system generates a tracking route for the patrol robot based on this predicted movement path. The tracking route includes: moving from the patrol robot's current position to a predicted intersection point on the predicted movement path, and following the mobile stall to be tracked along the predicted movement path to the coverage boundary of the next area within the target time period.

[0086] Finally, a dispatch command and the tracking characteristics of the mobile stalls to be tracked are sent to the selected patrol robot, thereby controlling the patrol robot to pause its current patrol task and perform a tracking task according to the tracking route.

[0087] When a fixed camera in the current area detects a mobile stall to be tracked, the persistence of the event and the direction of movement of the mobile stall are determined. Based on the persistence of the event and the direction of movement, the next area to which the mobile stall is going is determined. If the untrackable distance between the next area and the current area is greater than a preset distance, a patrol robot in the predetermined area is located. The preset distance is dynamically set according to the probability of event occurrence between the current area and the next area. When the probability of event detection by the patrol robot is less than the preset probability of event detection, the patrol robot is dispatched to perform the tracking task for the event to be tracked between the current area and the next area. This invention achieves seamless and continuous tracking of mobile stalls across areas by coordinating fixed cameras and patrol robots, combining the behavioral characteristics of mobile stalls and the probability of event occurrence between areas to dynamically adjust the tracking strategy, and optimizes the dispatching mechanism of patrol robots to improve resource utilization efficiency.

[0088] It is understood that in the specific implementation of this application, data related to behavioral data, graph data, content data, user data, etc. are involved. When the embodiments in this application are applied to specific products or technologies, user permission or consent is required. Furthermore, the collection, use and processing of related data, as well as the training, deployment and invocation of algorithm models, must comply with the relevant laws, regulations and standards of the relevant countries and regions.

[0089] Optionally, in the step of determining the event persistence of the mobile stall to be tracked, the dwell time of the mobile stall in the current area, the time interval since the last move, and the crowd density around the stall can be obtained; the event persistence can be calculated based on the dwell time, the time interval, and the crowd density.

[0090] In this embodiment of the invention, event persistence is a comprehensive indicator used to quantitatively characterize the stability and management importance of the business activities of a tracked mobile stall within the current monitoring area. A higher event persistence value indicates that the stall remains stable in the current area and has a greater impact on the surrounding environment; correspondingly, it deserves higher priority for continuous monitoring with more tracking resources. It should be noted that event persistence is not a fixed static value, but rather dynamically calculated based on the stall's real-time behavioral characteristics.

[0091] Dwell time refers to the duration for which a mobile stall remains continuously within the field of view of a fixed camera. Once the fixed camera identifies and locks onto a mobile stall using a target detection algorithm, a tracking record is created for that target, and a timer is started. The target tracking algorithm continuously performs inter-frame correlation matching on the stall, and the timer continues to accumulate as long as the target does not leave the current camera's coverage area. Once the target completely moves out of the field of view, the timer is reset to zero. A longer dwell time indicates more stable business operations at the stall, and a longer duration of its impact on the area's order.

[0092] The time interval since the last movement reflects the frequency of movement of the tracked mobile stall, i.e., the stall's mobility characteristics. The system continuously records the stall's movement status. When the stall is stationary, it records the start time of the stationary state; when the stall moves, the system records the start time of the movement. The time interval since the last movement is the time difference between the current moment and the most recent detected moment when the stall changed from stationary to moving. If the stall has not moved for a long time, this time interval is relatively large, indicating that the stall is currently in a relatively stable operating state; conversely, if the stall moves frequently, this time interval is relatively small, indicating that the stall is highly mobile and may leave the current area at any time.

[0093] The crowd density around a mobile stall is used to assess its actual impact on public order. Video streams from fixed cameras can be used, employing pedestrian detection algorithms (such as pedestrian detection models based on YOLOv8 or CenterNet) to count the number of people within a pre-defined area around the stall (e.g., a circular area with a radius of 3 meters centered on the stall's geometric center). The counted number of people is then normalized (e.g., divided by the historical maximum number of observed people in that area) and converted into a density value between 0 and 1. Higher crowd density indicates a greater impact of the stall on pedestrian traffic, traffic safety, or environmental hygiene, and consequently, its management importance increases accordingly.

[0094] In one possible implementation, a weighted summation method can be used to calculate event persistence. Specifically, the dwell time, the reciprocal of the time interval since the last move, and the crowd density are first normalized to map them to a uniform numerical range (e.g., between 0 and 1) to eliminate the influence of different units of measurement. For the time interval since the last move, since this value is positively correlated with stability, its reciprocal is taken before normalization to ensure consistency with the changing trend of event persistence.

[0095] Subsequently, preset weight coefficients are assigned to the three normalized parameters, and the weighted results are summed to obtain a comprehensive value for the event persistence. This calculation process can be expressed as:

[0096]

[0097] in, Indicates the duration of an event; This refers to the normalized dwell time. It is the reciprocal of the time interval between the last movement and the normalized distance; The normalized population density; , , These are the corresponding weighting coefficients, and their specific values ​​can be configured and dynamically adjusted according to the actual management scenario. For example, in areas where it is necessary to focus on rectifying long-term street vending, the following can be used: Set it higher; in areas where temporary gatherings of vendors need to be addressed, it can be... The setting is relatively high. In a typical configuration example, , , The values ​​can be 0.4, 0.3, and 0.3 respectively, so that the contribution of dwell time to persistence is slightly higher than that of the other two factors.

[0098] The event duration value calculated using the above method The value ranges from 0 to 1. The higher the value, the more stable the operation of the mobile stall to be tracked in the current area, the greater the impact on the surrounding environment, and the stronger the business intention of the mobile stall to be tracked. It may carry out business activities anytime and anywhere. Accordingly, the necessity of allocating resources to track it across regions in subsequent steps is also higher.

[0099] It is understood that the weighted summation described above is only one specific method for calculating event persistence. In other embodiments of the present invention, nonlinear fusion, fuzzy logic reasoning, or machine learning models (such as logistic regression, decision trees, etc.) can be used to replace the weighted summation method, outputting a comprehensive event persistence assessment result based on the dwell time, the time interval since the last move, and the crowd density. These alternative methods also fall within the scope of protection of the present invention, as long as their core lies in using the above three dimensions of basic parameters to quantitatively characterize the behavioral features of the stalls.

[0100] In this embodiment of the invention, the behavioral characteristics and importance of the mobile stall to be tracked can be quickly and accurately assessed in the initial stage when the fixed camera detects the mobile stall to be tracked.

[0101] Optionally, in the step of determining the next area to which the mobile stall to be tracked is going based on the duration of the event and the direction of movement, the time when the mobile stall to be tracked will go to the next area can be predicted based on the duration of the event; and the direction of the mobile stall to go to the next area can be determined based on the direction of movement.

[0102] In this embodiment of the invention, the estimated time point at which the mobile stall to be tracked is expected to leave the current area is determined, i.e., the moment the stall disappears from the field of view of the current fixed camera. The accuracy of this time prediction directly affects the rationality of subsequent scheduling timing judgments.

[0103] Obtain the baseline dwell time for the mobile stalls to be tracked. The baseline dwell time can be a general parameter obtained based on historical statistical data, such as the average dwell time of all mobile stalls in the area; or it can be a personalized parameter obtained based on the historical behavior data of the stall itself. For example, if the stall has been detected by the same fixed camera multiple times, its dwell time each time can be recorded and a weighted average can be calculated as the baseline dwell time.

[0104] The baseline dwell time can be adjusted using event persistence. Since event persistence comprehensively reflects the current behavioral characteristics of the stall, such as dwell time, movement frequency, and crowd gathering, using it as a correction factor allows the time prediction to better reflect the actual state of the stall. Specifically, the calculation method for predicted dwell time can be expressed as:

[0105]

[0106] in, The predicted dwell time is the length of time the stall will remain in the current area. The baseline stay time; For event persistence The correction function, which is related to They are positively correlated, that is The higher the value, the larger the correction coefficient, and the longer the predicted dwell time; The lower the value, the smaller the correction coefficient, and the shorter the predicted dwell time. In one possible embodiment, the correction function can be set to a linear relationship, for example... ,in This is the preset adjustment coefficient.

[0107] The current time can be added to the predicted dwell time to obtain the predicted departure time of the stall from the current area. This predicted departure time is used to determine whether a patrol robot needs to be dispatched and to determine the dispatch window for the patrol robot, which includes the target time period.

[0108] It is understood that the above-described method based on baseline dwell time and correction function is only one specific implementation. In other embodiments of the present invention, machine learning models (such as time series prediction models) can also be used to directly predict the departure time end-to-end based on event persistence and other real-time features (such as time period, weather, surrounding events, etc.).

[0109] Determine the possible spatial location of the mobile stalls to be tracked after leaving the current area, that is, predict which adjacent area their movement trend will point to.

[0110] The movement direction vector of the mobile stall to be tracked is obtained. This direction vector can be obtained by using continuous video frames captured by a fixed camera and employing a target tracking algorithm (such as optical flow, Kalman filter tracker, or deep learning-based tracking model) to extract the continuous position coordinates of the stall in the image coordinate system, forming a motion trajectory. By smoothing the motion trajectory (e.g., using moving average filtering to remove jitter noise) and calculating the directional change trend at the end of the trajectory, a stable movement direction vector is obtained.

[0111] To improve prediction accuracy, the entry direction of the stall can be used for comprehensive judgment. Specifically, the initial direction of the stall entering the current area can be recorded and weighted together with the current direction of movement to account for possible turning behavior of the stall within the area. For example, if the stall's entry direction is basically consistent with the current direction of movement, it indicates that the stall has maintained its original trend, and the prediction confidence is high; if there is a large deviation between the two, it indicates that the stall may have stopped or turned within the area, and the prediction confidence needs to be reduced.

[0112] After obtaining a stable movement direction vector, this vector can be matched with a pre-defined regional topology map. The regional topology map is a pre-constructed data structure that records the adjacency relationships, relative orientations, and connectivity information between various monitored areas. Specifically, the movement direction vector can be mapped to a geographic coordinate system, the azimuth angle pointed to by the movement vector can be calculated, and then adjacent areas matching the aforementioned azimuth angle can be retrieved from the regional topology map. These adjacent areas are then designated as the next area.

[0113] In a preferred embodiment of the present invention, when the movement direction vector points to the boundary between two adjacent areas (i.e., the direction is ambiguous), the event persistence can be further considered for auxiliary judgment. Specifically, if the event persistence is high (indicating a strong intention to operate the stall), a higher-risk adjacent area can be selected as the next area; if the event persistence is low (the stall may leave quickly), the area that best matches the current movement direction can be selected as the next area. This improves the robustness of the next area prediction.

[0114] After simultaneously obtaining the predicted time and spatial direction of the mobile stall to be tracked leaving the current area, the time and spatial predictions are combined to generate complete cross-regional prediction information, including: estimated departure time, estimated time of entry into the next area, and the target's next area identifier. This allows for precise judgments on whether to dispatch patrol robots at the appropriate time and for the appropriate adjacent areas.

[0115] In one possible embodiment, during the process of determining the next region based on the movement direction, when the movement direction vector points to the boundary region between two adjacent regions, a situation arises where the direction is ambiguous and the next region cannot be uniquely determined. The system further combines the event persistence to make auxiliary judgments to improve the accuracy and robustness of the prediction.

[0116] Specifically, when the angle between the detected movement direction vector and the boundary lines of two adjacent regions (the first candidate region and the second candidate region) is less than a preset angle threshold (e.g., 15 degrees), the direction fuzzing process can be initiated. At this point, the currently calculated event duration value can be obtained. And take different decision-making strategies based on the magnitude of the value.

[0117] When the event persists Greater than or equal to a preset persistent high threshold (e.g.) When the tracking status is positive, it indicates that the mobile stalls being tracked have a strong intention to operate in the current area. Such stalls tend not to simply leave along the shortest path, but rather tend to move to areas with higher population density and more active business activities to obtain more business opportunities.

[0118] Therefore, when the direction is unclear, if the event is likely to persist, the area with the higher risk level from the first and second candidate areas can be prioritized as the next area for prediction. The risk level classification can be pre-set based on historical data, such as the frequency of mobile stalls appearing in each area historically, the number of complaints, and traffic congestion indices. Areas with higher risk levels typically correspond to sections with high pedestrian traffic and a strong commercial atmosphere.

[0119] When the event persists Less than or equal to a preset persistent low threshold (e.g.) When the time frame is unclear, it indicates that the mobile stall being tracked has a short dwell time in the current area, moves frequently, and has strong mobility. Such stalls may be closing or opening, and are usually in a state of rapid movement. Their movement trajectory has strong inertia, and they will not deliberately change their direction of travel due to business intentions. Therefore, in cases where the direction is unclear, they are more likely to maintain their current inertia and travel along the path that best matches their current direction of movement vector.

[0120] Therefore, when the direction of movement is ambiguous, if the event duration is low, the region with the smaller angle to the current direction of movement vector between the first and second candidate regions is preferentially selected as the next region for prediction. This selection strategy is based on the principle of motion inertia, which states that a moving object tends to maintain its original direction of motion without external intervention.

[0121] For events with duration falling within an intermediate range (e.g., between the low duration threshold and the low duration threshold), the system can employ a weighted fusion approach for decision-making. Specifically, the combined score of the two candidate regions is calculated:

[0122]

[0123] in, Candidate region Similarity to the direction vector of movement (which can be based on the cosine of the angle between the directions). Candidate region Normalized risk level and These are dynamically adjusted weighting coefficients. When the event duration is low, The value is relatively large; when the event duration is high, A larger value indicates a higher overall score. Candidate regions with higher overall scores can be identified as the next region to be predicted.

[0124] By employing the above method, when the direction of movement information is insufficient to uniquely determine the next area, the event persistence behavioral characteristic parameter can be effectively used as a supplement. This ensures that the predicted result for the next area not only conforms to the stall's behavioral patterns but also possesses high robustness. The prediction mechanism, which combines behavioral and motion characteristics, effectively avoids prediction errors caused by directional ambiguity.

[0125] It should be noted that the determination of the next area can be further considered from more factors, such as the current load of fixed cameras in each adjacent area, the historical frequency of stalls appearing in each adjacent area, and the deployment density of patrol robots in each adjacent area. These factors can be integrated with the event's persistence and direction of movement to jointly optimize the prediction results for the next area.

[0126] Optionally, in the step of dynamically setting the preset distance based on the probability of events occurring between the current area and the next area, the historical frequency of mobile stalls appearing on the path between the current area and the next area can be obtained, and the probability of events occurring between the current area and the next area can be determined based on the historical frequency of mobile stalls appearing; according to the preset mapping relationship between the probability of events occurring and the distance value, the probability of events occurring is converted into the corresponding preset distance value, wherein the probability of events occurring and the preset distance value are negatively correlated, so that the preset distance corresponding to the path between areas with higher probability of events occurring is smaller.

[0127] In this invention, a preset distance is used to determine whether the monitoring blind spot between the current area and the next area is worth dispatching a patrol robot for continued tracking. The preset distance is dynamically adjusted based on the event risk along the path between the current area and the next area. When the probability of mobile stalls appearing on the path between the two areas is high, a robot will be dispatched for tracking even if the blind spot distance is small, therefore the preset distance value is small; conversely, when the probability of mobile stalls appearing on the path is low, a robot is only dispatched when the blind spot distance is large, therefore the preset distance value is large.

[0128] The probability of the aforementioned event refers to the likelihood that mobile vendors will appear and operate on the path between the current area and the next area. It indicates the actual risk level of a segment of the path as a blind spot in monitoring. A higher probability of the event indicates that the path between the current area and the next area is more likely to become a gathering or passageway for mobile vendors, and the greater the need for continuous tracking.

[0129] Specifically, path information between the current area and the next area can be obtained. This path refers to the actual connecting roads, passages, or public spaces between the coverage boundaries of the fixed cameras in the two areas. The path information can be obtained through a preset regional topology map, which pre-stores the connecting paths and their geographical coordinates between each pair of adjacent monitoring areas.

[0130] After determining the route, historical detection data of mobile stalls along that route is retrieved from the historical database. The historical database records mobile stall events detected by fixed cameras, patrol robots, or other sensing devices within past time periods. Each historical detection data entry includes information such as the time, specific location, and duration of the event.

[0131] Based on the historical monitoring data mentioned above, the frequency of occurrence of mobile stalls along the aforementioned paths is statistically analyzed. Specifically, this can be done by dividing the number of mobile stall events detected within the statistical period by the total number of days in the statistical period to obtain the average daily frequency; or by dividing the statistical period into several time windows (such as hourly time slots) and statistically analyzing the frequency of occurrence in each time slot to obtain the risk differences between different time slots.

[0132] Based on this, the probability of an event can be determined according to the historical frequency of occurrence of mobile stalls. In one specific embodiment of the invention, the normalized historical frequency of occurrence is directly used as the probability of event occurrence. For example, the historical maximum observed frequency of this path is set to... The frequency obtained from the current statistics is The probability of the event occurring is... The normalized probability value is between 0 and 1, with a higher value indicating a greater likelihood of mobile stalls appearing along the aforementioned path.

[0133] In one possible embodiment, to reflect path risks more in real time, probability correction can also be performed by combining real-time dynamic data. For example, the current real-time pedestrian density data on the path can be obtained, and the real-time probability can be calculated according to the preset pedestrian-stall mapping relationship. Then, the historical probability and the real-time probability can be weighted and fused to obtain the comprehensive event occurrence probability.

[0134] In obtaining the probability of the event occurring Then, based on the preset mapping relationship between the event occurrence probability and the preset distance value, the event occurrence probability can be converted into the corresponding preset distance value. The above mapping relationship is set so that the probability of an event occurring is negatively correlated with the preset distance value; that is, the higher the probability of an event occurring, the smaller the preset distance value; and the lower the probability of an event occurring, the larger the preset distance value.

[0135] In one specific embodiment of the present invention, the mapping relationship is implemented using a linear function:

[0136]

[0137] in, The maximum tolerable blind zone distance preset for the system is a configurable system parameter that represents the maximum blind zone distance the system can tolerate under the safest conditions (probability of event occurrence is 0). For example, It can be set to 200 meters. When the probability of the event occurring... hour, Meters, meaning the robot is only dispatched when the blind zone distance is greater than 200 meters; when the probability of an event occurring... hour, Meters, meaning that as long as there is any blind spot distance (greater than 0), the system will dispatch a robot to track.

[0138] In another specific embodiment of the present invention, in order to provide more flexible control, the mapping relationship can be implemented using a piecewise function:

[0139]

[0140] in, and These are preset high probability thresholds and low probability thresholds, respectively. , , This is the preset distance value for the corresponding gear, and it satisfies... For example, when the probability of an event occurring is greater than 0.7, the preset distance is set to 50 meters; when the probability of an event occurring is between 0.3 and 0.7, the preset distance is set to 100 meters; and when the probability of an event occurring is less than 0.3, the preset distance is set to 200 meters.

[0141] In one possible implementation, the above mapping relationship can also be implemented using nonlinear functions (such as exponential decay functions, sigmoid functions, etc.) to adapt to the differentiated requirements for scheduling sensitivity in different scenarios. For example, in high-risk areas such as core business districts, a mapping function with faster decay can be used, so that a slight increase in the probability of an event can significantly reduce the preset distance, thereby scheduling robots more actively; in general areas, a mapping function with slower decay can be used to maintain a relatively relaxed scheduling strategy.

[0142] Through the aforementioned dynamic settings, the scheduling sensitivity can be adaptively adjusted based on the actual risk level of paths between different areas. On high-risk paths, robots will be actively dispatched for tracking even if the blind zone distance is small, ensuring tracking continuity. On low-risk paths, robots will only be dispatched when the blind zone distance is large, avoiding unnecessary resource consumption. This refined, scenario-adaptive scheduling decision-making mechanism can improve the utilization efficiency of patrol robot resources.

[0143] Optionally, if the untrackable distance between the next area and the current area is greater than a preset distance, the step of finding patrol robots within the predetermined area can involve obtaining the first coverage boundary of the fixed camera in the current area and the second coverage boundary of the fixed camera in the next area; calculating the spatial straight-line distance or path distance between the first coverage boundary and the second coverage boundary to obtain the untrackable distance; comparing the untrackable distance with the preset distance; when the untrackable distance is greater than the preset distance, using the connection path between the current area and the next area as the center line, searching the planned area as the predetermined area with a preset search radius; and finding patrol robots within the predetermined area.

[0144] In this embodiment of the invention, the untrackable distance refers to the spatial gap between the coverage boundary of a fixed camera in the current area and the coverage boundary of a fixed camera in the next area. It describes the size of the blind spot in a fixed camera monitoring network between two adjacent monitoring areas. It should be noted that, because the monitoring range of fixed cameras is usually distributed in a fan shape or rectangle, and is affected by factors such as installation location, lens focal length, and obstructions, there are often gap areas between the coverage areas of two adjacent cameras that cannot be covered by any fixed camera. When a mobile stall enters this gap area, if no other monitoring equipment (such as a patrol robot) continues to track it, the tracking will be interrupted.

[0145] Untrackable distance does not simply refer to the straight-line distance between two cameras, but rather the distance calculated along the actual travel path (such as roads, sidewalks, etc.). This is because the movement of mobile stalls is constrained by the geographical environment, and their actual travel routes are usually along roads or open spaces. Therefore, using path distance rather than Euclidean distance to measure untrackable distance can more accurately reflect the actual length of the tracking blind spot that the patrol robot needs to fill.

[0146] The aforementioned predetermined area refers to the spatial range used to search for available patrol robots when it is determined that patrol robots need to be dispatched for cross-regional tracking. The determination of this spatial range is based on the connection path between the current area and the next area, taking into account the patrol robot's mobility and response time to ensure that the dispatched robot can reach the tracking path and meet the target within a reasonable time. This predetermined area is the patrol area of ​​the patrol robot.

[0147] Specifically, it can obtain the first coverage boundary of the fixed cameras in the current area and the second coverage boundary of the fixed cameras in the next area. The coverage boundary refers to the geographic boundary line of the effective range that the fixed cameras can monitor. The above boundary information can be pre-stored in a geographic information database. Each fixed camera is calibrated during deployment, and its coverage area is recorded in the form of polygons or sectors and mapped to a unified geographic coordinate system.

[0148] Based on the device identifier of the fixed camera in the current area, the corresponding coverage boundary polygon data can be retrieved from the geographic information database; similarly, based on the device identifier of the fixed camera in the next area, the corresponding coverage boundary polygon data can be retrieved. The two coverage boundary polygons represent the effective monitoring range of the two monitoring areas respectively.

[0149] The path distance between the first and second coverage boundaries can be calculated to obtain the untrackable distance. After determining the connecting path between the current area and the next area, a walkable path from the first coverage boundary to the second coverage boundary is identified based on preset road network data (such as urban road network GIS data). The walkable path can be a pedestrian path, a non-motorized vehicle lane, or a motorized vehicle lane, depending on the actual movement mode of the mobile stall. Then, the start and end points of the path are determined. The start point can be selected as the point on the first coverage boundary that is closest to the stall's current movement direction and connects to the next area, and the end point can be selected as the point on the second coverage boundary that is closest to the stall's predicted entry direction and connects to the current area. The shortest path distance along the road network from the start point to the end point is calculated using a path planning algorithm (such as Dijkstra's algorithm or A* algorithm), and this distance value is taken as the untrackable distance. The untraceable distance value mentioned above represents the actual length of the surveillance blind spot that a mobile stall needs to traverse between leaving the coverage area of ​​the current fixed camera and entering the coverage area of ​​the next fixed camera.

[0150] The calculated untrackable distance and preset distance Compare. Preset distance. It is dynamically set based on the probability of events occurring between the current area and the next area. The higher the probability of an event occurring, the smaller the preset distance; the lower the probability of an event occurring, the larger the preset distance. This indicates that the blind spot distance between fixed cameras is within an acceptable range. In this case, even if a stall enters the blind spot, due to its short movement distance or the low probability of mobile stall operations historically occurring in that blind spot, the patrol robot can be temporarily withheld, and tracking can continue once the stall naturally enters the coverage area of ​​the next fixed camera in the next area. If This indicates that the blind spot distance between fixed cameras exceeds an acceptable range. In this situation, without the intervention of mobile monitoring equipment, the stall may be operating within the blind spot and thus cannot be recorded. Therefore, it is necessary to dispatch patrol robots for continuous tracking.

[0151] When the judgment At this point, the patrol robot begins its search process. Specifically, the spatial range of the predetermined area is first determined. Using the connecting path between the current area and the next area as the center line, the space is expanded with a preset search radius to form the predetermined area. The search radius can be configured according to the patrol robot's mobility, response time requirements, and actual scenario needs. For example, in an open street scenario, the search radius can be set to 200 to 500 meters; in a dense commercial area scenario, the search radius can be appropriately reduced to improve search accuracy.

[0152] Specifically, the geometric line of the connecting path is used as the center line, and a preset search radius is extended to both sides along the vertical direction of the center line to form a strip-shaped area. The strip-shaped area covers the connecting path and the space within a certain range around it, ensuring that patrol areas with patrol robots that may be located near the path but not directly above the path can be included in the search range.

[0153] In one possible implementation, the search radius is not a fixed value, but can be dynamically adjusted based on the duration of the event. Specifically, when the event duration is high, it indicates that the stall may remain for a longer period of time, and the timeliness requirement for tracking is relatively high. In this case, a smaller search radius can be used, prioritizing the search for robots closer to the connection path. When the event duration is low, it indicates that the stall may move quickly, requiring a rapid response. In this case, a larger search radius can be used to expand the search area to find available robots.

[0154] After determining the designated area, available patrol robots are located within that area. The search is based on at least one of the following: the patrol robot's current location, current task status, and remaining battery power. Information on at least one of the following for all patrol robots within the designated area is obtained, and a list of candidate robots in a dispatchable state is generated.

[0155] By completing the entire process from obtaining coverage boundaries, calculating untrackable distances, comparing distances, determining predetermined areas, and locating patrol robots, more accurate resource data for patrol robots can be obtained.

[0156] Optionally, in the step of finding patrol robots within a predetermined area, the current location, current task status, and remaining battery power of all patrol robots within the predetermined area can be obtained; based on the current location, current task status, and remaining battery power, at least one selectable robot can be filtered.

[0157] In this embodiment of the invention, the aforementioned current location refers to the real-time coordinate information of the patrol robot in space. The patrol robot is typically equipped with a positioning module (such as GPS, BeiDou Navigation Satellite System, or a LiDAR / vision-based Simultaneous Localization and Mapping (SLAM) system), capable of reporting its precise geographic coordinates to the management platform at a preset frequency (e.g., once per second). The distance between the robot and the predicted path to the tracked stall, as well as the time required for the robot to move to the designated intersection point, can be assessed using the current location information. Robots that are closer in distance and have shorter estimated arrival times are more suitable for scheduling.

[0158] The current task status described above characterizes the type of task the patrol robot is performing and its progress. Task statuses can include idle (not performing any task), patrolling (performing a routine patrol task), tracking (tracking another target), en route to the task location, charging, and maintenance. For robots performing high-priority tasks (such as handling emergencies), the system should avoid scheduling them; for robots performing routine patrol tasks, the system can evaluate based on the task's importance and interruptibility; and for robots in an idle state, they are the highest priority for scheduling.

[0159] During the execution of their tasks, the patrol robots report information such as the current task identifier, task priority, and task progress to the management platform in real time. The management platform maintains a task status table for each robot and updates it when it receives new status changes.

[0160] The remaining battery power mentioned above refers to the current remaining capacity of the patrol robot's battery, usually expressed as a percentage. The robot needs sufficient power to complete its scheduled tracking tasks, including moving from its current location to the predicted meeting point, following a stall to the next area, and potentially returning to the original area or heading to a charging station. Robots with excessively low remaining battery power (e.g., below 20%) are not suitable for scheduling, as they may have to abandon tracking midway through the task due to battery depletion. The power consumption required to perform the tracking task can be estimated based on the robot's historical power consumption data and the current task path length, and compared with the remaining battery power to determine if the robot is capable of performing the task. The patrol robot monitors its battery power in real time through its built-in battery management system and reports it to the management platform at a preset frequency.

[0161] After acquiring the aforementioned status information of all patrol robots within the designated area, the system selects at least one suitable patrol robot to perform the tracking task according to preset filtering rules. Candidate robots can be initially filtered based on their current task status. Specifically, robots in an idle state are directly included in the candidate pool; for robots in a "patrolling" state, it is further determined whether their current patrol task can be interrupted or paused. If the patrol task is of low priority (e.g., routine inspection) and the task progress has not yet reached a critical node, the robot is included in the candidate pool; if the patrol task is of high priority (e.g., emergency response), or the task is about to be completed (e.g., the time remaining is less than a preset threshold), it is not included in the candidate pool. Robots in a tracking, charging, or maintenance state are not included in the candidate pool to avoid resource conflicts or task interruptions.

[0162] The power consumption of candidate robots, after being filtered by task status, can be further assessed. The estimated power consumption for each robot performing the tracking task can be calculated. This includes the power consumption when moving to the junction. Power consumption during the tracking process And the power consumption when returning to the original location or heading to the nearest charging station. If the robot currently has remaining battery power... satisfy ,in If the preset safe battery level (e.g., 10%) is reached, the robot is capable of performing the task and is retained in the candidate pool; otherwise, it is excluded.

[0163] The candidate robots selected through the above screening process can be ranked according to the degree of matching between their current location and the predicted tracking path. Specifically, the shortest distance between each robot and the predicted movement path of the stall to be tracked is calculated. And the estimated time required for the robot to move to the predicted intersection point. The system follows From childhood to adulthood or Candidate robots are sorted in ascending order of their estimated arrival time. Robots that are closer in distance and have shorter estimated arrival times are more suitable for scheduling.

[0164] After completing the above filtering and sorting, the system selects the robot with the highest ranking as the final scheduling target. If there are multiple robots in the candidate pool and their ranking results are similar, multiple robots can also be selected as candidates.

[0165] In one possible embodiment, the selection rules can be configured and adjusted according to the actual application scenario. For example, a weighted scoring mechanism can be introduced, assigning different weights to the current location, task status, and remaining battery power, calculating the comprehensive scheduling suitability score for each robot, and selecting the robot with the highest score. Alternatively, a machine learning model can be used to train a scheduling decision model based on historical scheduling data, automatically outputting the optimal scheduling object.

[0166] Optionally, when the event detection probability of the patrol robot is less than the preset event detection probability, before scheduling the patrol robot to perform the tracking task for the event to be tracked between the current area and the next area, the current task type of the patrol robot can be obtained; based on the task type, historical event detection data of the task type in the area where the patrol robot is located within the target time period can be found; based on the historical event detection data, the event detection probability of the patrol robot can be determined.

[0167] In this embodiment of the invention, the aforementioned event discovery probability is an indicator used to quantify the ability or expected value of a patrol robot to discover a specific type of event within its assigned area and within a specific time period. Its function is to assess the impact of scheduling the robot to perform tracking tasks on its original responsibilities. A lower event discovery probability indicates that the patrol robot could originally discover fewer events within the target time period, the opportunity cost of scheduling it to perform tracking tasks is lower, and the patrol robot is more suitable for scheduling. Conversely, a higher event discovery probability indicates that the patrol robot has higher value in its original task, the greater the impact of scheduling on its original task, and the less suitable it is for scheduling.

[0168] Specifically, the system obtains the current task type being performed by the candidate patrol robot. Generally, patrol robots typically undertake multiple different types of tasks in their daily work, with different task types corresponding to different management objectives and event detection patterns. Task types can include routine patrol tasks, special monitoring tasks, emergency response tasks, and fixed-point duty tasks. During task execution, the patrol robot reports information such as the current task identifier, task category, and task priority to the management platform in real time; the management platform maintains a task status table for each robot and updates it upon receiving new status changes.

[0169] After obtaining the current task type, you can further search for historical event discovery data of the patrol robot within its area for that task type. Historical event discovery data refers to the statistical records of various events (including but not limited to mobile stall events) that the patrol robot successfully discovered and reported during its historical operations when performing the same or similar task types.

[0170] Align historical data with the target time periods for tracking mobile stalls. Specifically, determine the target time periods by predicting when the stall is expected to leave the current area (i.e., the expected start time of the tracking task) and when the stall is expected to enter the next area (i.e., the expected end time of the tracking task). Then, search for time windows in historical data that match the target time period, such as the same hour segment, the same day of the week, the same holiday attributes, etc.

[0171] Align historical data with the robot's current location. Each patrol robot has an associated patrol area, which can be a fixed patrol route or a dynamically divided responsibility grid. Filter the historical data to find records of tasks performed by the robot within its current location, rather than data from other areas.

[0172] Align historical data with the current task type. For example, if the robot is currently performing a routine patrol task, the system will search for event discovery records of the robot performing routine patrol tasks in the same area during the same historical period; if the robot is currently performing a special monitoring task, the system will search for the historical records of special monitoring tasks.

[0173] Through the multi-dimensional alignment and filtering described above, a set of historical sample data most similar to the current state of the patrol robot can be obtained. After obtaining the aligned historical event discovery data, the event discovery probability can be calculated based on this data. Specifically, the average number of events discovered by the patrol robot per hour during the target time period (or the average number of events discovered per task) in the historical sample data can be counted, and this statistical value can be used as the event discovery probability. For example, if historical data shows that the robot discovers an average of 3 mobile stall events per hour when performing regular patrol tasks during weekday morning rush hours (7:00-9:00), then its event discovery probability can be determined as 3. The higher this value, the greater the value of the robot during that time period, and the higher the scheduling opportunity cost. Alternatively, the historical discovery frequency can be compared with the average level of all robots in the area to calculate the relative ratio. For example, if the robot's average discovery frequency during the target time period is 3 events / hour, while the average discovery frequency of all robots in the area is 2 events / hour, then its relative ratio (event discovery probability) is 1.5; if its average discovery frequency is 1 event / hour, the relative ratio is 0.5. The relative ratio can eliminate the influence of differences between regions, making the event detection probability of robots in different regions comparable.

[0174] In one possible implementation, the varying sensitivities of different task types to different event types are considered. For example, a robot performing a specific monitoring task might be more effective at detecting mobile stall events but less effective at detecting other types of events. Therefore, different weights can be assigned to different event types to calculate a weighted event detection probability. Specifically, higher weights are assigned to target event types related to the tracking task (i.e., mobile stall events), while lower weights are assigned to non-target event types. This weighted calculation more accurately reflects the impact of scheduling the robot on its ability to detect target events.

[0175] In another possible implementation, a machine learning model (such as a time series forecasting model or a regression model) can be used to predict the probability of event discovery within a target time period based on historical event discovery data. The machine learning model can comprehensively consider more influencing factors, such as weather conditions, holiday information, and surrounding activity information, to output more accurate predictions. The machine learning model can be trained using algorithms such as linear regression, random forest, gradient boosting tree, or long short-term memory networks, using historical event discovery data as labels and time period, region, task type, weather, and holidays as features to obtain the prediction model.

[0176] The event detection probability value of the patrol robot is finally determined by any of the above calculation methods or a combination thereof. The above probability values It can be the absolute number of discoveries, the relative ratio, or the normalized 0-1 probability value, as long as it can be used to compare with the preset event discovery probability threshold.

[0177] In one possible embodiment, the event discovery probability can be compared with a preset event discovery probability threshold. Comparison. When When the time frame indicates that the patrol robot has a low ability to detect expected events during the target time period, scheduling it will have little impact on the original task, therefore it is determined that the robot is suitable to be scheduled to perform the tracking task; when If the event detection probability of all candidate robots is greater than or equal to the threshold after screening, the anomaly handling process can be initiated. For example, the robot with the lowest event detection probability can be selected for scheduling and a resource shortage alert can be generated, or other robots can be waited for to become available.

[0178] like Figure 2 As shown, an embodiment of the present invention provides a tracking device for mobile stalls, the tracking device for mobile stalls comprising:

[0179] The event detection module 201 is used to determine the event persistence and movement direction of the mobile stall to be tracked when the fixed camera in the current area detects the mobile stall to be tracked.

[0180] The first processing module 202 is used to determine the next area to which the mobile stall to be tracked is going based on the duration of the event and the direction of movement.

[0181] The second processing module 203 is used to locate the patrol robot within a predetermined area if the untrackable distance between the next area and the current area is greater than a preset distance; wherein the preset distance is dynamically set according to the probability of events occurring between the current area and the next area.

[0182] The scheduling and tracking module 204 is used to schedule the patrol robot to perform a tracking task for the event to be tracked between the current area and the next area when the event detection probability of the patrol robot is less than the preset event detection probability.

[0183] Optionally, the event detection module 201 is further configured to obtain the dwell time of the mobile stall to be tracked in the current area, the time interval since the last move, and the crowd density around the stall; and calculate the event persistence based on the dwell time, the time interval, and the crowd density.

[0184] Optionally, the first processing module 202 is further configured to predict the time when the mobile stall to be tracked will go to the next area based on the duration of the event; and determine the direction of the mobile stall to be tracked going to the next area based on the direction of movement.

[0185] Optionally, the second processing module 203 is further configured to obtain the historical frequency of mobile stalls on the path between the current area and the next area, and determine the probability of an event between the current area and the next area based on the historical frequency of mobile stalls; and convert the probability of an event into a corresponding preset distance value according to a preset mapping relationship between the probability of an event and the distance value, wherein the probability of an event and the preset distance value are negatively correlated, so that the preset distance corresponding to the path between areas with a higher probability of an event is smaller.

[0186] Optionally, the second processing module 203 is further configured to obtain the first coverage boundary of the fixed camera in the current area and the second coverage boundary of the fixed camera in the next area; calculate the spatial straight-line distance or path distance between the first coverage boundary and the second coverage boundary to obtain the untrackable distance; compare the untrackable distance with the preset distance; when the untrackable distance is greater than the preset distance, take the connection path between the current area and the next area as the center line and search the planned area as the predetermined area with a preset search radius; and find the patrol robot in the predetermined area.

[0187] Optionally, the second processing module 203 is further configured to obtain the current location, current task status, and remaining battery power of all patrol robots within the predetermined area; and to select at least one robot based on the current location, current task status, and remaining battery power.

[0188] Optionally, the device further includes:

[0189] The acquisition module is used to acquire the type of task currently being performed by the patrol robot;

[0190] The search module is used to search for historical event discovery data of the task type within the area where the patrol robot is located during the target time period, based on the task type.

[0191] The third processing module is used to determine the event detection probability of the patrol robot based on the historical event detection data.

[0192] like Figure 3 As shown, this embodiment of the invention also provides an electronic device, including a processor, which can execute any of the above-described mobile stall tracking methods.

[0193] Specifically, it includes a processor 301 and a memory 302, as well as a computer program stored in the memory 302 and capable of running on the processor 301, which executes a method for tracking mobile stalls, wherein:

[0194] The processor 301 executes the calculator program for the mobile stall tracking method stored in the memory 302, and performs the following steps:

[0195] When a fixed camera in the current area detects a mobile stall to be tracked, determine the duration of the event and the direction of movement of the mobile stall to be tracked;

[0196] Based on the duration of the event and the direction of movement, determine the next area to which the mobile stall to be tracked is going;

[0197] If the untrackable distance between the next area and the current area is greater than the preset distance, then the patrol robot in the predetermined area is searched; wherein, the preset distance is dynamically set according to the probability of events occurring between the current area and the next area;

[0198] When the probability of an event being detected by the patrol robot is less than the preset probability of an event being detected, the patrol robot is scheduled to perform a tracking task for the event to be tracked between the current area and the next area.

[0199] Optionally, the determination of the event persistence of the mobile stall to be tracked, performed by processor 301, includes:

[0200] The system obtains the dwell time of the mobile stall to be tracked in the current area, the time interval since the last move, and the crowd density around the stall.

[0201] The duration of the event is calculated based on the dwell time, the time interval, and the crowd density.

[0202] Optionally, the step of processor 301 determining the next area to which the mobile stall to be tracked is going based on the event persistence and the direction of movement includes:

[0203] Based on the duration of the event, predict the time it will take for the tracked mobile stall to move to the next area;

[0204] Based on the direction of movement, determine the direction in which the mobile stall to be tracked will move to the next area.

[0205] Optionally, the preset distance executed by processor 301 is dynamically set according to the probability of events occurring between the current region and the next region, including:

[0206] Obtain the historical frequency of mobile stalls along the path between the current area and the next area, and determine the probability of events occurring between the current area and the next area based on the historical frequency of mobile stalls.

[0207] Based on the preset mapping relationship between the probability of an event and the distance value, the probability of an event is converted into a corresponding preset distance value. The probability of an event and the preset distance value are negatively correlated, so that the preset distance between the paths of regions with higher event probabilities is smaller.

[0208] Optionally, the step of processor 301 to locate patrol robots within a predetermined area if the untrackable distance between the next area and the current area is greater than a preset distance includes:

[0209] Obtain the first coverage boundary of the fixed camera in the current area and the second coverage boundary of the fixed camera in the next area;

[0210] Calculate the spatial straight-line distance or path distance between the first coverage boundary and the second coverage boundary to obtain the untrackable distance;

[0211] Compare the untrackable distance with the preset distance;

[0212] When the untraceable distance is greater than the preset distance, the planned area is searched with the connection path between the current area and the next area as the center line and the preset search radius as the predetermined area.

[0213] Locate the patrol robots within the designated area.

[0214] Optionally, the process of finding patrol robots within a predetermined area, executed by processor 301, includes:

[0215] Obtain the current location, current task status, and remaining battery power of all patrol robots within the predetermined area;

[0216] Based on the current location, current task status, and remaining battery power, at least one robot is selected.

[0217] Optionally, before scheduling the patrol robot to perform a tracking task for the event to be tracked between the current area and the next area when the event detection probability of the patrol robot is less than the preset event detection probability, the method executed by the processor 301 further includes:

[0218] Obtain the type of task currently being performed by the patrol robot;

[0219] Based on the task type, search for historical event data of the task type within the target time period in the area where the patrol robot is located;

[0220] Based on the historical event discovery data, the event discovery probability of the patrol robot is determined.

[0221] This invention also provides a computer-readable storage medium storing a computer program. When executed by a processor, the computer program implements the various processes of the mobile stall tracking method provided in this invention and achieves the same technical effect. To avoid repetition, it will not be described again here.

[0222] The above description is the preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principles described in this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for tracking mobile stalls, characterized in that, The method includes the following steps: When a fixed camera in the current area detects a mobile stall to be tracked, the event duration and direction of movement of the mobile stall to be tracked are determined. Based on the duration of the event and the direction of movement, determine the next area to which the mobile stall to be tracked is going; If the untrackable distance between the next area and the current area is greater than a preset distance, then a patrol robot within the predetermined area is located; wherein, the preset distance is dynamically set according to the probability of events occurring between the current area and the next area; When the probability of the patrol robot detecting an event is less than the preset probability of detecting an event, the patrol robot is scheduled to perform a tracking task for the event to be tracked between the current area and the next area.

2. The method for tracking mobile stalls as described in claim 1, characterized in that, Determining the event persistence of the mobile stall to be tracked includes: The system obtains the dwell time of the mobile stall to be tracked in the current area, the time interval since the last move, and the crowd density around the stall. The duration of the event is calculated based on the dwell time, the time interval, and the crowd density.

3. The method for tracking mobile stalls as described in claim 1, characterized in that, Determining the next area to which the tracked mobile stall is going based on the duration of the event and the direction of movement includes: Based on the duration of the event, predict the time it will take for the tracked mobile stall to move to the next area; Based on the direction of movement, determine the direction in which the mobile stall to be tracked will move to the next area.

4. The method for tracking mobile stalls as described in claim 1, characterized in that, The preset distance is dynamically set based on the probability of events occurring between the current region and the next region, including: Obtain the historical frequency of mobile stalls along the path between the current area and the next area, and determine the probability of events occurring between the current area and the next area based on the historical frequency of mobile stalls. Based on the preset mapping relationship between the probability of an event occurring and a preset distance value, the probability of an event occurring is converted into a corresponding preset distance value, and the higher the probability of an event occurring, the smaller the corresponding preset distance.

5. The method for tracking mobile stalls as described in claim 4, characterized in that, The step of locating patrol robots within a predetermined area if the untrackable distance between the next area and the current area is greater than a preset distance includes: Obtain the first coverage boundary of the fixed camera in the current area and the second coverage boundary of the fixed camera in the next area; Calculate the path distance between the first coverage boundary and the second coverage boundary to obtain the untraceable distance; Compare the untrackable distance with the preset distance; When the untraceable distance is greater than the preset distance, the planned area is searched with the connection path between the current area and the next area as the center line and the preset search radius as the predetermined area. Locate the patrol robots within the designated area.

6. The method for tracking mobile stalls as described in any one of claims 1 to 5, characterized in that, The process of locating patrol robots within a predetermined area includes: Obtain the current location, current task status, and remaining battery power of all patrol robots within the predetermined area; Based on the current location, current task status, and remaining battery power, at least one robot is selected.

7. The method for tracking mobile stalls as described in any one of claims 1 to 5, characterized in that, Before scheduling the patrol robot to perform a tracking task for the event to be tracked between the current area and the next area when the event detection probability of the patrol robot is less than a preset event detection probability, the method further includes: Obtain the type of task currently being performed by the patrol robot; Based on the task type, search for historical event data of the task type within the target time period in the area where the patrol robot is located; Based on the historical event discovery data, the event discovery probability of the patrol robot is determined.

8. A tracking device for mobile stalls, characterized in that, The tracking device for the mobile stall includes: The event detection module is used to determine the event duration and movement direction of the mobile stall to be tracked when the fixed camera in the current area detects the mobile stall to be tracked. The first processing module is used to determine the next area to which the mobile stall to be tracked is going based on the duration of the event and the direction of movement. The second processing module is used to locate patrol robots within a predetermined area if the untrackable distance between the next area and the current area is greater than a preset distance; wherein the preset distance is dynamically set according to the probability of events occurring between the current area and the next area. The scheduling and tracking module is used to schedule the patrol robot to perform a tracking task for the event to be tracked between the current area and the next area when the event detection probability of the patrol robot is less than the preset event detection probability.

9. An electronic device, characterized in that, include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the computer program, implements the steps of the method for tracking mobile stalls as claimed in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method for tracking mobile stalls as described in any one of claims 1 to 7.

Images