Patrol artificial intelligence model-based patrol method, system, equipment and medium

Abstract

The invention relates to a patrol artificial intelligence model-based patrol method, system, device and medium, and the method comprises the steps: receiving the correlation information of a first abnormal event sent by at least one unmanned patrol car, the first abnormal event being recognized by a vehicle-mounted patrol artificial intelligence model of the unmanned patrol car; determining associated information of an edge abnormal event according to the associated information of the first abnormal event, wherein the edge abnormal event comprises an event which is encountered in the patrol process of the unmanned patrol vehicle and cannot be accurately recognized by the patrol artificial intelligence model; a patrol artificial intelligence model of a cloud server is optimized according to the associated information of the edge abnormal event, and an optimized patrol artificial intelligence model is obtained; and upgrading the vehicle-mounted patrol artificial intelligence model of the at least one unmanned patrol vehicle by using the optimized patrol artificial intelligence model. According to the invention, the adaptability of the patrol artificial intelligence model to the patrol environment and various types of abnormal events is improved.

Description

Technical Field

[0001] This disclosure relates to the field of automated patrol technology, and in particular to a patrol method, system, device, and medium based on a patrol artificial intelligence model. Background Technology

[0002] With the increasing popularity of autonomous vehicles, more and more autonomous vehicles are being used in security scenarios in closed areas (such as airports, logistics parks, ports, nuclear power plants, and industrial zones).

[0003] Currently, in applications where security patrols are conducted using autonomous vehicles (referred to herein as unmanned patrol vehicles), these vehicles collect information about the surrounding environment and then use artificial intelligence models to identify abnormal events based on the collected data. However, this patrol method has at least the following problems: once the artificial intelligence model is deployed in the autonomous vehicle, it becomes fixed and cannot self-adjust according to the actual patrol environment. When faced with new types of abnormal events or significant changes in the patrol environment, the accuracy of abnormal event identification decreases significantly.

[0004] In view of the above, this application is hereby submitted. Summary of the Invention

[0005] To address or at least partially address the aforementioned technical problems, this disclosure provides a patrol method, system, device, and medium based on a patrol artificial intelligence model. This enables the automatic collection of edge anomalies and the continuous optimization of the patrol artificial intelligence model based on these events. This significantly improves the patrol artificial intelligence model's adaptability to the patrol environment and various types of anomalies, thereby enhancing patrol effectiveness.

[0006] In a first aspect, embodiments of this disclosure provide a patrol method based on a patrol artificial intelligence model, applied to a cloud server, the method comprising:

[0007] Receive association information of a first abnormal event from at least one unmanned patrol vehicle, wherein the first abnormal event is identified by the patrol artificial intelligence model onboard the unmanned patrol vehicle;

[0008] The association information of edge anomalies is determined based on the association information of the first anomaly event. The edge anomalies include events encountered by the unmanned patrol vehicle during patrol and which the patrol artificial intelligence model cannot accurately identify.

[0009] The patrol AI model of the cloud server is optimized based on the correlation information of the edge anomaly events to obtain an optimized patrol AI model.

[0010] The patrol AI model on the vehicle of at least one unmanned patrol vehicle is upgraded using the optimized patrol AI model.

[0011] Secondly, this disclosure also provides a patrol system based on a patrol artificial intelligence model, including: a cloud server and at least one unmanned patrol vehicle;

[0012] The cloud server is configured to receive association information of a first abnormal event from at least one unmanned patrol vehicle, wherein the first abnormal event is identified by the patrol AI model onboard the unmanned patrol vehicle; determine association information of edge abnormal events based on the association information of the first abnormal event, wherein the edge abnormal events include events encountered by the unmanned patrol vehicle during patrol and which the patrol AI model cannot accurately identify; optimize the patrol AI model on the cloud server based on the association information of the edge abnormal events to obtain an optimized patrol AI model; and upgrade the patrol AI model onboard the at least one unmanned patrol vehicle using the optimized patrol AI model, so that the at least one unmanned patrol vehicle can identify abnormal events using the onboard patrol AI model.

[0013] The at least one unmanned patrol vehicle is used to conduct mobile patrols of the target area, and when a first abnormal event is identified by the onboard patrol artificial intelligence model, it sends the associated information of the first abnormal event to the cloud server, and receives the optimized patrol artificial intelligence model sent by the cloud server, and upgrades the onboard patrol artificial intelligence model using the optimized patrol artificial intelligence model.

[0014] Thirdly, embodiments of this disclosure also provide an electronic device, the electronic device comprising: one or more processors; a storage device for storing one or more programs; and when the one or more programs are executed by the one or more processors, causing the one or more processors to implement the patrol method based on the patrol artificial intelligence model as described above.

[0015] Fourthly, embodiments of this disclosure also provide a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the patrol method based on the patrol artificial intelligence model as described above.

[0016] This disclosure provides a patrol method based on a patrol artificial intelligence model. A cloud server receives correlation information of a first abnormal event identified by an onboard patrol artificial intelligence model from at least one unmanned patrol vehicle. The server then filters out correlation information of edge abnormal events, including events encountered by the unmanned patrol vehicle during patrol that the patrol artificial intelligence model cannot accurately identify. The correlation information of these edge abnormal events is then used to optimize the patrol artificial intelligence model on the cloud server, resulting in an optimized patrol artificial intelligence model. Finally, the optimized patrol artificial intelligence model is used to upgrade the onboard patrol artificial intelligence model of the at least one unmanned patrol vehicle, forming a closed-loop optimization process for the patrol artificial intelligence model and enabling continuous optimization. Specifically, by automatically collecting edge abnormal events and using them to specifically optimize the patrol artificial intelligence model, the abnormal event identification capability of the patrol artificial intelligence model is significantly improved, thereby enhancing patrol effectiveness. Attached Figure Description

[0017] The above and other features, advantages, and aspects of the embodiments of this disclosure will become more apparent from the accompanying drawings and the following detailed description. Throughout the drawings, the same or similar reference numerals denote the same or similar elements. It should be understood that the drawings are schematic, and the originals and elements are not necessarily drawn to scale.

[0018] Figure 1 This is a flowchart of a patrol method based on a patrol artificial intelligence model, as described in an embodiment of this disclosure.

[0019] Figure 2 This is a schematic diagram of the structure of a patrol system based on a patrol artificial intelligence model according to an embodiment of this disclosure;

[0020] Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this disclosure. Detailed Implementation

[0021] Embodiments of this disclosure will now be described in more detail with reference to the accompanying drawings. While some embodiments of this disclosure are shown in the drawings, it should be understood that this disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of this disclosure. It should be understood that the accompanying drawings and embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of protection of this disclosure.

[0022] It should be noted that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules or units, and are not used to limit the order of functions performed by these devices, modules or units or their interdependencies.

[0023] The names of messages or information exchanged between multiple devices in the embodiments of this disclosure are for illustrative purposes only and are not intended to limit the scope of such messages or information.

[0024] Figure 1 This is a flowchart illustrating a patrol method based on a patrol artificial intelligence model, as described in an embodiment of this disclosure. The method can be executed by a patrol device based on the patrol artificial intelligence model. This device can be implemented using software and / or hardware and can be configured in an electronic device, such as a cloud server. Figure 1 As shown, the method may specifically include the following steps:

[0025] S110, Receive the associated information of the first abnormal event sent by at least one unmanned patrol vehicle.

[0026] The first anomalous event is identified by the patrol AI model onboard the unmanned patrol vehicle. There can be one or more unmanned patrol vehicles, each equipped with its own AI model. During patrols, the unmanned patrol vehicles collect data about their surroundings using onboard sensors (such as cameras and radar). Based on this environmental data, the patrol AI model identifies anomalous events and records the identified events as the first anomalous event. Examples of first anomalous events include "broken barbed wire" or "unusual person or animal entering the area."

[0027] The associated information of the first abnormal event can be sent directly to the cloud server by the unmanned patrol vehicle, or the unmanned patrol vehicle can send the associated information of the first abnormal event to a nearby edge computing node, and the edge computing node can then forward the associated information of the first abnormal event to the cloud server.

[0028] S120. Determine the association information of the edge abnormal event based on the association information of the first abnormal event.

[0029] Among them, edge anomalies include events encountered by unmanned patrol vehicles during patrols that cannot be accurately identified by the patrol AI model.

[0030] In some implementations, edge anomaly events are specifically low-confidence recognition events. These are events where the confidence level of the patrol AI model's anomaly recognition result is below a preset threshold (the default preset threshold is 85%, adjustable within the range of 80%-90%) for multiple consecutive frames. This means that the patrol AI model recognizes the existence of an anomaly, but its confidence level is low, reflecting the low accuracy of the patrol AI model's current recognition results. For example: a pedestrian wearing camouflage at dusk; the corresponding scene description: at sunset, the light is dim and yellowish; a pedestrian wearing camouflage clothing very similar in color to the surrounding environment is slowly moving beside the green belt in front of the unmanned patrol vehicle. For this scenario, the patrol AI model's recognition result is: an anomaly event exists, but the recognition confidence level is below the preset threshold for multiple consecutive frames. Therefore, this anomaly event belongs to the low-confidence recognition event category and will be identified as an edge anomaly event. Correspondingly, the association information of the first anomalous event includes the confidence level. The higher the confidence level, the higher the recognition accuracy of the first anomalous event. The association information of edge anomalous events is determined based on the association information of the first anomalous event, including: determining the association information of the first anomalous event with a confidence level less than a first threshold as the association information of edge anomalous events; or, determining the association information of the first anomalous event with a confidence level less than the first threshold in multiple consecutive frames as the association information of edge anomalous events. "Confidence level in multiple consecutive frames" can be understood as follows: taking an image as the input to a patrol AI model, for example, each time an image frame is input to the patrol AI model, the patrol AI model outputs the recognition result for that frame, which includes the confidence level and the identifier of the anomalous event.

[0031] In other implementations, a marginal anomaly specifically refers to a misidentified event, where the patrol AI model incorrectly identifies a normal target or phenomenon as a security anomaly, exhibiting a high confidence level but being determined to be an error after manual verification. For example, if a swaying bush in the wind is misjudged as an "intrusion" by the patrol AI model, this anomaly will be identified as a marginal anomaly, belonging to the category of events misidentified by the patrol AI model. Correspondingly, the association information of the marginal anomaly is determined based on the association information of the first anomaly, including: determining the association information of the first anomaly with a confidence level greater than a first threshold and which has been manually verified as a false alarm as the association information of the marginal anomaly.

[0032] In some implementations, edge anomalies are events missed by the patrol AI model, or events that the patrol AI model fails to identify as actually existing security anomalies, or events that the patrol AI model cannot identify. For example, when an anomaly occurs, the unmanned patrol vehicle is also at the location, but it fails to identify the anomaly, reflecting insufficient recognition capability of the patrol AI model. For example, an intruder climbs over a wall with complex mesh decorations. Due to the intruder's rapid movement and frequent obscuring of parts of its body by the mesh decorations, the unmanned patrol vehicle happens to be patrolling that section of the road, but its onboard patrol AI model fails to identify the anomaly. Correspondingly, the association information of the edge anomaly is determined based on the association information of the first anomaly, including:

[0033] The location and time of the second abnormal event are obtained. The second abnormal event is identified and reported to the cloud server by security devices other than unmanned patrol vehicles (e.g., fixed roadside security devices, patrol drones, etc.). A matching time period is determined based on the occurrence time, and a matching area is determined based on the occurrence location, wherein the occurrence time is within the matching time period and the occurrence location is within the matching area. A spatiotemporal matching algorithm is used to determine whether an unmanned patrol vehicle is located within the matching area during the matching time period. If an unmanned patrol vehicle is located within the matching area during the matching time period and there is no first abnormal event matching the second abnormal event, then the association information of the second abnormal event is determined as the association information of a marginal abnormal event.

[0034] The matching time period can be a time interval obtained by extending a certain period forward and backward from the occurrence time of the second anomalous event. For example, if the second anomalous event occurs at 8:00 AM on a certain day, the matching time period could be the time interval from 7:59 AM to 8:01 AM on that day. Similarly, the matching area can be a circular area with the location of the second anomalous event as the center and a set radius. Alternatively, the matching area can be a geographical area where the location of the second anomalous event can be observed. Matching areas associated with locations with a high frequency of historical anomalous events can be pre-generated, and the corresponding matching areas can be directly retrieved or searched in subsequent applications.

[0035] Specifically, the way other security devices identify the second anomaly can be the same as or different from the way the unmanned patrol vehicle identifies the first anomaly. Because different types of devices have different hardware systems and compatibility issues, they typically identify anomaly events differently. By having different types of security devices patrol the same area collaboratively, the accuracy and coverage of anomaly identification can be improved, thus ensuring the security of the patrolled area.

[0036] Optionally, for edge-related abnormal events, a continuous image sequence within a 5-second time window before and after the identification time is saved to form a complete context data packet, which includes metadata such as the original image sequence, timestamp, GPS location, and vehicle ID, to facilitate downstream applications.

[0037] S130. Optimize the patrol AI model of the cloud server based on the correlation information of the edge abnormal events to obtain the optimized patrol AI model.

[0038] Since edge anomalies are events encountered by unmanned patrol vehicles during patrols that cannot be accurately identified by the patrol AI model, the correlation information of edge anomalies is a high-quality positive sample for optimizing the identification ability of the patrol AI model. By using the correlation information of edge anomalies to optimize the patrol AI model, the goal of targeted optimization of the patrol AI model's weaknesses can be achieved, thereby significantly improving the identification ability of the patrol AI model.

[0039] Specifically, the patrol AI model of the cloud server and the patrol AI model on the unmanned patrol vehicle are the same model. When the cloud server collects a certain number of edge anomaly events, it uses the correlation information of the edge anomaly events to optimize the patrol AI model of the cloud server, obtaining an optimized patrol AI model. The cloud server then distributes the optimized patrol AI model to the unmanned patrol vehicle to update the patrol AI model on the unmanned patrol vehicle, enabling the unmanned patrol vehicle to continue to perform patrol tasks using the optimized patrol AI model. This forms an optimization closed loop for the patrol AI model, achieving the goal of continuous optimization of the patrol AI model.

[0040] In some implementations, the patrol AI model of the cloud server is optimized based on the correlation information of edge anomaly events to obtain an optimized patrol AI model, including the following sub-steps 131-134:

[0041] 131. Determine the characteristic factors based on the correlation information of the aforementioned edge anomaly events.

[0042] The associated information of the edge anomaly event includes images detected by the onboard sensors of the unmanned patrol vehicle, targets identified by the patrol artificial intelligence model based on the images, the physical location of the targets, the position of the targets in the images, the speed of the targets, and the motion information of the unmanned patrol vehicle at the corresponding time (such as speed, acceleration, heading angle, etc.).

[0043] Feature factors are determined based on the correlation information of the edge anomaly events, including determining the image's brightness and contrast. Specifically, if the image's brightness is high, it indicates overexposure, which can interfere with the model's inference. Therefore, data augmentation operations can be used to preprocess the image to improve image quality. Similarly, if the image's brightness is low, it indicates underexposure, which can also interfere with the model's inference. Both excessively high and low contrast can affect the model's inference. Therefore, brightness and contrast can be identified as feature factors, and then matching data augmentation operations can be called to preprocess the image based on these feature factors to improve image quality and enhance the effectiveness of subsequent image-based model training.

[0044] The relative angle between the target and the unmanned patrol vehicle is determined based on the target's physical location and the corresponding physical location of the unmanned patrol vehicle. Based on this relative angle, geometric transformation enhancement operations (such as rotation, scaling, cropping, affine transformation, etc.) can be applied to the image to make the target more easily observable by the unmanned patrol vehicle, thereby improving image quality and facilitating subsequent model training.

[0045] If the target or the unmanned patrol vehicle is moving at high speed, motion-related data augmentation processing can be applied to the image, such as Gaussian noise reduction or Gaussian blurring. The goal is to enhance the image data, improve image quality, and facilitate subsequent model training.

[0046] In summary, one or more of the following are identified as feature factors: brightness value, contrast, relative angle, target position in the image, target speed, and motion information of the unmanned patrol vehicle at the corresponding time. These feature factors are used to determine data augmentation operations that match the image.

[0047] 132. Determine the matching data augmentation operation based on the aforementioned feature factors.

[0048] Optionally, you can look up the mapping relationship between feature factors and data augmentation operations to determine the matching data augmentation operations.

[0049] 134. Apply the matching data augmentation operation to the correlation information of the edge anomaly events to obtain training samples.

[0050] Specifically, the matching data augmentation operation is applied to the images in the associated information of the edge anomaly events to obtain training samples.

[0051] 135. The Elastic Weight Consolidation (EWC) algorithm is used to train the patrol AI model of the cloud server using the training samples to obtain an optimized patrol AI model.

[0052] Among them, the EWC (Elastic Weight Consolidation) algorithm is essentially a "memory-preserving" incremental learning algorithm that solves the "catastrophic forgetting" problem in incremental learning. Its core logic is to add a penalty term to the weights in the model that are important to old knowledge when training on new data, so that these weights are not significantly modified, and only the weights that are useful to new knowledge are adjusted, so as to achieve the goal of learning new knowledge without forgetting old knowledge.

[0053] S140. Upgrade the onboard patrol artificial intelligence model of the at least one unmanned patrol vehicle using the optimized patrol artificial intelligence model.

[0054] Specifically, the onboard patrol AI model of the unmanned patrol vehicle can be silently upgraded using OTA (Over-The-Air) technology, which utilizes the optimized patrol AI model.

[0055] In some implementations, a canary release strategy is used to gradually expand the model upgrade scope (10% → 20% → 50% → 100%). For example, first, the onboard patrol intelligence model of 10% of the unmanned patrol vehicles is upgraded, and then the performance of the upgraded patrol intelligence model is verified. If anomalies are found, the upgrade is immediately rolled back and the upgrade operation is canceled. If no anomalies are found and the upgraded patrol intelligence model performs well, the upgrade scope is expanded to upgrade the onboard patrol intelligence model of 20% of the unmanned patrol vehicles, and then the performance of the upgraded patrol intelligence model is verified, until the patrol intelligence model of all unmanned patrol vehicles has been upgraded. In summary, the optimized patrol AI model is deployed to a predetermined number of unmanned patrol vehicles, enabling these vehicles to use the optimized model for anomaly detection. Performance metrics (e.g., false positive rate, false negative rate) of the unmanned patrol vehicles using the optimized model are statistically analyzed over a set period. If the performance metrics meet the set requirements, the number of unmanned patrol vehicles deploying the optimized model is increased, and the performance metrics are again statistically analyzed over the set period until all unmanned patrol vehicles have the optimized model deployed. If the performance metrics do not meet the set requirements, the unmanned patrol vehicles revert to using the unoptimized patrol AI model for anomaly detection.

[0056] In summary, a dual-model hot-switching technology is employed. After receiving a new model, the unmanned patrol vehicle first loads the new model into memory for backup. At the start of the next patrol cycle, it switches to the new model, while the old model is retained in memory as a backup. If the new model malfunctions, the system automatically switches back to the old model, ensuring uninterrupted service and achieving seamless upgrades with "zero downtime."

[0057] The updated patrol AI model has been put into operation on unmanned patrol vehicles, continuing to identify and report abnormal events in new patrol missions. This forms a self-evolutionary closed loop of "data-driven optimization, optimization improving capabilities, and capabilities generating new data," achieving continuous improvement in the capabilities of the patrol AI model and fundamentally solving the industry problem of the performance "degradation" of traditional patrol AI models over time after deployment.

[0058] In some implementations, the patrol method based on the patrol AI model further includes: if the confidence level is greater than a first threshold, indicating a high probability of an abnormal event, the patrol strategy is determined to be target tracking to facilitate the tracing of the abnormal event and achieve longer-term evidence retention. If the confidence level is less than the first threshold, indicating a low probability of an abnormal event, the patrol strategy is determined to be one or more of the following: adjusting the relative angle between the unmanned patrol vehicle and the target (by adjusting the angle, the unmanned patrol vehicle can repeatedly identify the abnormal event to verify the previous identification result and improve the identification reliability), adjusting the relative distance between the unmanned patrol vehicle and the target (by adjusting the angle, the unmanned patrol vehicle can repeatedly identify the abnormal event to verify the previous identification result), dispatching other unmanned patrol vehicles to move to the corresponding location for collaborative patrol (by collaborative patrol, the identification accuracy of abnormal events is improved), and sending control commands to the associated roadside equipment to drive the roadside equipment to collaboratively track and stare at the target; and sending control commands to the unmanned patrol vehicle according to the patrol strategy. By determining the matching patrol strategy based on the confidence level, the identification accuracy and reliability of abnormal events can be improved.

[0059] In some implementations, a patrol system based on a patrol artificial intelligence model is also provided, such as... Figure 2 As shown, the patrol system based on the patrol artificial intelligence model includes a cloud server 210 and at least one unmanned patrol vehicle 220.

[0060] The cloud server 210 is configured to receive association information of a first abnormal event from at least one unmanned patrol vehicle 220, wherein the first abnormal event is identified by the patrol AI model onboard the unmanned patrol vehicle; determine association information of edge abnormal events based on the association information of the first abnormal event, wherein the edge abnormal events include events encountered by the unmanned patrol vehicle during patrol and which the patrol AI model cannot accurately identify; optimize the patrol AI model on the cloud server based on the association information of the edge abnormal events to obtain an optimized patrol AI model; and upgrade the patrol AI model onboard the at least one unmanned patrol vehicle using the optimized patrol AI model, so that the at least one unmanned patrol vehicle can identify abnormal events through the patrol AI model onboard the vehicle.

[0061] The at least one unmanned patrol vehicle 220 is used to conduct mobile patrols of the target area, and when a first abnormal event is identified by the onboard patrol artificial intelligence model, it sends the associated information of the first abnormal event to the cloud server, and receives the optimized patrol artificial intelligence model sent by the cloud server, and upgrades the onboard patrol artificial intelligence model using the optimized patrol artificial intelligence model.

[0062] The associated information of the first abnormal event includes a confidence level. A higher confidence level indicates higher accuracy in identifying the first abnormal event. The at least one unmanned patrol vehicle 220 is further configured to: if the confidence level is greater than a first threshold, determine the patrol strategy as tracking the target; if the confidence level is less than the first threshold, determine the patrol strategy as one or more of the following: adjusting the relative angle between the unmanned patrol vehicle and the target, adjusting the relative distance between the unmanned patrol vehicle and the target, dispatching other unmanned patrol vehicles to corresponding locations for collaborative patrolling, and sending control commands to associated roadside equipment (e.g., sending control commands to surrounding roadside equipment via vehicle-to-everything (V2X) wireless communication technology) to drive the roadside equipment to collaboratively track and stare at the target; and continue patrolling according to the patrol strategy. By determining a matching patrol strategy based on the confidence level, the accuracy of abnormal event identification can be improved.

[0063] This disclosure overcomes the blind spots of fixed monitoring systems by leveraging the autonomous movement and real-time environmental perception of unmanned patrol vehicles, transforming passive monitoring into proactive discovery and early warning, thus improving the real-time performance of security responses. By constructing an automated closed loop from data collection at the terminal (including the unmanned patrol vehicle) to model training in the cloud, the system can continuously optimize its artificial intelligence model using edge case data (i.e., correlation information of edge anomalies) generated during operation. This allows it to adapt to dynamically changing patrol environments and new anomalies, maintaining and improving long-term identification accuracy. Through automated model optimization and online upgrade pipelines, the model iteration process is transformed from a manual project dependent on human intervention into a routine process completed automatically by the system, significantly reducing reliance on professional algorithm engineers for later maintenance and downtime for system upgrades. Through information interaction and collaborative decision-making between vehicles, between vehicles and the cloud, and between vehicles and fixed security facilities, coordinated handling of security incidents and optimized scheduling of global resources are achieved, forming a systematic security capability and overcoming the limitations of single-point capabilities.

[0064] The patrol system based on the patrol artificial intelligence model provided in this disclosure can execute the steps in the patrol method based on the patrol artificial intelligence model provided in this disclosure, and has the execution steps and beneficial effects, which will not be repeated here.

[0065] Figure 3 This is a schematic diagram of the structure of an electronic device according to an embodiment of this disclosure. See below for details. Figure 3 It shows a schematic diagram of a structure suitable for implementing the electronic device 300 in the embodiments of this disclosure. Figure 3 The electronic device shown is merely an example and should not be construed as limiting the functionality and scope of the embodiments disclosed herein.

[0066] like Figure 3 As shown, the electronic device 300 may include a processing device 301, a ROM 302, a RAM 303, a bus 304, an input / output (I / O) interface 305, an input device 306, an output device 307, a storage device 308, and a communication device 309. The processing device (e.g., a central processing unit, a graphics processor, etc.) 301 can perform various appropriate actions and processes to implement the methods of the embodiments described herein, based on a program stored in the read-only memory (ROM) 302 or a program loaded from the storage device 308 into the random access memory (RAM) 303. The RAM 303 also stores various programs and data required for the operation of the electronic device 300. The processing device 301, the ROM 302, and the RAM 303 are interconnected via the bus 304. The input / output (I / O) interface 305 is also connected to the bus 304.

[0067] In particular, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a non-transitory computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts, thereby implementing the patrol method based on the patrol artificial intelligence model as described above. In such embodiments, the computer program can be downloaded and installed from a network via communication device 309, or installed from storage device 308, or installed from ROM 302. When the computer program is executed by processing device 301, it performs the functions defined in the methods of embodiments of this disclosure.

[0068] It should be noted that the computer-readable medium described in this disclosure can be a computer-readable signal medium or a computer-readable storage medium, or any combination thereof. A computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of a computer-readable storage medium may include, but are not limited to: an electrical connection having one or more wires, a portable computer disk, a hard disk, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage device, magnetic storage device, or any suitable combination thereof. In this disclosure, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this disclosure, a computer-readable signal medium can include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can be any computer-readable medium other than a computer-readable storage medium, which can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. The program code contained on the computer-readable medium can be transmitted using any suitable medium, including but not limited to: wires, optical fibers, RF (radio frequency), etc., or any suitable combination thereof.

[0069] The aforementioned computer-readable medium may be included in the aforementioned electronic device; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable medium carries one or more programs that, when executed by the electronic device, cause the electronic device to: receive association information of a first abnormal event from at least one unmanned patrol vehicle, wherein the first abnormal event is identified by the patrol AI model onboard the unmanned patrol vehicle; determine association information of edge abnormal events based on the association information of the first abnormal event, the edge abnormal events including events encountered by the unmanned patrol vehicle during patrol and which the patrol AI model cannot accurately identify; optimize the patrol AI model on the cloud server based on the association information of the edge abnormal events to obtain an optimized patrol AI model; and upgrade the patrol AI model onboard the at least one unmanned patrol vehicle using the optimized patrol AI model.

[0070] Optionally, when one or more of the above-described procedures are executed by the electronic device, the electronic device may also perform other steps described in the above embodiments.

[0071] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include, based on electrical connections of one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fibers, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0072] The above description is merely a preferred embodiment of this disclosure and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this disclosure is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features disclosed in this disclosure that have similar functions.

Claims

1. A patrol method based on a patrol artificial intelligence model, applied to a cloud server, characterized in that, include: Receive association information of a first abnormal event from at least one unmanned patrol vehicle, wherein the first abnormal event is identified by the patrol artificial intelligence model onboard the unmanned patrol vehicle; The association information of edge anomalies is determined based on the association information of the first anomaly event. The edge anomalies include events encountered by the unmanned patrol vehicle during patrol and which the patrol artificial intelligence model cannot accurately identify. The patrol AI model of the cloud server is optimized based on the correlation information of the edge anomaly events to obtain an optimized patrol AI model. The patrol AI model on the vehicle of at least one unmanned patrol vehicle is upgraded using the optimized patrol AI model.

2. The patrol method based on a patrol artificial intelligence model according to claim 1, characterized in that, Also includes: Receive the associated information of the second abnormal event; wherein the second abnormal event is identified by security devices other than the at least one unmanned patrol vehicle; The step of determining the association information of the edge anomaly event based on the association information of the first anomaly event includes: The location and time of the second abnormal event are obtained, and a matching time period is determined based on the occurrence time, and a matching region is determined based on the occurrence location, wherein the occurrence time is within the matching time period and the occurrence location is within the matching region; A spatiotemporal matching algorithm is used to determine whether an unmanned patrol vehicle is located within the matching area during the matching time period. If an unmanned patrol vehicle is located within the matching area during the matching time period and there is no first abnormal event matching the second abnormal event, then the association information of the second abnormal event is determined as the association information of the edge abnormal event.

3. The patrol method based on a patrol artificial intelligence model according to claim 1, characterized in that, The associated information of the first abnormal event includes the confidence level. The higher the confidence level, the higher the recognition accuracy of the first abnormal event. The step of determining the association information of the edge anomaly event based on the association information of the first anomaly event includes: The association information of the first abnormal event with a confidence level less than the first threshold is determined as the association information of the marginal abnormal event; And / or, the association information of the first abnormal event with a confidence level greater than the first threshold and which is manually verified as a false alarm is determined as the association information of the marginal abnormal event.

4. The patrol method based on a patrol artificial intelligence model according to claim 1, characterized in that, The associated information of the first abnormal event includes the confidence level. The higher the confidence level, the higher the recognition accuracy of the first abnormal event. The patrol method based on the patrol artificial intelligence model also includes: If the confidence level is greater than the first threshold, then the patrol strategy is determined to be the target tracking strategy; If the confidence level is less than the first threshold, the patrol strategy is determined to be one or more of the following: adjusting the relative angle between the unmanned patrol vehicle and the target, adjusting the relative distance between the unmanned patrol vehicle and the target, dispatching other unmanned patrol vehicles to move to the corresponding location for cooperative patrol, and sending control commands to the associated roadside equipment to drive the roadside equipment to perform cooperative tracking and staring at the target. Control commands are sent to the unmanned patrol vehicle according to the patrol strategy.

5. The patrol method based on a patrol artificial intelligence model according to claim 1, characterized in that, The step of optimizing the patrol AI model of the cloud server based on the correlation information of the edge anomaly events to obtain an optimized patrol AI model includes: Characteristic factors are determined based on the correlation information of the aforementioned edge anomaly events; Based on the aforementioned characteristic factors, determine the matching data augmentation operation; The matching data augmentation operation is applied to the correlation information of the edge anomaly events to obtain training samples; The Elastic Weight Consolidation (EWC) algorithm is used to train the patrol AI model on the cloud server using the training samples, thereby obtaining an optimized patrol AI model.

6. The patrol method based on a patrol artificial intelligence model according to claim 5, characterized in that, The associated information of the edge anomaly event includes images detected by the onboard sensors of the unmanned patrol vehicle, targets identified by the patrol artificial intelligence model based on the images, the physical location of the targets, the position of the targets in the images, the speed of the targets, and the motion information of the unmanned patrol vehicle at the corresponding time. The step of determining feature factors based on the correlation information of the edge anomaly events includes: Determine the brightness and contrast of the image; The relative angle between the target and the unmanned patrol vehicle is determined based on the physical location of the target and the physical location of the unmanned patrol vehicle at the corresponding moment. One or more of the following are identified as the feature factors: brightness value, contrast, relative angle, target position in image, target speed, and motion information of unmanned patrol vehicle at corresponding time.

7. A patrol system based on a patrol artificial intelligence model, characterized in that, include: Cloud servers and at least one unmanned patrol vehicle; The cloud server is configured to receive association information of a first abnormal event from at least one unmanned patrol vehicle, wherein the first abnormal event is identified by the patrol AI model onboard the unmanned patrol vehicle; determine association information of edge abnormal events based on the association information of the first abnormal event, wherein the edge abnormal events include events encountered by the unmanned patrol vehicle during patrol and which the patrol AI model cannot accurately identify; optimize the patrol AI model on the cloud server based on the association information of the edge abnormal events to obtain an optimized patrol AI model; and upgrade the patrol AI model onboard the at least one unmanned patrol vehicle using the optimized patrol AI model, so that the at least one unmanned patrol vehicle can identify abnormal events using the onboard patrol AI model. The at least one unmanned patrol vehicle is used to conduct mobile patrols of the target area, and when a first abnormal event is identified by the onboard patrol artificial intelligence model, it sends the associated information of the first abnormal event to the cloud server, and receives the optimized patrol artificial intelligence model sent by the cloud server, and upgrades the onboard patrol artificial intelligence model using the optimized patrol artificial intelligence model.

8. The patrol system based on a patrol artificial intelligence model according to claim 7, characterized in that, The associated information of the first abnormal event includes the confidence level. The higher the confidence level, the higher the recognition accuracy of the first abnormal event. The at least one unmanned patrol vehicle is also used for: If the confidence level is greater than the first threshold, then the patrol strategy is determined to be the target tracking strategy; If the confidence level is less than the first threshold, the patrol strategy is determined to be one or more of the following: adjusting the relative angle between the unmanned patrol vehicle and the target, adjusting the relative distance between the unmanned patrol vehicle and the target, dispatching other unmanned patrol vehicles to move to the corresponding location for cooperative patrol, and sending control commands to the associated roadside equipment to drive the roadside equipment to perform cooperative tracking and staring at the target. Continue patrol operations according to the patrol strategy.

9. An electronic device, characterized in that, The electronic device includes: one or more processors; a storage device for storing one or more programs; and when the one or more programs are executed by the one or more processors, the one or more processors implement the patrol method based on the patrol artificial intelligence model as described in any one of claims 1-6.

10. A computer-readable storage medium, characterized in that, It stores a computer program that, when executed by a processor, implements the patrol method based on a patrol artificial intelligence model as described in any one of claims 1-6.

Images