An event arbitration method, a parking monitoring system, electronic devices, and a storage medium.

By performing correlation analysis and cumulative confidence assessment on the metadata of multiple camera units, the problems of missed event reports and false wake-ups caused by unreasonable confidence threshold settings in the existing technology are solved, thereby improving the robustness and reliability of the parking monitoring system.

CN123123543APending Publication Date: 2026-11-17HANGZHOU HUACHENG SOFTWARE TECH CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611256457.8
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-18
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing event arbitration methods, the confidence threshold is not set reasonably, which leads to unresolved issues such as missed event reporting or frequent wake-up of the master controller.

Method used

By acquiring metadata reported by multiple camera units, performing correlation analysis, determining whether the target events are the same event, caching metadata with confidence levels within a preset range to a queue, dynamically evaluating the cumulative confidence level, and generating a wake-up command when the cumulative confidence level reaches a threshold.

Benefits of technology

It effectively reduced the false alarm rate, avoided false wake-ups caused by single-frame noise interference, and improved the robustness and reliability of the parking monitoring system in complex scenarios.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123123543A_ABST
    Figure CN123123543A_ABST
Patent Text Reader

Abstract

This application relates to an event arbitration method, a parking monitoring system, an electronic device, and a storage medium. The arbitration terminal of the parking monitoring system is applied, and the parking monitoring system also includes a data acquisition terminal; the data acquisition terminal includes multiple camera units. The method includes: acquiring metadata reported by each camera unit; performing correlation analysis on the acquired metadata to determine whether the target events corresponding to different metadata are the same target event; when the target events corresponding to different metadata are not the same target event, caching the target metadata with a confidence level within a preset target range to a target queue, and determining the cumulative confidence level of the target events in each camera unit based on the caching results; when the cumulative confidence level of the target event is greater than a preset cumulative confidence level threshold, generating a wake-up command. This method can solve the problem of existing event arbitration methods having issues such as unrealistic confidence level threshold settings leading to missed event reporting or frequent wake-ups of the main controller.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This application relates to the field of vehicle-mounted intelligent monitoring technology, and in particular to an event arbitration method, a parking monitoring system, electronic equipment, and a storage medium. Background Technology

[0002] With the rapid development of in-vehicle intelligent monitoring, intelligent driving, and vehicle networking technologies, the demand for parking monitoring while vehicles are parked is also increasing. Parking monitoring systems can continuously sense the surrounding environment and the interior of the vehicle while it is parked, and issue alarms and preserve evidence when a security event is detected. During the monitoring process, the detected data needs to be arbitrated to identify events requiring alarms and evidence preservation from a large amount of data, thus activating the main controller to record video and issue alarms.

[0003] However, existing event arbitration methods use a fixed confidence threshold to evaluate events. When the confidence level of an event does not reach the threshold, the data is discarded. When the confidence level of an event reaches the threshold, the master controller is woken up for further confirmation. This approach has the problem of unreasonably set confidence thresholds, which can lead to missed events or frequent wake-ups of the master controller.

[0004] The existing event arbitration methods have problems such as unwarranted event reporting or frequent master control wake-ups due to unreasonable confidence threshold settings. No effective solution has been proposed yet. Summary of the Invention

[0005] Therefore, it is necessary to provide an event arbitration method, a parking monitoring system, an electronic device, and a storage medium to address the aforementioned technical problems.

[0006] Firstly, this application provides an event arbitration method. Applied to the arbitration end of a parking monitoring system, the parking monitoring system further includes a data acquisition end; the data acquisition end includes multiple camera units installed at different locations, and the method includes:

[0007] Acquire the different metadata reported by each camera unit of the data acquisition terminal; the metadata includes at least the confidence level of the target event;

[0008] Perform correlation analysis on the different metadata obtained to determine whether the target events corresponding to the different metadata are the same target event;

[0009] When the target events corresponding to different metadata are not the same target event, the target metadata with a confidence level within a preset target range is cached in the target queue, and the cumulative confidence level of the target events in each camera unit is determined based on the caching results; the target queue is the queue in the target camera unit that stores the target events corresponding to the target metadata; the target camera unit is the camera unit that reports the target metadata; the preset target range is the confidence range where there is a risk of missed reports or false wake-ups;

[0010] A wake-up command is generated when the cumulative confidence value of the target event is greater than a preset cumulative confidence threshold.

[0011] In one embodiment, the step of performing correlation analysis on the acquired different metadata to determine whether the target events corresponding to the different metadata are the same target event includes:

[0012] Calculate the spatiotemporal correlation between different metadata to obtain the spatiotemporal correlation results between different metadata;

[0013] Based on the spatiotemporal correlation results between different metadata, it is determined whether the target events corresponding to different metadata are the same target event.

[0014] In one embodiment, the metadata further includes the time, location, and direction of movement of the target event; the calculation of the spatiotemporal correlation between different metadata to obtain the spatiotemporal correlation result between different metadata includes:

[0015] Based on the occurrence time of the target event in different metadata, determine the time difference of the target event between different metadata.

[0016] Determine the spatial distance between different metadata where the target event occurs, based on the location of the target event in different metadata.

[0017] Based on the movement direction of the target object in different metadata, determine the angle between the movement directions of the target object in different metadata.

[0018] In one embodiment, determining whether the target events corresponding to different metadata are the same target event based on the spatiotemporal correlation results between different metadata includes:

[0019] When the spatiotemporal correlation results between different metadata satisfy the preset criteria for determining the same target event, the target events corresponding to the different metadata are determined to be the same target event; otherwise, the target events corresponding to the different metadata are determined not to be the same target event. The preset criteria for determining the same target event are that the time difference between the target events occurring between different metadata is less than or equal to a preset time difference threshold, the spatial distance between the target events occurring between different metadata is less than or equal to a preset spatial distance threshold, and the angle between the motion directions of the target objects between different metadata is less than or equal to a preset angle threshold.

[0020] In one embodiment, the method further includes:

[0021] When the target event corresponding to different metadata is the same target event, the overall confidence level of the target event is determined based on the confidence level of the target event in different metadata.

[0022] A wake-up command is generated when the overall confidence level of the target event is greater than a preset overall confidence level threshold.

[0023] In one embodiment, when the target events corresponding to different metadata are the same target event, determining the overall confidence level of the target event based on the confidence levels of the target events in the different metadata includes:

[0024] When different metadata correspond to the same target event, the associated reward value of the target event is determined based on the angle between the movement directions of the target objects in the different metadata.

[0025] The sum of the target confidence level and the associated reward value is determined as the comprehensive confidence level of the target event; the target confidence level is the maximum confidence level among all confidence levels corresponding to the target event.

[0026] In one embodiment, when the target events corresponding to different metadata are not the same target event, the target metadata with a confidence level within a preset target range is cached in the target queue, and the cumulative confidence level of the target events in each camera unit is determined based on the caching results, including:

[0027] When the target events corresponding to different metadata are not the same target event, the target metadata with a confidence level in the preset target range is cached in the target queue to obtain the updated target queue;

[0028] The cumulative confidence level of the target event corresponding to the target queue before the update is determined by multiplying the historical cumulative confidence level and the decay factor, and then adding the sum of the confidence levels of the target events in the target metadata. The cumulative confidence level of the target event corresponding to the updated target queue is then determined as the cumulative confidence level of the target event in the target camera unit. The decay factor is a parameter related to the risk level of the target event and / or the priority of the location where the target event occurs. When the target queue is not updated within a preset time period, the historical cumulative confidence level decays according to a preset decay rule.

[0029] In one embodiment, the method further includes:

[0030] Based on the acquired vehicle status information, the arbitration mode is switched; the arbitration mode includes parking mode and temporary parking mode; different arbitration modes are configured with different target parameters; the target parameters include one or more of the following: preset time difference threshold, preset spatial distance threshold, preset cumulative confidence threshold, and attenuation factor.

[0031] Secondly, this application also provides a parking monitoring system. The system includes a data acquisition terminal, an arbitration terminal, and a main control terminal;

[0032] The data acquisition terminal includes multiple camera units installed at different locations, each camera unit being used to acquire images of various target areas; using a preset lightweight model, target event detection is performed on the acquired images to determine the confidence level of each target event; when the confidence level of a detected target event is greater than a preset reporting threshold, metadata of the target event is generated and reported to the arbitration terminal; the preset reporting threshold is lower than the lower limit of a preset target interval.

[0033] The arbitration end is used to execute the event arbitration method described in the first aspect above;

[0034] The main control terminal is used to record the target event in high definition and / or send alarm information when it receives the wake-up command sent by the arbitration terminal.

[0035] Thirdly, this application also provides an electronic device. The device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the event arbitration method described in the first aspect.

[0036] Fourthly, this application also provides a computer-readable storage medium. The computer-readable storage medium stores a computer program thereon, which, when executed by a processor, implements the event arbitration method described in the first aspect.

[0037] The aforementioned event arbitration method, parking monitoring system, electronic equipment, and storage medium acquire different metadata reported by various camera units at the data acquisition terminal. Then, they perform correlation analysis on the acquired metadata to determine whether the target events corresponding to different metadata are the same target event. When the target events corresponding to different metadata are not the same target event, the target metadata with a confidence level within a preset target range is cached in the target queue. Based on the caching results, the cumulative confidence level of the target event in each camera unit is determined. When the cumulative confidence level of the target event is greater than a preset cumulative confidence level threshold, an event is generated. The wake-up command accumulates and caches metadata that may have the risk of missed or false wake-ups. Based on the cached results, it evaluates the cumulative confidence of the target event. By continuously tracking and dynamically judging low-confidence events, it effectively reduces the missed rate of target events by accumulating confidence over time when the confidence of a single frame is insufficient. At the same time, it avoids false wake-ups caused by noise interference in a single frame, improves the robustness and reliability of the parking monitoring system in complex scenarios, and solves the problem of missed events or frequent wake-ups of the main controller caused by unreasonable confidence threshold settings in existing event arbitration methods.

[0038] Details of one or more embodiments of this application are set forth in the following drawings and description to make other features, objects and advantages of this application more readily apparent. Attached Figure Description

[0039] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0040] Figure 1 A hardware structure block diagram of a terminal for an event arbitration method provided in an embodiment of this application;

[0041] Figure 2 A flowchart of an event arbitration method provided in an embodiment of this application;

[0042] Figure 3 A flowchart of an event arbitration method provided in a preferred embodiment of this application;

[0043] Figure 4 This is a structural block diagram of a parking monitoring system provided in an embodiment of this application. Detailed Implementation

[0044] To better understand the purpose, technical solution, and advantages of this application, the application is described and illustrated below in conjunction with the accompanying drawings and embodiments.

[0045] Unless otherwise defined, the technical or scientific terms used in this application shall have the general meaning understood by one of ordinary skill in the art to which this application pertains. Words such as “a,” “an,” “an,” “the,” “the,” and “these” used in this application do not indicate quantitative limitation and may be singular or plural. The terms “comprising,” “including,” “having,” and any variations thereof used in this application are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or device that comprises a series of steps or modules (units) is not limited to the listed steps or modules (units) but may include steps or modules (units) not listed, or may include other steps or modules (units) inherent to these processes, methods, products, or devices. Words such as “connected,” “linked,” and “coupled” used in this application are not limited to physical or mechanical connections but may include electrical connections, whether direct or indirect. “Multiple” used in this application refers to two or more. “And / or” describes the relationship between related objects, indicating that three relationships may exist; for example, “A and / or B” can represent: A alone, A and B simultaneously, and B alone. Normally, the character " / " indicates that the objects before and after it are in an "or" relationship. The terms "first," "second," "third," etc., used in this application are merely to distinguish similar objects and do not represent a specific order of objects.

[0046] The method embodiments provided in this example can be executed on a terminal, computer, or similar computing device. For example, it can run on a terminal. Figure 1 This is a hardware structure block diagram of the terminal for the event arbitration method in this embodiment. For example... Figure 1 As shown, a terminal may include one or more ( Figure 1 Only one is shown in the diagram. A processor 102 and a memory 104 for storing data are also included. The processor 102 may be, but is not limited to, a microprocessor (MCU) or a programmable logic device (FPGA). The terminal may also include a transmission device 106 for communication functions and an input / output device 108. Those skilled in the art will understand that… Figure 1 The structure shown is for illustrative purposes only and does not limit the structure of the terminal described above. For example, the terminal may also include components that are larger than... Figure 1 The more or fewer components shown, or having the same Figure 1 The different configurations shown are illustrated.

[0047] The memory 104 can be used to store computer programs, such as application software programs and modules, like the computer program corresponding to the event arbitration method in this embodiment. The processor 102 executes various functional applications and data processing by running the computer programs stored in the memory 104, thereby implementing the aforementioned method. The memory 104 may include high-speed random access memory and may also include non-volatile memory, such as one or more magnetic storage devices, flash memory, or other non-volatile solid-state memory. In some instances, the memory 104 may further include memory remotely located relative to the processor 102, and these remote memories can be connected to the terminal via a network. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0048] The transmission device 106 is used to receive or send data via a network. This network includes a wireless network provided by the terminal's communication provider. In one example, the transmission device 106 includes a Network Interface Controller (NIC), which can connect to other network devices via a base station to communicate with the Internet. In another example, the transmission device 106 can be a Radio Frequency (RF) module used for wireless communication with the Internet.

[0049] This embodiment provides an event arbitration method applied to the arbitration end of a parking monitoring system. The parking monitoring system also includes a data acquisition end; the data acquisition end includes multiple camera units installed at different locations. Figure 2 This is a flowchart of the event arbitration method in this embodiment, as follows: Figure 2 As shown, the process includes the following steps:

[0050] Step S210: Obtain the different metadata reported by each camera unit at the data acquisition terminal; the metadata includes at least the confidence level of the target event.

[0051] The aforementioned multiple cameras can be multiple edge intelligent cameras. The multiple camera units installed in different locations can be cameras installed inside the vehicle or cameras installed outside the vehicle. For example, they can include multiple or all of the following: exterior surround view cameras (front, rear, left, and right), and interior cabin cameras such as those located on the A-pillar, center console, rearview mirrors, and rear roof. The aforementioned different metadata can be metadata reported by different camera units, or metadata reported by the same camera unit at different times.

[0052] The aforementioned arbitration end monitors each camera unit in real time to obtain different metadata reported by each camera unit at the data acquisition end. This metadata characterizes information about the target event, including at least the confidence level of the target event, which can be represented by a value between 0 and 100. The confidence level represents the probability that the target event is a real event, used to measure the credibility of the target event occurring in the footage captured by the camera unit. The confidence level of the target event is calculated using the internal recognition algorithm of the camera unit. Optionally, the metadata may also include one or more target fields. These target fields include: camera unit identifier (camera_id, used to distinguish camera units in different locations), the time of the target event, the type of the target event (e.g., movement, human, vehicle, collision, debris, child, pet, etc.), the location of the target event, the direction of movement of the target object, the location of the target object, and the speed of movement of the target object. It should be noted that the metadata generated by each camera uses a unified format. That is, the aforementioned metadata is standardized metadata after format conversion. The aforementioned target events refer to physical targets and their related behaviors or states that may pose a safety threat to the vehicle. The aforementioned target objects can be the entities that trigger the camera unit to generate an alarm signal. The specific objects of the target objects can be set according to the specific application scenario and requirements; this embodiment does not impose specific limitations. For example, the aforementioned target objects can be one or more of the following: a person, a vehicle, an animal, or other objects.

[0053] Furthermore, the positions of the aforementioned target objects and the locations of the target events are all within a world coordinate system. This world coordinate system is determined as follows: the external camera maps its image coordinates to a two-dimensional plane with the vehicle's center ground projection as the origin, using pre-calibrated camera extrinsic parameters (such as a homography matrix or a 3D projection matrix); the internal camera maps its image coordinates to a two-dimensional plane inside the vehicle cabin with the center of the dashboard or the bottom of the B-pillar as the origin, using pre-calibrated fixed feature points such as seats and door sills. This calibration method allows for the comparison of positional information in the metadata reported by different camera units within the same spatial reference system.

[0054] It should be noted that in the parking monitoring system provided in this embodiment, the arbitration end runs on the strategy engine MCU (Microcontroller Unit). That is, the logic function of the arbitration end is deployed on a low-power microcontroller. This design ensures that the arbitration end can always be online (with milliwatt or microwatt-level power consumption), thus reducing the burden on the vehicle's overall power consumption when parked. Furthermore, by deploying the logic function of the arbitration end on a low-power microcontroller, after generating a wake-up command, the arbitration end can re-evaluate the target event. Once the target event is confirmed to be real, the wake-up command is sent to the main control end, waking the main control SoC from sleep mode and initiating high-power operations such as full-function HD recording.

[0055] Step S220: Perform correlation analysis on the different metadata obtained to determine whether the target events corresponding to the different metadata are the same target event.

[0056] Specifically, when new metadata is received, the arbitration end performs correlation analysis between the newly received metadata and metadata received from other camera units within the same time window (e.g., within the previous 2 seconds) to determine whether the target events corresponding to the different metadata are the same target event. That is, the purpose of correlation analysis on the acquired different metadata is to determine whether the target events detected by different camera units within a similar time frame are the same. In one embodiment, the aforementioned correlation analysis on the acquired different metadata to determine whether the target events corresponding to the different metadata are the same target event can be performed by calculating the spatiotemporal correlation between different metadata through three dimensions: time, space, and motion consistency. This yields the spatiotemporal correlation results between the different metadata, and then, based on these results, it is determined whether the target events corresponding to the different metadata are the same target event.

[0057] Step S230: When the target events corresponding to different metadata are not the same target event, cache the target metadata with a confidence level in the preset target range to the target queue, and determine the cumulative confidence level of the target events in each camera unit based on the caching results; the target queue is the queue in the target camera unit that stores the target events corresponding to the target metadata; the target camera unit is the camera unit that reports the target metadata; the preset target range is the confidence range where there is a risk of missed reporting or false wake-up.

[0058] The aforementioned preset target range can be specifically set according to specific circumstances and needs, and this embodiment does not impose specific limitations here. The aforementioned preset target range is the value between the upper limit and the lower limit of the preset target range. When the confidence level is greater than the upper limit of the preset target range, it is clear that the target event exists, and there is no risk of false wake-up. When the confidence level is less than the lower limit of the preset target range, it is clear that the target event does not exist, and there is no risk of missed reporting. For example, the aforementioned preset target range can be [40, 70]. It should be noted that the numerical range of [40, 70] is only a specific implementation method provided by this application for ease of understanding, and the protection scope of the aforementioned preset target range includes but is not limited to this range. In practical applications, the upper and lower limits of the aforementioned preset target range can be adaptively set or pre-calibrated by those skilled in the art according to specific scenarios and parameters. When the confidence level is within the preset target range, the arbitration end cannot immediately make a decision on the target event reported by the metadata, and needs to put the metadata into the target queue (confidence accumulation queue) of the corresponding camera unit. The arbitration end is equipped with a storage queue (target queue) for each target event of each camera unit. When a camera unit reports new metadata for this target event, it is stored in the target queue. For each target queue, a confidence threshold needs to be configured. When the cumulative confidence of the target events determined by the target queue reaches the confidence threshold, accumulation stops, a confirmation event is triggered, and the system enters the confirmation state. The aforementioned confidence threshold can be specifically set according to specific circumstances and requirements. This embodiment does not impose a specific limitation here. Generally, the aforementioned confidence threshold can be set to a value between 150 and 300. For example, the aforementioned confidence threshold can be 200. It should be noted that the aforementioned value range of 150 to 300 is only a specific implementation method provided by this application for ease of understanding and does not constitute any limitation on the protection scope of the aforementioned confidence threshold.

[0059] The above-mentioned method of caching target metadata with confidence levels within a preset target range to a target queue and determining the cumulative confidence level of target events in each camera unit based on the caching results can be based on the historical cumulative confidence level of the target queue before the update and the confidence level of target events in the target metadata (database to be updated) to determine the cumulative confidence level of target events corresponding to the updated target queue; and the cumulative confidence level of target events corresponding to the updated target queue is determined as the cumulative confidence level of target events in the target camera unit.

[0060] Step S240: When the cumulative confidence value of the target event is greater than the preset cumulative confidence threshold, a wake-up command is generated.

[0061] The aforementioned preset cumulative confidence threshold can be specifically set according to specific application scenarios and requirements, and this embodiment does not impose specific limitations here. For example, the aforementioned preset cumulative confidence threshold can be set to 80. It should be noted that the aforementioned value of 80 is only a specific implementation method provided by this application for ease of understanding, and the protection scope of the aforementioned preset cumulative confidence threshold includes but is not limited to this value. The aforementioned wake-up command can be sent to the strategy engine MCU for secondary verification, or directly sent to the main control SoC (main control terminal) to start high-definition recording and / or send alarm information.

[0062] Steps S210 to S240 above involve acquiring different metadata reported by each camera unit at the data acquisition terminal, then performing correlation analysis on the acquired metadata to determine whether the target events corresponding to the different metadata are the same target event. When the target events corresponding to the different metadata are not the same target event, the target metadata with a confidence level within a preset target range is cached in the target queue, and the cumulative confidence level of the target event in each camera unit is determined based on the caching results. When the cumulative confidence level of the target event is greater than a preset cumulative confidence level threshold, a wake-up command is generated. It accumulates and caches metadata that may have the risk of missed detections or false wake-ups. Based on the cached results, it evaluates the cumulative confidence of the target event. By continuously tracking and dynamically judging low-confidence events, it effectively reduces the missed detection rate of the target event through confidence accumulation compensation over time when the confidence of a single frame detection is insufficient. At the same time, it avoids false wake-ups caused by noise interference in a single frame, improves the robustness and reliability of the parking monitoring system in complex scenarios, and solves the problem of existing event arbitration methods that have the problem of missed events or frequent wake-ups of the master controller due to unreasonable confidence threshold settings.

[0063] In one embodiment, step S220 involves performing correlation analysis on the acquired different metadata to determine whether the target events corresponding to the different metadata are the same target event, including:

[0064] Step S222: Calculate the spatiotemporal correlation between different metadata to obtain the spatiotemporal correlation results between different metadata.

[0065] The aforementioned spatiotemporal correlation results include one or more of the following: the time difference of target events occurring between different metadata, the spatial distance between target events occurring between different metadata, and the angle between the motion directions of target objects between different metadata. The calculation of the spatiotemporal correlation between different metadata, yielding the spatiotemporal correlation results, allows the arbitration end to extract information such as the time, location, and / or motion direction of target events from different metadata. It also allows for the calculation of the time difference of target events occurring between different metadata, the spatial distance between target events, and the angle between the motion directions of target objects. These calculation results serve as the spatiotemporal correlation results, which are used to determine whether target events detected by different camera units are the same target event.

[0066] Step S224: Based on the spatiotemporal correlation results between different metadata, determine whether the target events corresponding to different metadata are the same target event.

[0067] The aforementioned determination of whether target events corresponding to different metadata are the same target event based on the spatiotemporal correlation results between different metadata can be achieved by pre-setting preset conditions for determining the same target event. When the spatiotemporal correlation results between different metadata satisfy the preset conditions for determining the same target event, the target events corresponding to different metadata are determined to be the same target event; otherwise, the target events corresponding to different metadata are determined not to be the same target event. The aforementioned preset conditions for determining the same target event can include: the time difference between the occurrence of target events between different metadata being less than or equal to a preset time difference threshold; the spatial distance between the occurrence of target events between different metadata being less than or equal to a preset spatial distance threshold; and the angle between the motion directions of the target objects between different metadata being less than or equal to a preset angle threshold.

[0068] Preferably, in order to enhance the reliability of the association, other judgment conditions can be added based on the preset judgment conditions for the same target event, according to the security level of the target event. For example, for high security level event types, semantic consistency judgment conditions can be added in addition to the preset judgment conditions for the same target event. Only when the event types of different metadata are consistent or belong to the preset compatible type group are the target events corresponding to the metadata determined to be the same target event; otherwise, the target events corresponding to different metadata are determined to be different target events.

[0069] Steps S222 to S224 above calculate the spatiotemporal correlation between different metadata to obtain the spatiotemporal correlation results between different metadata. Then, based on the spatiotemporal correlation results between different metadata, it is determined whether the target events corresponding to different metadata are the same target event. By determining whether the target events corresponding to different metadata are the same target event, it is convenient to determine different arbitration processing paths based on the judgment results. When the judgment result is that the target events corresponding to different metadata are the same target event, it indicates that multiple camera units have observed the same target event in similar time, similar space, and similar motion direction. The arbitration end fuses the confidence in each metadata to determine the comprehensive confidence of the target event and generates a wake-up command based on the comprehensive confidence. When the judgment result is that the target events corresponding to different metadata are not the same target event, it indicates that the target event corresponding to the current metadata has not been synchronously verified by other camera units. The arbitration end further determines whether the confidence of the target event in the metadata falls into the preset target range. If so, the metadata is cached in the target queue of the corresponding camera unit and a wake-up command is generated based on the accumulated confidence. By differentiating and precisely arbitrating different types of target events, it ensures that real events are captured in a timely manner while avoiding false wake-ups caused by environmental interference from a single camera unit.

[0070] Specifically, in one embodiment, the metadata further includes the time, location, and direction of movement of the target event; step S222 calculates the spatiotemporal correlation between different metadata to obtain the spatiotemporal correlation results between different metadata, including:

[0071] Step S2222: Determine the time difference of the occurrence of the target event between different metadata based on the occurrence time of the target event in different metadata.

[0072] As described above, by determining the time difference between the occurrence of the target event in different metadata based on the time of the target event in different metadata, the timestamp and world coordinates of the new metadata can be extracted for the arbitration end and compared with the metadata from other camera units within the time window.

[0073] Step S2224: Determine the spatial distance between different metadata where the target event occurs, based on the location of the target event in different metadata.

[0074] Step S2226: Determine the angle between the movement directions of the target objects in different metadata based on the movement directions of the target objects in different metadata.

[0075] The method described above, which determines the angle between the movement directions of target objects in different metadata based on their movement directions, can be achieved by first converting the movement directions of target objects in different metadata into direction vectors in the same coordinate system. This yields the direction vectors corresponding to the movement directions of target objects in different metadata. Then, using a preset method, the angle between these direction vectors is determined, and this angle is defined as the angle between the movement directions of target objects in different metadata. Similarly, determining the angle between two direction vectors using a preset method can be achieved by first calculating the cosine of the two vectors using the dot product formula, and then using the inverse cosine function to obtain the angle between the two vectors.

[0076] Steps S2222 to S2226 above determine the time difference of target events between different metadata based on the occurrence time of the target events in different metadata. Then, they determine the spatial distance between target events between different metadata based on the location of the target events in different metadata. Finally, they determine the angle between the movement directions of target objects in different metadata based on the movement direction of the target objects in different metadata. By determining the spatiotemporal correlation results between different metadata, they judge whether the target events corresponding to different metadata are the same target event. Based on the judgment result, they determine the event arbitration processing path. When the judgment result is that the target events corresponding to different metadata are the same target event... When the events are the same target event, it indicates that multiple camera units have observed the same target event in similar time, space, and motion direction. The arbitrator fuses the confidence levels in each metadata to determine the overall confidence level of the target event and generates a wake-up command based on this overall confidence level. When the judgment result is that the target events corresponding to different metadata are not the same target event, it indicates that the target event corresponding to the current metadata has not been synchronously verified by other camera units. The arbitrator further determines whether the confidence level of the target event in the metadata falls within a preset target range. If so, the metadata is cached in the target queue of the corresponding camera unit, and a wake-up command is generated based on the accumulated confidence level. By arbitrating different types of target events in a differentiated and precise manner, it ensures that real events are captured in a timely manner while avoiding false wake-ups caused by environmental interference from a single camera unit.

[0077] In another embodiment, step S224, based on the spatiotemporal correlation results between different metadata, determines whether the target events corresponding to the different metadata are the same target event, including:

[0078] Step S2242: When the spatiotemporal correlation results between different metadata satisfy the preset same target event determination conditions, the target events corresponding to the different metadata are determined to be the same target event; otherwise, the target events corresponding to the different metadata are determined not to be the same target event. The preset same target event determination conditions are: the time difference between the occurrence of the target events between different metadata is less than or equal to the preset time difference threshold, the spatial distance between the occurrence of the target events between different metadata is less than or equal to the preset spatial distance threshold, and the angle between the motion directions of the target objects between different metadata is less than or equal to the preset angle threshold.

[0079] In a parking monitoring system, multiple camera units are installed at different locations on the vehicle, each independently acquiring images and detecting target events in AOV mode. When the same physical target (such as a pedestrian approaching the vehicle) passes through the field of view of multiple camera units in sequence, each camera unit will detect the target at different times. Therefore, only by setting the preset time difference threshold within a reasonable range can the accurate reporting of the same target event by different camera units before and after be determined.

[0080] The aforementioned preset time difference threshold can be dynamically adjusted based on whether a pre-warning wake-up flag is received, and can also be specifically set and adjusted according to specific application scenarios and requirements. For example, the aforementioned preset time difference threshold can also be determined based on the target's movement speed, the shooting areas of different camera units, etc. For example, when no pre-warning wake-up flag is received, the aforementioned preset time difference threshold is 500ms. When a pre-warning wake-up flag is received, the aforementioned preset time difference threshold is 200ms. It should be noted that the aforementioned 500ms or 200ms is only a specific implementation method provided by this application for ease of understanding, and the protection scope of the aforementioned preset time difference threshold includes but is not limited to this value. In addition, the aforementioned pre-warning wake-up flag is a cross-camera unit early activation signal. When a camera unit detects that a target is moving towards an adjacent camera unit, it generates a pre-warning wake-up signal and sends it to the arbitration end. The arbitration end then forwards the pre-warning wake-up signal to the adjacent camera unit, causing the adjacent camera unit to switch from low-power AOV mode to a higher frame rate mode in advance. Therefore, when a pre-warning wake-up flag is received, the preset time difference threshold will be set smaller to achieve more accurate correlation.

[0081] Furthermore, the aforementioned preset time difference threshold can be set to different values ​​depending on the specific installation location of the camera unit, and can also be specifically set and adjusted according to specific application scenarios and requirements. For example, if the camera unit is located outside the vehicle, the preset time difference threshold can be set to a larger value; if the camera unit is located inside the vehicle, the preset time difference threshold can be set to a smaller value. For example, if the camera unit is located outside the vehicle, the preset time difference threshold can be set to 100cm; if the camera unit is located inside the vehicle, the preset time difference threshold can be set to 30cm. It should be noted that the aforementioned 100cm or 30cm is merely a specific implementation method provided by this application for ease of understanding, and the protection scope of the aforementioned preset time difference threshold includes but is not limited to this value.

[0082] The aforementioned preset angle threshold can be set to different values ​​based on the state of the target object, and can also be specifically set and adjusted according to specific application scenarios and needs. For example, the aforementioned preset angle threshold can be 45°. It should be noted that the aforementioned angle value of 45° is only a specific implementation method provided by this application for ease of understanding, and the aforementioned preset angle threshold includes but is not limited to this value. For reference, for stationary targets, this condition can be appropriately relaxed and can be set to a larger angle value.

[0083] It should be noted that the aforementioned criteria for determining whether the target events corresponding to different metadata are the same can be at least one of the following three conditions: the time difference between the occurrence of target events in different metadata is less than or equal to a preset time difference threshold; the spatial distance between the occurrence of target events in different metadata is less than or equal to a preset spatial distance threshold; and the angle between the movement directions of target objects in different metadata is less than or equal to a preset angle threshold. It should also be noted that the more conditions are met simultaneously, the more accurate the determination of whether the target events corresponding to different metadata are the same target event.

[0084] In one embodiment, the event arbitration method further includes:

[0085] Step S250: When the target events corresponding to different metadata are the same target event, determine the comprehensive confidence level of the target event based on the confidence level of the target event in different metadata.

[0086] In this embodiment, when the arbitration end determines that the target events corresponding to different metadata are the same target event, a high-confidence event confirmation needs to be triggered. That is, at this time, the overall confidence level of the target event can be jointly determined by observing the same target event from different angles using different camera units. The overall confidence level of the target event can be calculated by fusing the confidence levels in the metadata related to the target event to obtain a fused confidence level value that comprehensively reflects the credibility of the target event. The above-mentioned determination of the overall confidence level of the target event based on the confidence levels of the target event in different metadata can be achieved by obtaining the confidence levels corresponding to each metadata that is determined to be the same target event, then determining the maximum confidence level among the confidence levels corresponding to each metadata, then determining the associated reward value corresponding to the target event based on the angle between the movement directions of the target objects in different metadata, and finally determining the overall confidence level of the target event by summing the maximum value and the associated reward value.

[0087] Step S260: When the overall confidence level of the target event is greater than the preset overall confidence level threshold, a wake-up command is generated.

[0088] The aforementioned preset comprehensive confidence threshold can be specifically set according to specific application scenarios and needs, and this embodiment does not impose specific limitations here. For example, the aforementioned preset comprehensive confidence threshold can be set to 80. It should be noted that the aforementioned value of 80 is only a specific implementation method provided by this application for ease of understanding, and the protection scope of the aforementioned preset comprehensive confidence threshold includes but is not limited to this value. For vehicle monitoring systems, according to different working processes, they can be divided into four states: idle state, listening state, confirmation state, and wake-up state. The aforementioned idle state is a state where there are no events and only metadata is being listened to. In the idle state, if the confidence of the target event of the generated metadata is less than the lower limit of the preset target interval, it is directly discarded; if the confidence of the target event of the generated metadata is greater than the upper limit of the preset target interval, it directly enters the confirmation state. The aforementioned listening state is a state where low-confidence metadata is received and more evidence is awaited. At this time, the listening window timer is started to wait for metadata from other camera units or for the accumulated threshold to be reached. The above confirmation state indicates successful spatiotemporal correlation or sufficient confidence accumulation. In this state, a mild wake-up command is output to activate the relevant camera unit and enter high-definition verification mode. The above wake-up state indicates successful secondary verification, confirming the target event as a real event. In this state, a full wake-up command is output to wake up the main control SoC and high-definition recording module, and an alarm can be triggered via the vehicle network. During the monitoring state, if the monitoring window exceeds a preset time and the conditions for entering the confirmation state are not met, the system returns to the idle state and performs cumulative attenuation.

[0089] In this embodiment, when the overall confidence level of the target event exceeds a preset overall confidence level threshold, a wake-up command is generated. At this point, the system directly enters the confirmation state and records the ID (Identification) of the associated camera unit. After the parking monitoring system enters the confirmation state, it does not directly wake up the main control unit. Instead, it first performs a mild wake-up: activating only 1-2 cameras (external or internal) in the area related to the target event, capturing 3-5 seconds of short video at medium resolution (e.g., 720P) and medium frame rate (15fps), and running a slightly larger verification model (which can run on the MCU or camera). For in-vehicle scenarios, the verification model can perform detailed identification of high-security event types (e.g., children / pets left behind, intrusion, etc.), such as judging the target's breathing micro-movements and persistence. The verification model outputs a secondary verification confidence level. When the secondary verification confidence level exceeds a preset threshold (e.g., 80, range 70-90), the arbitration unit enters the wake-up state and outputs a full wake-up command. If the confidence level of the secondary verification is less than or equal to the preset threshold, the system returns to an idle state and records the metadata characteristics of this false alarm for subsequent parameter adaptive adjustment.

[0090] Steps S250 to S260 above determine the overall confidence level of the target event based on the confidence level of the target event in different metadata when the target event corresponding to different metadata is the same target event. Then, when the overall confidence level of the target event is greater than a preset overall confidence level threshold, a wake-up command is generated. The overall confidence level of the target event is determined by comprehensively considering the confidence level of the target event in different metadata that report the same target data. This allows for obtaining an accurate overall confidence level of the target event based on the observation results from different angles of different camera units. The wake-up command for the target event is generated when the overall confidence level is greater than the preset overall confidence level threshold, or the multiple metadata corresponding to the target event are stored or discarded when the overall confidence level of the target event is less than a preset discard threshold (the threshold for determining whether metadata needs to be discarded or cached).

[0091] In another embodiment, step S250, when the target events corresponding to different metadata are the same target event, determines the overall confidence level of the target event based on the confidence levels of the target events in the different metadata, including:

[0092] Step S252: When the target events corresponding to different metadata are the same target event, determine the associated reward value corresponding to the target event based on the angle between the movement directions of the target objects between the different metadata.

[0093] The aforementioned associated reward value can be used as a quantitative reward for multiple camera units simultaneously observing the same target object or the same target event from the same direction. The smaller the angle between the motion directions of the target object in different metadata, and the more consistent the motion directions, the greater the probability that different camera units have observed the same target, and the larger the corresponding associated reward value. Conversely, the larger the angle between the motion directions of the target object in different metadata, and the more inconsistent the motion directions, the lower the probability that different camera units have observed the same target, and the smaller the corresponding associated reward value. When the angle between the motion directions of the target object in different metadata reaches a certain value, the associated reward value can decay to 0. Therefore, the aforementioned associated reward value is dynamically determined based on the degree of consistency of the motion directions of the target object in different metadata.

[0094] In this step, the calculation process for determining the associated reward value corresponding to the target event based on the angle between the movement directions of target objects between different metadata is as follows:

[0095] ;

[0096] Wherein, bonus_max is the preset maximum reward value, θ_max is the maximum allowed angle (generally 45°; it should be noted that 45° is only a specific implementation method provided by this application for ease of understanding, and the maximum allowed angle includes but is not limited to this value), and θ is the angle between the movement directions of the target objects between different metadata.

[0097] The aforementioned preset maximum reward value can be the associated reward value corresponding to a 0° (completely identical) angle between the movement directions of the various metadata. The setting of the aforementioned preset maximum reward value is related to the location where the target event occurs; that is, the preset maximum reward value varies depending on the location of the target event, with the preset maximum reward value for events occurring outside the vehicle being less than that for events occurring inside the vehicle. For example, when the target event occurs outside the vehicle, its preset maximum reward value can be 20, and when the target event occurs inside the vehicle, its preset maximum reward value can be 30. It should be noted that the aforementioned values ​​of 20 or 30 are merely a specific implementation provided by this application for ease of understanding, and the protection scope of the aforementioned preset maximum reward value includes, but is not limited to, this value.

[0098] In a preferred embodiment, a segmented mapping table can also be used to determine the correspondence of associated reward values. This table pre-defines the correspondence between the motion direction angles of target objects between different metadata and their associated reward values. The associated reward value corresponding to the motion direction angles of target objects between different metadata is determined by looking up the table. This segmented mapping table allows the confirmation accuracy of spatiotemporal correlation to change smoothly with motion consistency, avoiding the instability of the decision boundary caused by hard thresholds.

[0099] Step S254: The sum of the target confidence level and the associated reward value is determined as the comprehensive confidence level of the target event; the target confidence level is the maximum confidence level among all the confidence levels corresponding to the target event.

[0100] In this embodiment, the calculation process for the comprehensive confidence level of the event identified as the target event is as follows:

[0101] ;

[0102] max(conf1, conf2) represents the target confidence level, where conf1 and conf2 represent different confidence levels corresponding to the target event.

[0103] Steps S252 to S254 above determine the associated reward value of the target event based on the angle between the movement directions of the target objects in different metadata when the target events corresponding to different metadata are the same target event. Then, the sum of the target confidence and the associated reward value is determined as the comprehensive confidence of the target event. By determining the comprehensive confidence of the target event, it is convenient to determine whether to generate a wake-up command based on the comprehensive confidence of the target event.

[0104] Further, in one embodiment, step S230, when the target events corresponding to different metadata are not the same target event, caches the target metadata with confidence levels within a preset target range to the target queue, and determines the cumulative confidence level of the target events in each camera unit based on the caching results, including:

[0105] Step S232: When the target events corresponding to different metadata are not the same target event, cache the target metadata with confidence in the preset target range to the target queue to obtain the updated target queue.

[0106] Step S234: The product of the historical cumulative confidence level and the decay factor corresponding to the target queue before the update, plus the sum of the confidence levels of the target events in the target metadata, is used to determine the cumulative confidence level of the target events corresponding to the updated target queue; the cumulative confidence level of the target events corresponding to the updated target queue is used to determine the cumulative confidence level of the target events in the target camera unit; the decay factor is a parameter related to the risk level of the target event and / or the priority of the location where the target event occurs; when the target queue is not updated within a preset time period, the historical cumulative confidence level is decayed according to a preset decay rule.

[0107] The calculation process for the cumulative confidence C_acc(t+1) of the target events corresponding to the updated target queue is as follows:

[0108] ;

[0109] Where C_acc(t) is the historical cumulative confidence level of the target queue before the update, decay_rate is the decay factor, and confidence_new is the confidence level of the target event in the target metadata.

[0110] When the target queue is not updated within a preset time period, the historical cumulative confidence level is decayed according to a preset decay rule. This means that if the target queue does not receive new metadata within the preset time period, its cumulative confidence level is automatically multiplied by a decay factor. It should be noted that the preset time period can be specifically set according to specific circumstances and needs, and this embodiment does not impose any specific limitations. For example, the preset time period can be 1 second. It should be noted that the 1 second mentioned above is merely a specific implementation method provided for ease of understanding in this application and does not constitute any limitation on the scope of protection of the preset time period. The preset decay rule can also be based on a multiple of the decay factor or other parameters, and this embodiment does not impose any specific limitations here.

[0111] Furthermore, the values ​​of the aforementioned attenuation factors can be differentiated based on the event type or the risk level of the target event's location in the metadata. When the event type is high-risk or the target event occurs in a preset high-priority area (such as the rear child safety seat area), an attenuation factor closer to 1 (e.g., 0.98) is used, causing its accumulated confidence to decay more slowly. This allows the parking monitoring system to retain the memory of such events for a longer period, making it easier to accumulate to the threshold and thus reducing the risk of missed alarms. For ordinary event types, a normal attenuation factor (e.g., 0.95) is maintained to ensure reasonable filtering of routine low-confidence events and avoid false alarms. This embodiment achieves a differentiated balance between false alarm control and safety response through the aforementioned differentiated attenuation factor mechanism, where high-risk events accumulate faster and decay more slowly, while the filtering capability for ordinary events remains unchanged.

[0112] In addition, if the metadata in the target queue does not reach the preset minimum threshold within the timeout window (e.g., 10 seconds; it should be noted that 10 seconds is only a specific implementation method provided by this application for ease of understanding, and the timeout window includes but is not limited to this value), the historical cumulative confidence of the target queue will gradually decay until it is cleared to zero, and the vehicle monitoring system will return to the idle state.

[0113] Steps S232 to S234 above involve caching target metadata with confidence levels within a preset target range to a target queue when the target events corresponding to different metadata are not the same target event, thus obtaining an updated target queue. Then, the product of the historical cumulative confidence level and the decay factor corresponding to the target queue before the update, plus the sum of the confidence levels of the target events in the target metadata, is determined as the cumulative confidence level of the target events corresponding to the updated target queue. The cumulative confidence level of the target events corresponding to the updated target queue is then determined as the cumulative confidence level of the target events in the target camera unit. By determining the cumulative confidence level of the target events, it is easier to generate wake-up commands based on the cumulative confidence level of the target events in the future.

[0114] In one embodiment, the event arbitration method further includes:

[0115] Step S270: Based on the acquired vehicle status information, switch the arbitration mode; the arbitration mode includes parking mode and temporary parking mode; different arbitration modes are configured with different target parameters; the target parameters include one or more of the following: preset time difference threshold, preset spatial distance threshold, preset cumulative confidence threshold, and attenuation factor.

[0116] The vehicle status information mentioned above includes one or more of the following: vehicle speed, door lock status, armed status, and window position. The parking mode mentioned above can be an arbitration mode corresponding to long-term vehicle parking. In this mode, compared to scenarios without arbitration, when equipped with corresponding target parameters, one or more of the following can be implemented: reducing the preset time difference threshold, reducing the preset spatial distance threshold, activating only key cameras, establishing strong correlation with door / window status signals, reducing the preset cumulative confidence threshold, and activating attenuation factors corresponding to high-risk events. The aforementioned temporary parking mode can be the arbitration mode corresponding to temporary vehicle parking. In this mode, compared to the scenario without arbitration mode, when equipped with the corresponding target parameters, the preset spatial distance threshold can be adaptively adjusted, for example, the preset spatial distance threshold can be tightened from 100cm to 20cm; the maximum reward value for semantic association determination can be adaptively adjusted, for example, the maximum reward value for semantic association determination can be increased from 30 to 40; the attenuation factor for regular events can be adaptively adjusted, for example, the attenuation factor for regular events can be adjusted from 0.95 to 0.97; the attenuation factor for high-risk events can be adaptively adjusted, for example, the attenuation factor for high-risk events can be adjusted from 0.98 to 0.99; the preset time difference threshold can remain unchanged (for example, 500ms; it should be noted that 500ms here is only a specific implementation method provided by this application for ease of understanding, and the preset time difference threshold includes but is not limited to this value). The target parameters mentioned above can also be adjusted based on environmental scenario information (such as being close to a wall, in an open area, near a car, or under a tree), based on historical false alarm rate statistics, and can also be adaptively adjusted based on needs and application scenarios.

[0117] In one embodiment, after the master control terminal completes event recording and confirms the end of the event (such as the target leaving or the vehicle owner disarming), it sends an event end signal to the arbitration terminal. The arbitration terminal clears the accumulated queue of the relevant camera units, resets the state machine to the idle state, and can fine-tune the confidence assessment model of the relevant camera units based on whether the current event is a real event.

[0118] The present embodiment will now be described and illustrated through preferred embodiments.

[0119] Figure 3 This is a flowchart of an event arbitration method provided in a preferred embodiment of this application. The arbitration terminal is applied to a parking monitoring system, which also includes a data acquisition terminal; the data acquisition terminal includes multiple camera units installed in different locations, such as... Figure 3 As shown, the arbitration method for this event includes the following steps:

[0120] Step S301: Obtain the different metadata reported by each camera unit of the data acquisition terminal; the metadata includes at least the confidence level of the target event;

[0121] Step S302: Calculate the spatiotemporal correlation between different metadata to obtain the spatiotemporal correlation results between different metadata;

[0122] Step S303: Based on the spatiotemporal correlation results between different metadata, determine whether the target events corresponding to different metadata are the same target event;

[0123] Step S304: When the target events corresponding to different metadata are not the same target event, cache the target metadata with a confidence level in the preset target range to the target queue, and determine the cumulative confidence level of the target events in each camera unit based on the caching results; the target queue is the queue in the target camera unit that stores the target events corresponding to the target metadata; the target camera unit is the camera unit that reports the target metadata; the preset target range is the confidence range where there is a risk of missed reporting or false wake-up.

[0124] Step S305: When the target events corresponding to different metadata are the same target event, determine the comprehensive confidence level of the target event based on the confidence levels of the target events in different metadata.

[0125] Step S306: When the cumulative confidence value of the target event is greater than the preset cumulative confidence threshold, and / or when the comprehensive confidence value of the target event is greater than the preset comprehensive confidence threshold, a wake-up command is generated.

[0126] Steps S301 to S306 above involve acquiring different metadata reported by each camera unit at the data acquisition terminal, then performing correlation analysis on the acquired metadata to determine whether the target events corresponding to the different metadata are the same target event. When the target events corresponding to the different metadata are not the same target event, the target metadata with a confidence level within a preset target range is cached in the target queue, and the cumulative confidence level of the target event in each camera unit is determined based on the caching results. When the cumulative confidence level of the target event is greater than a preset cumulative confidence level threshold, a wake-up command is generated. It accumulates and caches metadata that may have the risk of missed detections or false wake-ups. Based on the cached results, it evaluates the cumulative confidence of the target event. By continuously tracking and dynamically judging low-confidence events, it effectively reduces the missed detection rate of the target event through confidence accumulation compensation over time when the confidence of a single frame detection is insufficient. At the same time, it avoids false wake-ups caused by noise interference in a single frame, improves the robustness and reliability of the parking monitoring system in complex scenarios, and solves the problem of existing event arbitration methods that have the problem of missed events or frequent wake-ups of the master controller due to unreasonable confidence threshold settings.

[0127] This application provides an example of spatiotemporal correlation analysis and confidence accumulation of multiple cameras outside a vehicle.

[0128] For example, a vehicle is parked on the roadside in a parked armed state. The left-side camera detects a human-shaped target with a confidence level of 75, world coordinates (-150cm, 50cm), and the motion vector direction is pointing towards the vehicle. Almost simultaneously (with a time difference of 150ms), the front camera also detects the same target with a confidence level of 70, world coordinates (-120cm, 80cm), and the motion vector direction angle θ=10°.

[0129] The arbitration panel performs spatiotemporal correlation analysis on the two sets of metadata: the time difference between the target events between the metadata sets is Δt = 150ms, which is less than the preset time difference threshold of 500ms; the spatial distance between the target events is Δd ≈ 42cm, which is less than the preset spatial distance threshold of 100cm; and the angle of motion of the target object is θ = 10°, which is less than the preset angle threshold of 45°. All three conditions are met, indicating it is a multi-view observation of the same target. The overall confidence score is calculated as: max(75,70) = 75, dynamic reward value = 20 × (1 - 10 / 45) ≈ 15.6, overall confidence score = 75 + 15.6 ≈ 91, which exceeds the preset cumulative confidence score threshold of 80. Therefore, it directly enters the confirmation state without needing to proceed to the accumulation process.

[0130] The arbitration end then outputs a mild wake-up command, activating the front camera to capture a 3-second short video at 720P / 15fps. The verification model confirms the target as a human figure, and the secondary verification confidence level is 88, exceeding the threshold of 80. It then enters the S3 (fully wake-up) state, waking up the main control SoC to start full-function HD recording.

[0131] The aforementioned spatiotemporal correlation mechanism also possesses false alarm suppression capabilities. Taking the aforementioned insect interference as an example, if two cameras each detect an insect, their time difference may be close, but if the spatial distance or the angle between their movement directions exceeds a threshold, the arbitrator will determine it as an unrelated event and will not trigger a wake-up call. Similarly, when two pedestrians pass behind a vehicle simultaneously, moving in opposite directions (with an angle of 110°), although they may be close in time and space, the condition of motion consistency is not met, and they will not be mistakenly identified as the same target.

[0132] When a target is continuously observed by only a single camera and a multi-camera correlation cannot be established, the arbitrator initiates a confidence accumulation process. Taking the detection of a slowly moving target at night by a later camera as an example, the single-frame confidence is 45, entering the accumulation interval. The event type is "humanoid," using a standard attenuation factor of 0.95. At the subsequent 1, 2, 3, and 4 seconds, the camera continuously detects the same target, with confidence levels of 48, 50, 52, and 55 respectively. After accumulation and updating, the accumulated confidence C_acc of the target event reaches 227.3 at the 4-second mark, exceeding the accumulated confidence threshold of 200, triggering a confirmation state. After secondary verification, the main controller is awakened. If the target only appears briefly (e.g., flashes by), the accumulated value will not reach the threshold due to attenuation, thus avoiding invalid wake-ups.

[0133] This application provides an example of in-vehicle multi-view stereo cross-validation and differential confidence accumulation.

[0134] For example, after the vehicle is turned off and locked, it is in a long-term parked armed state. The center console camera detected a suspected child target on the left rear seat with a confidence level of 60%, but from an oblique side view, it is difficult to distinguish the clothing on the seat from a real child. At the same time, the rear roof camera detected a three-dimensional target in the same area from a top-down view with a confidence level of 75%. The two cameras reported the following metadata: cam10 (child, 60, coordinates 15, -70, motion vector 0, 0), cam11 (child, 75, coordinates 18, -72, motion vector 0, 0).

[0135] The arbitration panel performs spatiotemporal and semantic correlation analysis on the two sets of metadata: the time difference of the target event Δt = 150ms, which is less than the in-vehicle time window threshold (preset time difference threshold) of 400ms; the spatial distance of the target event Δd ≈ 3.6cm, which is less than the in-vehicle space threshold (preset spatial distance threshold) of 30cm; the event_type of both cameras is "child", which is semantically consistent; the target is stationary, and the motion consistency condition is automatically relaxed. All three conditions and the semantic enhancement condition are met, and it is determined to be a multi-view stereoscopic observation of the same target. The overall confidence level = 75 + 30 = 105, exceeding the high confidence threshold of 80, and directly enters the confirmation state. After a slight wake-up, high-definition verification confirms it is a child, and a full wake-up is initiated, sending a child abandonment alarm.

[0136] If the rear seats only contain folded clothing, the overhead camera, with its top-down view, can observe that the target does not possess the three-dimensional height and body features of a human. Therefore, its reporting confidence level will be significantly lower than that of a child target, and the overall confidence level will not reach the threshold, thus avoiding false alarms.

[0137] When a target is observed by only a single camera with low confidence, the differential attenuation factor ensures that high-risk events reach the cumulative threshold more quickly. Continuing the scenario above, at night, there is a child partially covered by a blanket in the back seat. Only the rearview mirror camera can detect the target, with a confidence level of only 40-55 per frame, and no other cameras can establish a correlation. This target type is "child," triggering a differential attenuation factor of 0.98 (slower attenuation). Furthermore, the rear seat area is a preset high-priority area, reducing the cumulative threshold to 180.

[0138] The cumulative value starts at t=0s, C_acc=45; at t=2s, a second detection (confidence 48%) yields C_acc≈91.2; at t=4s, a third detection (confidence 50%) yields C_acc≈137.6; at t=6s, a fourth detection (confidence 52%) yields C_acc≈184.1, exceeding the threshold of 180, triggering state S2. Verification confirms the presence of a child, activating the alarm. If the target is a normal movement event, using a standard attenuation factor of 0.95, the cumulative value at t=6s under the same conditions is approximately 168.6, which has not yet reached the threshold, requiring a longer time to trigger. The differentiated attenuation mechanism advances the response time for high-risk events by approximately 2 seconds, which is significant for safety-related scenarios such as children being left behind.

[0139] This application provides an example of in-vehicle intrusion detection linked to door status.

[0140] For example, in a vehicle armed state, the driver's side window is broken, and an intruder climbs into the vehicle through the window. The footwell camera detects a "moving" and then a "human-shaped" target, with the confidence level gradually increasing from 55 to 78. The front roof camera then detects the same target, with a confidence level of 72. The arbitrator performs spatiotemporal correlation analysis: Δt = 200ms, satisfying the temporal correlation condition; Δd ≈ 25cm, satisfying the spatial correlation condition; θ = 12°, satisfying the motion consistency condition. The overall confidence level exceeds 80, and the system transitions to state S2.

[0141] At the same time, the arbitrator reads through the CAN bus that the left front window is unauthorized (the window position signal changes but the door lock status does not change, and there is no legitimate unlocking signal). Combined with the trajectory chain information from the footwell camera—the target enters the vehicle from the window position and moves towards the cabin—it is determined to be an illegal intrusion. The event confidence level is raised to the highest level, and the system directly enters S3 (fast track) from S2, fully waking up the main control SoC to start high-definition recording and sending an intrusion alarm through the vehicle network.

[0142] If the vehicle owner unlocks the door normally using the remote key, the CAN bus will detect a valid door unlock signal and a sequence of door lock status changes (unlock-open-enter-close-lock). In this case, even if the in-car camera detects a human figure, the arbitrator automatically suppresses the intrusion detection logic based on the presence of a valid door unlock signal, remaining idle and not triggering any alarms. This mechanism completely avoids false alarms when the owner returns to retrieve items normally. The distinction between valid and invalid access is automatically completed through the timing logic of the door lock status change sequence and window status changes, without requiring the user to manually switch modes.

[0143] This application provides an example of dynamic scene adaptation and parameter linkage adjustment.

[0144] For example, the arbitrator (arbitration end) automatically switches to global arbitration mode based on vehicle status signals such as vehicle speed, door lock status, and arming status, covering the entire life cycle of the vehicle from driving to parking.

[0145] When the vehicle is in motion, the arbitration mode switches to driving mode, the confidence threshold is reduced (the high threshold is reduced from 80 to 60), the time window is shortened to 200ms, the driver area camera is mainly used, the spatiotemporal correlation can be turned off, and the single camera is mainly used for rapid accumulation.

[0146] When temporarily stopping at a red light, the arbitration mode switches to temporary stopping mode, enhancing the sensitivity of the rear-seat surveillance system. The in-vehicle space threshold is tightened from 30cm to 20cm, and the child event attenuation factor is increased to 0.99 (due to the extremely high risk of a child being left alone in the car during a temporary stop). The normal event attenuation factor is adjusted to 0.97. If a pre-warning wake-up signal is received, the time window automatically tightens from 500ms to 200ms.

[0147] After the engine is turned off and the car is locked, the arbitration mode switches to the parking mode, and the anti-theft and residual detection functions are activated. Only the key view camera is retained, and unnecessary equipment is turned off to reduce power consumption. It is strongly correlated with the door / window status signal. The cumulative threshold of high-risk areas is reduced to 180, and the differential attenuation factor is enabled (0.98). At the same time, it is linked with the vibration sensor and activates the surrounding camera verification when abnormal vibration is detected.

[0148] Building upon this foundation, the arbitrator can further fine-tune based on environmental perception information and predicted wake-up signals. Taking predicted wake-up coordination as an example, if the vehicle's left-side camera detects a target motion vector pointing forward, the system predicts the target will soon enter the front camera's field of view and wakes the front camera in advance. Upon receiving the predicted wake-up flag, the arbitrator tightens the time correlation window from 500ms to 200ms. The left-side and front cameras subsequently report the same target metadata with a time difference of 180ms, satisfying the tightened window requirement and successfully entering spatial and motion correlation. The tightened window, while ensuring correct correlation, eliminates irrelevant events with large time deviations in the candidate set, improving correlation purity.

[0149] Taking collaborative environmental modeling as an example, the image semantic segmentation module identifies the left side of the vehicle as a wall-hugging area (extremely low risk). The arbitrator accordingly increases the spatial threshold (reduced from 100cm to 50cm) and cumulative threshold (increased from 200 to 300) of the camera on that side, reduces its wake-up priority, and effectively avoids false alarms caused by wall reflections and shadow changes.

[0150] Taking dynamic parameter adaptation as an example, the swaying of tree shadows causes the left camera to frequently generate motion detections with a confidence level of 40-50, occasionally accumulating to the threshold and causing false wake-ups. The arbitrator continuously calculates the false alarm rate. After detecting that the false alarm rate of the left camera is too high, it automatically increases its accumulation threshold to 250 and simultaneously increases the attenuation factor to 0.97. After the adjustment, the false alarm rate decreases significantly, and this adjustment only applies to the left camera independently and does not affect the sensitivity of other cameras.

[0151] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0152] Based on the same inventive concept, this embodiment also provides a parking monitoring system for implementing the above embodiments and preferred embodiments, which will not be repeated hereafter. The terms "module," "unit," "subunit," etc., used below refer to combinations of software and / or hardware that perform a predetermined function. Although the apparatus described in the following embodiments is preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0153] In one embodiment, Figure 4 This is a structural block diagram of a parking monitoring system provided in one embodiment of this application, as shown below. Figure 4 As shown, the parking monitoring system includes: a data acquisition terminal 42, an arbitration terminal 44, and a main control terminal 46;

[0154] The data acquisition terminal 42 includes multiple camera units installed at different locations, each camera unit being used to acquire images of each target area; using a preset lightweight model, target event detection is performed on the acquired images to determine the confidence level of each target event; when the confidence level of the detected target event is greater than a preset reporting threshold, metadata of the target event is generated and reported to the arbitration terminal 44; the preset reporting threshold is lower than the lower limit of the preset target interval.

[0155] Arbitration terminal 44 is used to execute the event arbitration method provided in the embodiments of this application;

[0156] The main control terminal 46 is used to record the target event in high definition and / or send alarm information when it receives the wake-up command sent by the arbitration terminal 44.

[0157] The aforementioned preset reporting threshold can be set according to specific application scenarios and needs, and this embodiment does not impose specific limitations. The goal is simply to report suspected target events using the preset reporting threshold. For example, the preset reporting threshold could be 30. Each camera unit is used to acquire images of each target area. Each camera unit can continuously acquire images of each target area according to a preset AOV (Always-On Video) frame interval. Specifically, the camera unit can continuously acquire images of the target area at an extremely low frame rate (e.g., 1-5 frames / second). Using this method, the camera unit is in a sleep state most of the time, only waking up momentarily at a preset acquisition time to complete the acquisition and processing of one frame, thereby achieving an average power consumption at the milliwatt or even microwatt level.

[0158] Furthermore, the aforementioned preset lightweight model is a lightweight artificial intelligence model (such as a trimmed version of backbone networks like MobileNet or TinyYOLO) running on the lightweight NPU (Neural Processing Unit) or MCU built into each camera unit. It should be noted that the aforementioned preset lightweight model can be specifically configured according to the specific application scenario and requirements. This embodiment does not impose specific limitations here, as long as the preset lightweight model can be used to detect target events in the acquired images and determine the confidence level of each target event. Additionally, the model inference frame rate of the aforementioned preset lightweight model is consistent with the AOV acquisition frame rate. The aforementioned preset lightweight model can detect various target event types such as movement, human figures, vehicles, collisions, children, and pets. The aforementioned report to the arbitration end can be submitted to the arbitration end on the policy engine MCU via a low-bandwidth bus (such as I2C or SPI).

[0159] It should be noted that the above modules can be functional modules or program modules, and can be implemented through software or hardware. For modules implemented through hardware, the above modules can reside in the same processor; or the above modules can be located in different processors in any combination.

[0160] In one embodiment, an electronic device is provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement any of the event arbitration methods described in the above embodiments.

[0161] In one embodiment, a computer-readable storage medium is provided having program instructions stored thereon, which, when executed by a processor, implement any of the event arbitration methods described in the above embodiments.

[0162] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM). The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0163] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0164] The embodiments described above are merely illustrative of several implementation methods of this application, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of this patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. An event arbitration method, characterized by, An arbitration terminal is used in a parking monitoring system, which further includes a data acquisition terminal; the data acquisition terminal includes multiple camera units installed at different locations, and the method includes: Acquire the different metadata reported by each camera unit of the data acquisition terminal; the metadata includes at least the confidence level of the target event; Perform correlation analysis on the different metadata obtained to determine whether the target events corresponding to the different metadata are the same target event; When the target events corresponding to different metadata are not the same target event, the target metadata with a confidence level within a preset target range is cached in the target queue, and the cumulative confidence level of the target events in each camera unit is determined based on the caching results; the target queue is the queue in the target camera unit that stores the target events corresponding to the target metadata; the target camera unit is the camera unit that reports the target metadata; the preset target range is the confidence range where there is a risk of missed reports or false wake-ups; A wake-up command is generated when the cumulative confidence value of the target event is greater than a preset cumulative confidence threshold.

2. The event arbitration method of claim 1, wherein, The step of performing correlation analysis on the acquired different metadata to determine whether the target events corresponding to the different metadata are the same target event includes: Calculate the spatiotemporal correlation between different metadata to obtain the spatiotemporal correlation results between different metadata; Based on the spatiotemporal correlation results between different metadata, it is determined whether the target events corresponding to different metadata are the same target event.

3. The event arbitration method of claim 2, wherein, The metadata also includes the time, location, and direction of movement of the target event; The calculation of the spatiotemporal correlation between different metadata, to obtain the spatiotemporal correlation results between different metadata, includes: Based on the occurrence time of the target event in different metadata, determine the time difference of the target event between different metadata. Determine the spatial distance between different metadata where the target event occurs, based on the location of the target event in different metadata. Based on the movement direction of the target object in different metadata, determine the angle between the movement directions of the target object in different metadata.

4. The event arbitration method of claim 3, wherein, The determination of whether the target events corresponding to different metadata are the same target event based on the spatiotemporal correlation results between different metadata includes: When the spatiotemporal correlation results between different metadata satisfy the preset criteria for determining the same target event, the target events corresponding to the different metadata are determined to be the same target event; otherwise, the target events corresponding to the different metadata are determined not to be the same target event. The preset criteria for determining the same target event are that the time difference between the target events occurring between different metadata is less than or equal to a preset time difference threshold, the spatial distance between the target events occurring between different metadata is less than or equal to a preset spatial distance threshold, and the angle between the motion directions of the target objects between different metadata is less than or equal to a preset angle threshold.

5. The event arbitration method of claim 1, wherein, The method further includes: When the target event corresponding to different metadata is the same target event, the overall confidence level of the target event is determined based on the confidence level of the target event in different metadata. A wake-up command is generated when the overall confidence level of the target event is greater than a preset overall confidence level threshold.

6. The event arbitration method of claim 5, wherein, When the target events corresponding to different metadata are the same target event, the overall confidence level of the target event is determined based on the confidence levels of the target event in the different metadata, including: When different metadata correspond to the same target event, the associated reward value of the target event is determined based on the angle between the movement directions of the target objects in the different metadata. The sum of the target confidence level and the associated reward value is determined as the comprehensive confidence level of the target event; the target confidence level is the maximum confidence level among all confidence levels corresponding to the target event.

7. The event arbitration method of claim 1, wherein, When the target events corresponding to different metadata are not the same target event, the target metadata with a confidence level within a preset target range is cached in the target queue, and the cumulative confidence level of the target events in each camera unit is determined based on the caching results, including: When the target events corresponding to different metadata are not the same target event, the target metadata with a confidence level in the preset target range is cached in the target queue to obtain the updated target queue; The cumulative confidence level of the target event corresponding to the target queue before the update is determined by multiplying the historical cumulative confidence level and the decay factor, and then adding the sum of the confidence levels of the target events in the target metadata. The cumulative confidence level of the target event corresponding to the updated target queue is then determined as the cumulative confidence level of the target event in the target camera unit. The decay factor is a parameter related to the risk level of the target event and / or the priority of the location where the target event occurs. When the target queue is not updated within a preset time period, the historical cumulative confidence level decays according to a preset decay rule.

8. The event arbitration method of claim 1, wherein, The method further includes: Based on the acquired vehicle status information, the arbitration mode is switched; the arbitration mode includes parking mode and temporary parking mode; different arbitration modes are configured with different target parameters; the target parameters include one or more of the following: preset time difference threshold, preset spatial distance threshold, preset cumulative confidence threshold, and attenuation factor.

9. A parking monitoring system characterized by comprising: The system includes a data acquisition terminal, an arbitration terminal, and a main control terminal; The data acquisition terminal includes multiple camera units installed at different locations, each camera unit being used to acquire images of various target areas; using a preset lightweight model, target event detection is performed on the acquired images to determine the confidence level of each target event; when the confidence level of a detected target event is greater than a preset reporting threshold, metadata of the target event is generated and reported to the arbitration terminal; the preset reporting threshold is lower than the lower limit of a preset target interval. The arbitration end is used to execute the event arbitration method according to any one of claims 1 to 8; The main control terminal is used to record the target event in high definition and / or send alarm information when it receives the wake-up command sent by the arbitration terminal.

10. An electronic device, comprising: The device includes a memory and a processor; The memory is connected to the processor and is used to store programs; The processor is configured to implement the event arbitration method as described in any one of claims 1 to 8 by running a program in the memory.

11. A computer readable storage medium having stored thereon program instructions, wherein, When the program instructions are executed by the processor, they implement the event arbitration method as described in any one of claims 1 to 8.