Early warning method and device, electronic equipment and storage medium

Abstract

The invention relates to the field of intelligent monitoring, and discloses an early warning method, which is suitable for a monitoring system, and the monitoring system comprises a sensor, a camera and a computing power analysis module. The method comprises the following steps: determining an initial weight and a position coordinate of each sensor; when the sensor detects an abnormal event, an alarm signal is sent; after receiving the alarm signal, the camera captures an image according to a preset strategy based on an image capture time sequence; inputting the picture group into a computing power analysis module, and selecting an image with the highest confidence coefficient; according to the picture confidence coefficient and the detection rate, whether the event reaches an early warning level is judged, and early warning information is uploaded to a system; and dynamically adjusting the weight of the trigger sensor according to the analysis result of the alarm event. According to the method, the sensor weight and the image capturing strategy are dynamically optimized, double evaluation of the confidence coefficient and the detection rate is combined, accurate judgment and response of the abnormal event are achieved, and the method has the advantages of being efficient, flexible and high in robustness and is suitable for monitoring requirements of various complex scenes.

Description

Technical Field

[0001] The present invention relates to the field of intelligent monitoring technology, and specifically to an early warning method, device, electronic equipment and storage medium. Background Art

[0002] With the widespread application of intelligent monitoring technology, monitoring systems have played an important role in security, industrial monitoring and other fields. Traditional monitoring systems usually rely on the collaboration of sensors and cameras to trigger alarms by monitoring abnormal conditions in the environment. However, the existing technology still has some shortcomings in practical applications, which affects the response efficiency and detection accuracy of the monitoring system.

[0003] In existing monitoring systems, the reliability of sensor-triggered alarm signals is low, and system operating efficiency is often reduced due to false alarms or missed alarms. On the one hand, the trigger priority of sensors is usually fixed, lacking a dynamic adjustment mechanism, making it difficult to timely optimize its weight allocation according to actual changes in the monitoring scene (such as sensor deployment location, risk area level, etc.). On the other hand, relying solely on simple trigger rules of sensors cannot fully consider multi-dimensional information such as the spatial location and time correlation of sensor signals, making it difficult for the system to accurately screen out truly important alarm events.

[0004] In addition, in the existing technology, cameras usually use fixed or simple time intervals to capture images, and fail to flexibly adjust the capture strategy based on the dynamic characteristics of abnormal events. This method not only leads to a waste of resources, but also may miss key images related to abnormal events, affecting the accuracy of event analysis and judgment.

[0005] Furthermore, existing image analysis methods mostly focus on processing a single image, ignoring the use of time series information and event correlation features in a group of images. In current technologies, event determination is mostly based on a single indicator (such as confidence), but lacks comprehensive analysis of multi-dimensional indicators (such as confidence, detection rate, etc.), which leads to easy misjudgment in complex scenes.

[0006] In addition, the static design of the sensor trigger mechanism limits the long-term adaptability of the system. Existing monitoring systems find it difficult to adjust the trigger priority and weight allocation of sensors based on event feedback or environmental changes, which causes the system to gradually lose robustness in dynamic scenarios and increases maintenance and operating costs.

[0007] In summary, the existing monitoring system still has many deficiencies in abnormal event detection and response, mainly manifested in the difficulty of dynamically adjusting the sensor trigger priority, the lack of flexibility of the camera capture strategy, the imperfect image analysis and event determination methods, and the lack of long-term adaptability and stability of the system. Therefore, there is an urgent need for an early warning method that can achieve efficient collaboration among sensors, cameras, and computing power analysis modules to solve the above defects in the existing technology. Summary of the invention

[0008] In view of the shortcomings of the prior art, the present invention provides an early warning method, device, electronic device and storage medium to solve the problems in the existing monitoring system that the sensor trigger priority is difficult to dynamically adjust, the camera capture strategy is inflexible, the event analysis is not accurate enough, and the long-term adaptability of the system is poor, thereby improving the accuracy and efficiency of abnormal event detection and response.

[0009] To achieve the above objectives, the present invention is implemented through the following technical solutions: an early warning method is applicable to a monitoring system, the monitoring system includes a sensor, a camera and a computing power analysis module, and the early warning method includes the following steps: Determine the initial weight and position coordinates of each sensor; When the sensor detects an abnormal event, it calculates the trigger priority according to the preset rules. If the trigger priority meets the threshold condition, an alarm signal is sent; After receiving the alarm signal, the camera captures images according to the calculated capture time sequence and the preset continuous capture strategy; the captured image group is input into the computing power analysis module, and the image with the highest confidence is selected according to the event relevance score of the image; based on the image confidence and detection rate, it is determined whether the event has reached the warning level, and the warning information is uploaded to the system; According to the analysis results of the alarm events, the weights of the triggered sensors are optimized and adjusted.

[0010] Preferably, the initial weight of the sensor is set according to the type of sensor, perception range, historical trigger records and its importance in the target monitoring area. The position coordinates of the sensor are pre-calibrated by physical position measurement during deployment or based on the coordinate system of the monitoring area. The position coordinates are used to calculate the spatial distance between the sensor and the target camera.

[0011] Preferably, the calculation of the trigger priority satisfies the following rules: The higher the importance weight of a sensor, the higher the trigger priority; The closer the sensor is to the camera, the higher the trigger priority; The shorter the last trigger interval, the lower the trigger priority.

[0012] Preferably, after the camera receives the alarm signal, the step of capturing images according to the calculated capture time sequence and a preset continuous capture strategy comprises: The camera's capture function is activated according to the alarm signal, and the capture time series is calculated according to the event characteristics; The snapshot time series is dynamically generated based on the priority and temporal relevance of the event status, and each time point is associated with the expected change of the event; Trigger the camera to capture images in time sequence, and adjust the camera's exposure parameters, clarity settings, and lighting compensation during capture to improve capture quality; The capture of multiple continuous images is completed within a preset time range, and the image group is passed as input to the computing power analysis module for subsequent processing.

[0013] Preferably, the step of inputting the captured image group into the computing power analysis module and selecting the image with the highest confidence according to the event relevance score of the image comprises: Receive the image group captured by the camera and extract the basic features of the image, including target clarity, object integrity and background interference; Based on the feature extraction results, each image is scored for event relevance, and the score is calculated comprehensively based on the clarity of the target in the image, the completeness of the object features, and the temporal correlation with the alarm event; Sort the event relevance scores of all images and select the image with the highest score as the image with the highest confidence; The image with the highest confidence level is used as the final basis for event determination.

[0014] Preferably, the step of determining whether the event reaches the warning level according to the image confidence and detection rate, and uploading the warning information to the system includes: Receive the image with the highest confidence and its corresponding confidence value transmitted by the computing power analysis module; Obtain the detection rate of the current event, which is based on the overall evaluation result of the computing power analysis module on the event correlation in the picture group; compare the confidence value with the preset confidence threshold, and at the same time compare the detection rate with the preset detection rate threshold; when the confidence value is higher than the confidence threshold and the detection rate is higher than the detection rate threshold, determine that the current event reaches the warning level, and generate warning information; The warning information including the warning event type, occurrence time, sensor location and the image with the highest confidence is uploaded to the system for processing or storage.

[0015] Preferably, the step of optimizing and adjusting the weight of the triggered sensor according to the analysis result of the alarm event includes: Get the actual effect score of the alarm event, which is evaluated based on the accuracy, missed alarm rate and false alarm rate of the event judgment; get the expected effect score of the trigger sensor, which is calculated based on the sensor type, historical trigger data and its importance in event detection; Compare the actual effect score with the expected effect score, and calculate the difference between the two to reflect the deviation degree of sensor triggering; adjust the weight of the triggering sensor according to the score difference. When the actual effect score is higher than the expected score, increase the sensor weight; when the actual effect score is lower than the expected score, reduce the sensor weight; Update the sensor weights to the system so that the optimization results are reflected in subsequent trigger priority calculations.

[0016] The present invention also provides an early warning device, comprising: Sensor module, used to monitor environmental events in real time and calculate trigger priorities; The camera module is used to receive the alarm signal from the sensor module and execute the image capture task according to the image capture optimization strategy; The computing power analysis module is used to analyze the captured image group, calculate the event confidence and select the image with the highest confidence; the event determination module is used to determine the event level based on the confidence and detection rate; The optimization feedback module is used to dynamically adjust the weights of sensors according to the analysis results of alarm events.

[0017] The present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the above method is implemented.

[0018] The present invention also provides a storage medium on which a computer program is stored. When the computer program is executed by a processor, the method described above is implemented.

[0019] The present invention provides an early warning method, device, electronic device and storage medium. It has the following beneficial effects: 1. The present invention takes into account the weight, location and trigger conditions of the sensor, dynamically calculates the trigger priority, and ensures that the alarm signal is sent with higher accuracy and reliability. By avoiding interference from irrelevant or low-importance trigger events, the false alarm rate and missed alarm rate can be effectively reduced, thereby significantly improving the accuracy of abnormal event detection.

[0020] 2. The present invention designs a capture strategy based on dynamic adjustment, which avoids redundant capture and resource waste by calculating the capture time series and image analysis priority, while ensuring that images at key time points are effectively captured. The computing power analysis module performs in-depth analysis on the image group and selects the key images with the highest confidence, thereby reducing the system computing power consumption and improving the overall resource utilization efficiency.

[0021] 3. The present invention dynamically optimizes sensor weights and continuously adjusts weight distribution according to historical data and real-time event feedback, so that the monitoring system can adapt to environmental changes (such as risk level, sensor status changes, etc.), thereby maintaining good performance in complex and changing scenarios. This mechanism can effectively improve the long-term stability and flexibility of the system.

[0022] 4. The early warning method of the present invention has good scalability and is suitable for multi-target monitoring and complex dynamic scenarios. For example, through multi-target analysis and collaborative optimization, it can efficiently capture multiple abnormal behaviors, and classify and prioritize different types of abnormal events (such as intrusion, fire, etc.), providing technical support for a wide range of practical applications.

[0023] 5. Through the dynamic adjustment of sensor weights and the introduction of system feedback mechanisms, the present invention can identify sensors with false alarms or performance degradation, and mark potential problem sensors during the weight optimization process, prompting operation and maintenance personnel to conduct inspections and maintenance in a timely manner. This mechanism reduces unnecessary resource consumption and manual intervention in system operation, and reduces overall maintenance and operating costs.

[0024] 6. The present invention adopts the dual judgment rules of confidence and detection rate, and realizes rapid response to key abnormal events on the basis of ensuring the accuracy of warning events. The warning information contains detailed contents such as event location, type, time and key images, supports efficient subsequent processing and decision-making response, and greatly improves the timeliness and practicality of the system's warning. BRIEF DESCRIPTION OF THE DRAWINGS

[0025] Figure 1 It is a schematic diagram of the method flow of the present invention; Figure 2 It is a schematic diagram of the structure of the device of the present invention; Figure 3 It is a schematic diagram of the computer device structure of the present invention.

[0026] Among them, 100, sensor module; 200, camera module; 300, computing power analysis module; 400, event determination module; 500, optimization feedback module. DETAILED DESCRIPTION

[0027] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0028] Please see attached Figure 1The present invention provides an early warning method, which is applicable to a monitoring system, wherein the monitoring system includes a sensor, a camera, and a computing power analysis module. Through reasonable division of labor, cooperation, and optimized calculation, it is possible to achieve efficient early warning of events, reduce system resource consumption, and improve the accuracy of early warning. The steps of the method of the present invention are described below in conjunction with the accompanying drawings.

[0029] like Figure 1 As shown, the early warning method may include the following steps: S1. Determine the initial weight and position coordinates of each sensor; S2, the sensor detects abnormal events and calculates the trigger priority; S3, camera captures image; S4,computing analysis of the image group; S5. Event level determination and warning information upload; S6. Optimize and adjust sensor weights.

[0030] As for step S1, in this embodiment, this step is used to determine the initial weight and position coordinates of each sensor to achieve reasonable deployment of the sensors in the monitoring system and provide necessary initial conditions for subsequent trigger priority calculations.

[0031] In some embodiments, the initial weight setting of the sensor is based on a comprehensive consideration of the following factors: sensor type, importance of the monitoring target area, historical trigger records, and sensor perception range.

[0032] Specifically, the types of sensors may include infrared sensors, sound sensors, and vibration sensors. Different types of sensors have different sensitivities and applicability in event detection. For example, infrared sensors are more sensitive to temperature anomaly detection, while sound sensors are more suitable for monitoring abnormal sounds in the environment. Based on the importance of sensor types, initial weights can be assigned to different sensors through preset rules. For example, infrared sensors in temperature-sensitive areas can be assigned higher initial weights, while sound sensors in low-risk areas can be assigned lower initial weights.

[0033] As an option, the importance of the target area monitored by the sensor is determined by the risk assessment results in the area. Exemplarily, the target area can be divided into high-risk area, medium-risk area and low-risk area, where the high-risk area includes places where abnormal events are prone to occur, such as crowded areas or areas where dangerous goods are stored. For sensors in these high-risk areas, their initial weights are usually set higher to increase their triggering priority.

[0034] It should be noted that the historical trigger records of the sensor are also an important basis for determining the initial weight. In one possible implementation, the system adjusts the initial weight by counting the number of event triggers during the past operation of the sensor and the accuracy of the trigger (i.e., the proportion of valid trigger events). For sensors with higher trigger accuracy, higher weights can be assigned; for sensors with more false triggers, their weights can be appropriately reduced.

[0035] In some embodiments, the sensing range of the sensor may also affect the initial weight. A sensor with a larger sensing range usually has the ability to cover multiple monitoring areas, so its initial weight may be higher than that of a sensor with a smaller sensing range. The specific sensing range may be set according to the hardware capability of the sensor or the actual deployment environment.

[0036] After determining the initial weight of the sensor, its position coordinates need to be calibrated. In a possible implementation, the position coordinates of the sensor can be manually measured by its physical installation position in the monitoring area, for example, calibrated in a two-dimensional plane as (x i ,y i ), or in three-dimensional space, it is labeled (x i ,y i ,z i ). As an option, the monitoring area can adopt a preset coordinate system, and the position of the sensor can be mapped into the coordinate system to facilitate subsequent calculations.

[0037] It is understandable that the spatial position of the sensor is an important basis for the subsequent calculation of the distance factor between the sensor and the camera. In some embodiments, the spatial distance between the sensor and the target camera can be calculated according to the following formula: Among them, (x cam ,y cam ) is the coordinate of the target camera, (x i ,y i ) are the coordinates of the sensor.

[0038] When three-dimensional coordinate calculation is required, the distance formula can be expanded to: This spatial distance will directly affect the calculation of subsequent trigger priorities.

[0039] As a technical extension, in some complex monitoring systems, the rapid calibration of sensor positions can be achieved through automated positioning technology. For example, using wireless sensor network technology, the relative position of the sensor is determined through signal strength or time delay measurement technology, and further combined with existing fixed reference points (such as camera positions) for absolute coordinate calibration.

[0040] It should be noted that the setting of initial weights and position coordinates has an important impact on the role of sensors in the monitoring system. By reasonably configuring weights and accurately calibrating positions, reliable data support can be provided for the dynamic calculation of trigger priorities and the accuracy of alarm signals in subsequent steps.

[0041] In some embodiments, for environments that require dynamic adjustment, such as situations where the sensor may be moved or the monitored area may change, the system can set up a periodic re-inspection mechanism to adjust the sensor's weight and location coordinate information by re-measuring the sensor position or analyzing the latest historical data, thereby further improving the flexibility and reliability of the system.

[0042] Regarding step S2, in this embodiment, when the sensor detects an abnormal event, the trigger priority is calculated according to the set rules, and based on the calculation result, it is determined whether to send an alarm signal to the camera to start the subsequent capture task.

[0043] In some embodiments, sensors are used to monitor physical quantities related to abnormal events in the environment, such as infrared sensors for monitoring temperature changes, sound sensors for monitoring sudden changes in noise levels, and vibration sensors for monitoring the vibration amplitude of the ground or objects. When the sensor detects that its physical quantity exceeds a set threshold, the event is recorded as an abnormal event.

[0044] It should be noted that the sensor's ability to detect abnormal events is directly limited by its hardware design and the thresholds set. As an option, the sensor's threshold can be adjusted according to the characteristics of the monitoring scenario. For example, a lower noise threshold can be set in a low-noise environment, while the temperature threshold can be appropriately increased in a high-temperature area. Such threshold settings can increase the sensor's sensitivity to environmental changes.

[0045] In a possible implementation, the calculation of the trigger priority comprehensively considers the importance weight of the sensor, the spatial distance between the sensor and the target camera, and the trigger frequency of the abnormal event. The specific calculation rules are as follows: Weight factor: The higher the importance weight of a sensor, the higher its trigger priority. For example, important sensors in high-risk areas are usually assigned higher weights.

[0046] Distance factor: The closer the spatial distance between the sensor and the camera, the higher its trigger priority.

[0047] Time interval factor: The shorter the time interval between a sensor and the last trigger event, the lower its trigger priority. Specifically, the impact of the time interval can be represented by a time decay factor, which is used to reduce the priority of frequent triggers.

[0048] Specifically, the trigger priority Pi The calculation formula is: Among them, w i is the importance weight of the sensor, d i is the distance between the sensor and the camera, t i is the time interval between the sensor and the last trigger time, and α, β and γ are adjustment coefficients used to balance the influence of different factors.

[0049] In some embodiments, the adjustment coefficients α, β and γ can be dynamically adjusted through system training. For example, the influence of various factors on the trigger priority can be analyzed through the data of historical trigger events, and the values ​​of these coefficients can be optimized to meet the needs of different monitoring scenarios.

[0050] It should be noted that the calculation result of the trigger priority will be consistent with the set priority threshold P threshold When P i ≥P threshold When the sensor generates an alarm signal and sends it to the system; otherwise, the current abnormal event is regarded as an invalid trigger and no alarm signal is sent. Such a priority determination mechanism can effectively filter out low-importance or meaningless trigger events and reduce the waste of system resources.

[0051] In a possible implementation, the alarm signal may also include identification information of the triggering sensor (such as sensor number), sensor position, and timestamp of the triggering event. For example, the sensor position may be used for camera positioning, and the timestamp may be used for synchronizing subsequent image capture tasks.

[0052] It is understandable that the calculation of the trigger priority is a key link in achieving accurate sending of alarm signals. In some embodiments, the trigger mechanism can be optimized by periodically updating the weights of the sensors or dynamically adjusting the priority thresholds. For example, when the environment in the monitored area changes (such as new risk factors or sensor position adjustments), the weights of the sensors can be reallocated to adapt to new monitoring needs.

[0053] It should be noted that the trigger priority calculation method of this embodiment has high flexibility, and the weights of the weights, distances, and time intervals can be adjusted according to the actual scene requirements to adapt to different monitoring environments. In some complex scenes, the calculation rules of the trigger priority can be further enriched by introducing more environmental factors (such as ambient temperature and humidity, light intensity, etc.) to improve the adaptability and accuracy of the system.

[0054] For step S3, in this embodiment, after the camera receives the alarm signal sent by the sensor, it performs the image capture task according to the calculated capture time sequence and the preset continuous capture strategy to capture key images related to the abnormal event and provide support for subsequent analysis and event judgment.

[0055] In some embodiments, after receiving the alarm signal, the camera first activates the working mode of the camera and adjusts the shooting parameters, including image resolution, exposure time, and lighting compensation, to ensure the clarity and quality of the captured image. For example, the camera can adjust the brightness through automatic gain control to adapt to changes in light intensity in the monitoring environment, such as increasing the exposure time at night or in shadow areas.

[0056] As an option, after the camera is activated, the system calculates the snapshot time series based on the timestamp of the alarm signal and the sensor position. In a possible implementation, the snapshot time series is generated based on the dynamic change characteristics of the event, aiming to capture the key moment of the event. Specifically, the snapshot moments in the time series can be determined by the following principles: 1. The snapshot time points must cover the entire time period of the event, including the initial time, duration, and end time of the event trigger.

[0057] 2. The time interval is dynamically adjusted according to the rate of change of the event. For example, for events that change quickly (such as movement or vibration in a short period of time), the time interval should be shortened appropriately; for events that change slowly (such as a slow rise in temperature), the time interval can be extended appropriately.

[0058] In a possible implementation, the optimization of the screenshot time series can be achieved through the Markov process model. Specifically, it is assumed that the state change of the abnormal event can be represented as a discrete time Markov chain, whose state set is {s1, s2, …, s n}, the state transition probability matrix is ​​P = [P (s j |s i )], where P(s j |s i ) indicates that the event is from state s i Transfer to state s j probability.

[0059] The optimization goal of the screenshot time series is to maximize the utility function U(s j ,t), the specific optimization formula is: in: U(s j ,t) means in state s j The utility value of the screenshot at time t; Utility value U(sj ,t) comprehensively considers the image quality Q(t) and the capture cost C(t), and its calculation formula is: U(s j ,t)=к·Q(t)-λ·C(t) Among them, к and λ are adjustment coefficients used to balance image quality and capture cost.

[0060] Specifically, the image quality Q(t) can be evaluated based on indicators such as clarity and lighting conditions, such as using edge detection algorithms to evaluate image clarity or using brightness histograms to evaluate lighting uniformity. The capture cost C(t) includes the energy consumption of the camera, computing power usage, and network bandwidth and other resource consumption.

[0061] It should be noted that the optimized capture time sequence can ensure that the camera captures images at key time points, thereby avoiding redundant capture and waste of resources.

[0062] In some embodiments, after generating the image capture time sequence, the camera executes the image capture task in sequence. As an option, the total number of consecutive image captures can be calculated based on the duration and time interval of the event, and the specific calculation formula is: in: T duration is the duration of the event; T interval The time interval between consecutive captures.

[0063] In one possible implementation, to further enhance the flexibility of capturing images, the camera can also adjust the shooting angle in combination with a motion tracking algorithm. For example, when an event involves a moving target, the camera can adjust the field of view in real time based on the target's position information to ensure that the captured image always includes the target.

[0064] It can be understood that the capture strategy of this embodiment can efficiently capture key images during the event by optimizing the time sequence and dynamically adjusting the shooting parameters. In addition, for different types of abnormal events, such as sudden events or continuous events, the capture parameters and strategies of the camera can be dynamically adapted to meet the diverse needs of the monitoring system.

[0065] It should be noted that in some complex scenes (such as multi-person sports scenes or multi-target monitoring scenes), the camera can also combine multi-target detection algorithms to analyze images in real time and dynamically adjust the capture strategy based on the analysis results. For example, in crowded areas, only the target area where abnormal behavior is detected is captured to reduce unnecessary resource consumption.

[0066] In some embodiments, to ensure the integrity of the captured image group, the camera can also perform simple post-processing operations on the image after the image is captured, such as image denoising, contrast enhancement or edge sharpening, to further improve the image analysis effect. The processed image group will be passed to the computing power analysis module for subsequent event analysis and early warning judgment.

[0067] For step S4, in this embodiment, the image group captured by the camera is input into the computing power analysis module, and the image with the highest confidence is selected according to the event relevance score of the image for subsequent event determination. The core of this step is to extract and comprehensively analyze the features of the image and select the image that best reflects the characteristics of the abnormal event.

[0068] In some embodiments, the computing power analysis module first receives a group of images transmitted by the camera and preprocesses each image to extract key features. The preprocessing steps include operations such as image format conversion, size normalization, denoising, and contrast enhancement. As an option, denoising can be based on a Gaussian filtering algorithm to reduce noise interference in the image, while contrast enhancement can use a histogram equalization method to improve the visual effect of the image and the accuracy of feature extraction.

[0069] Specifically, the computing power analysis module scores the event relevance of each image based on the clarity of the target in the image, the integrity of the object, and the temporal correlation with the abnormal event. In some embodiments, the target clarity is quantified by edge detection algorithms or gradient amplitude statistics. For example, the proportion of gradient change areas in the image can be calculated, and a higher gradient amplitude generally indicates that the target edge is clear.

[0070] In one possible implementation, object integrity can be evaluated by an object detection algorithm. The object detection algorithm can be based on a deep learning model, such as a convolutional neural network (CNN) or an object detection framework (such as YOLO or FasterR-CNN). By detecting the object boundary in the image, calculating the area ratio of the object to the image or the complete boundary of the object, the object is evaluated to see whether it is complete.

[0071] It should be noted that the temporal correlation with the abnormal event is another important factor in the image relevance score. As an option, the system will assign a higher relevance weight based on the match between the image capture time and the sensor alarm time. For example, the closer the image is to the alarm time, the higher its temporal correlation with the event is usually.

[0072] In this embodiment, the event correlation score C(I i ) is calculated based on the weighted sum of multiple features of the image, and its calculation formula is as follows: in: f k (I i ) is picture I i The k-th feature score of , including clarity score, completeness score and temporal relevance score; ω k Feature weights are used to balance the importance of different features in scoring. These weights can be set through historical data training or experience.

[0073] As an option, the feature weight ω k The specific value of can be adjusted dynamically according to different monitoring scenarios. For example, in a low-light environment, the weight of the clarity score can be appropriately increased to highlight the requirements for image quality; while in a high-dynamic environment, the weight of the temporal correlation score can be increased to better capture the behavioral characteristics of the target.

[0074] In some embodiments, the computing power analysis module sorts the relevance score results of the image group and selects the image with the highest score. max This picture best reflects the key features of an abnormal event and usually contains a clear target object and a complete event status.

[0075] It should be noted that in order to further improve the analysis accuracy, the computing power analysis module can combine multiple high-scoring images for joint analysis. For example, when an event involves a complex dynamic scene (such as multi-target abnormal behavior), the target features of multiple images can be aggregated to generate a comprehensive score to more comprehensively reflect the event status.

[0076] It is understandable that the image analysis method in this embodiment is applicable to both single-target monitoring scenarios and multi-target complex scenarios. In some special scenarios, for example, if there are many interference factors (such as shadows or reflections) in the monitoring area, the system can also use the background modeling algorithm to eliminate the influence of non-target areas on the relevance score.

[0077] In one possible implementation, the computing power analysis module can also optimize the feature weights of the correlation score through a feedback mechanism. For example, for historical event data, analyze whether the highest-scoring images effectively reflect the essential characteristics of abnormal events, and adjust the feature weights based on the analysis results, thereby continuously improving the robustness and adaptability of the scoring model.

[0078] It should be noted that the image analysis step of this embodiment fully utilizes computing resources to perform in-depth analysis on captured images to ensure that the selected key images have high relevance and high confidence. This process provides a reliable foundation for subsequent event determination and warning level classification, and also lays a data support for the efficient operation of the monitoring system.

[0079] For step S5, in this embodiment, based on the confidence and detection rate of the image, it is determined whether the event reaches the warning level, and the warning information is uploaded to the system. This step aims to accurately determine the importance and urgency of the event through comprehensive evaluation of multiple indicators, thereby triggering the subsequent processing flow.

[0080] In some embodiments, after the computing power analysis module completes the relevance scoring and confidence selection of the image group, the image with the highest score I max and its corresponding confidence value C max It is passed to the event determination module. It should be noted that the confidence C of the image max It is the result of comprehensive multi-dimensional feature calculation, indicating the reliability of the image in reflecting abnormal events.

[0081] As an option, the event determination module also obtains the detection rate calculated by the computing power analysis module. Specifically, the detection rate R det It is the ratio of the number of pictures related to the event to the total number of pictures in the picture group, and its calculation formula is: in: N rel is the number of pictures related to the event; N total is the total number of picture groups.

[0082] It is understandable that the confidence level C max and detection rate R det These are two key indicators for determining whether an event has reached the warning level. Specifically, if and only if both meet the preset threshold conditions, the system will identify the event as an abnormal event that requires warning.

[0083] In a possible implementation, the warning level may be determined by the following rules: Judgment condition: C max ≥C threshold And R det ≥R threshold in: C threshold is the confidence threshold; R threshold is the detection rate threshold.

[0084] As an option, the threshold C threshold and R threshold It can be adjusted dynamically according to the security level of the monitored area. For example, in high-risk areas, the threshold can be appropriately lowered to increase the capture rate of abnormal events; while in low-risk areas, the threshold can be increased to reduce the false alarm rate.

[0085] In some embodiments, when an event reaches a warning level, the system generates a warning report containing detailed information. Exemplarily, the warning information includes the following: Event type: The event classification result output by the computing power analysis module, such as fire, intrusion, or equipment failure.

[0086] Occurrence time: The timestamp when the sensor triggers the abnormal event.

[0087] Event location: The geographic coordinates of the sensor or camera.

[0088] Key image: the image with the highest confidence I max and its associated relevance score.

[0089] Confidence and detection rate: represent the importance of the event and the coverage of the image analysis results respectively.

[0090] It should be noted that the structured design of early warning information can effectively support the subsequent processing flow of the system, such as alarm, storage or operator notification.

[0091] In one possible implementation, the way to upload the warning information to the system may include wired network transmission or wireless communication. Specifically, in the wireless transmission scenario, data transmission can be based on a cellular network (such as 4G / 5G) or a low-power wide area network (such as LoRa or NB-IoT) to ensure that the warning information can be quickly delivered to the upper system.

[0092] It is understandable that in order to improve the flexibility of event determination, the system can also introduce a feedback mechanism. For example, when the operator manually confirms the warning event, the confirmation result can be fed back to the computing power analysis module to further optimize the calculation model of confidence and detection rate.

[0093] In some embodiments, in order to reduce the waste of system resources caused by false alarms, the system will also record events when they do not reach the warning level, but only store basic information without triggering an alarm. For example, low-confidence events can be saved as system logs for subsequent statistical analysis or model training.

[0094] It should be noted that this embodiment achieves accurate determination and efficient early warning of abnormal events by comprehensively analyzing the confidence and detection rate of images. The flexible adjustment of determination rules and thresholds further enhances the system's ability to adapt to complex scenarios, while reducing the waste of resources caused by false positives or false negatives.

[0095] For step S6, in this embodiment, the weights of the triggered sensors are optimized and adjusted according to the analysis results of the alarm events. By analyzing the actual effects of the alarm events, the weight distribution of the sensors is dynamically optimized to improve the overall response efficiency and early warning accuracy of the system.

[0096] In some embodiments, after the event determination is completed, the system will record the analysis results of the alarm event. Specifically, the analysis results include indicators such as the accuracy of the event determination (such as whether it is a valid event), the false alarm rate and the missed alarm rate of the event. Exemplarily, the accuracy can be obtained through the manual confirmation results of the operator, and the false alarm rate and the missed alarm rate can be calculated by counting the inconsistency between the sensor trigger signal and the actual event occurrence.

[0097] It should be noted that the core of sensor weight adjustment is to compare the weight with the actual effect and dynamically correct it through optimization rules. In a possible implementation, the sensor weight adjustment formula is as follows: in: is the adjusted sensor weight; is the sensor weight before adjustment; S actual Score the actual effect of the alarm event; S expected Score the expected performance of the sensor; α is the adjustment coefficient, which is used to control the amplitude of weight adjustment.

[0098] As an option, the actual effect score S actual The following indicators can be combined for calculation: Trigger accuracy: the degree of match between the trigger signal and the actual event.

[0099] Trigger coverage: The capture rate of events within the sensor's monitoring range.

[0100] Signal strength: The magnitude of the change in physical quantity caused by an abnormal event, such as temperature change, vibration amplitude, etc.

[0101] Specifically, the expected effect score S expected It is calculated based on the initial weight of the sensor, the importance of the monitoring target area, and the historical records. For example, sensors in high-risk areas usually have higher expected scores, while sensors with more historical false alarms have relatively lower expected scores.

[0102] In a possible implementation, the weight adjustment rules can be optimized for different scenarios. For example: When S actual >S expected , the system increases the weight of the sensor to enhance its influence in the priority calculation; When S actual <Sexpected When the sensor is triggered, the system will reduce the weight of the sensor to reduce the impact of its false triggering on the system.

[0103] It should be noted that the weight adjustment is not completed in one go, but is achieved through multiple iterations. The system can dynamically optimize the adjustment coefficient α in the weight adjustment formula based on the statistical results of multiple events. For example, in an environment with frequent false alarms, the value of α can be appropriately reduced to reduce the amplitude of the weight adjustment, thereby avoiding excessive fluctuations.

[0104] In some embodiments, in order to improve the real-time performance of weight adjustment, the system can introduce a weight update mechanism based on time decay. Specifically, for historical events with a long time span, their actual effect score S actual The weight will gradually decrease. Its calculation formula is: S actual =S actual ·e -λt in: λ is the time decay coefficient; t is the time interval between the event occurrence and the current moment.

[0105] In another possible implementation, the system can combine the results of weight optimization with the self-diagnosis function of the sensor. For example, when the weight of a sensor continues to drop to a set minimum threshold, the system can mark the sensor as abnormal and send a notification to the maintenance personnel. Exemplarily, the self-diagnosis function can detect the hardware status of the sensor (such as battery power, signal strength) or environmental adaptability (such as the impact of temperature and humidity changes).

[0106] It is understandable that this embodiment not only optimizes the priority allocation of sensors in the system through dynamic adjustment of sensor weights, but also can gradually improve the overall performance of the early warning system. In some embodiments, the weight adjustment mechanism can be further extended to weight optimization of multi-sensor collaboration. For example, for multiple sensors covering the same monitoring area, the system can automatically adjust their collaboration mode according to the weight allocation strategy to ensure efficient use of sensor resources.

[0107] It should be noted that the weight adjustment method of this embodiment has strong adaptability and can optimize system parameters in real time according to environmental changes and event feedback, and is particularly suitable for complex and dynamic monitoring scenarios. This dynamic optimization mechanism provides stability and flexibility for the long-term operation of the early warning system.

[0108] In a preferred embodiment of the present invention, the early warning method not only relies on sensors, but can also be optimized in combination with the actual installation environment and monitoring scenarios of the camera. For example, in an industrial scenario, cameras are usually installed at the entrances and exits to monitor whether workers at the entrances and exits are wearing safety helmets. However, due to poor camera installation positions or the dynamic characteristics of the monitoring screen (for example, people passing through the entrances and exits quickly, causing the portrait in the screen to flash by), the camera may not be able to accurately capture clear images, or even miss reports.

[0109] In order to solve this problem, a fence or other blocking facilities can be installed in the door area, and the physical structure design can be used to guide people to stop or slow down at the entrance and exit, so that the camera can capture the picture under more stable conditions, improving the image clarity and recognition accuracy. At the same time, combined with the method of the present invention, the camera can still intelligently start the capture task through the trigger mechanism, and continuously capture multiple pictures when people pass by according to the optimized capture strategy, and then send the pictures to the computing power analysis module for confidence evaluation and event level determination.

[0110] By combining physical facilities with intelligent early warning methods, this preferred embodiment significantly reduces the false alarm rate caused by blurred images or failure to capture key images, while further improving the adaptability of the camera installation environment. This method is particularly suitable for areas with dense traffic or complex monitoring conditions, and can effectively improve the practicality and reliability of the early warning system.

[0111] In general, the present invention realizes efficient and accurate abnormal event detection and response by dynamically calculating trigger priority, optimizing capture strategy, deeply analyzing picture groups and comprehensively determining event warning levels. The method of the present invention enables the monitoring system to adapt to complex and changing environmental requirements by dynamically adjusting the sensor weights, while improving resource utilization and system robustness. The present invention can not only reduce false alarms and missed alarms, improve the accuracy of early warnings, but also has good scalability and is applicable to a variety of complex scenarios, providing an efficient and reliable solution for the field of intelligent monitoring.

[0112] The early warning device described below and the early warning method described above can be referred to in correspondence with each other.

[0113] Please see attached Figure 2 The present invention also provides an early warning device, including: The sensor module 100 is used to monitor environmental events in real time and calculate trigger priorities; The camera module 200 is used to receive the alarm signal of the sensor module and execute the image capture task according to the image capture optimization strategy; the computing power analysis module 300 is used to analyze the captured image group, calculate the event confidence and select the image with the highest confidence; An event determination module 400 is used to determine the event level according to the confidence level and the detection rate; The optimization feedback module 500 is used to dynamically adjust the weight of the sensor according to the alarm event analysis result.

[0114] The device of this embodiment can be used to execute the above method embodiment, and its principles and technical effects are similar, which will not be repeated here.

[0115] Please see attached Figure 3 The present invention further provides a computer device 40, comprising: a processor 41 and a memory 42, wherein the memory 42 stores a computer program executable by the processor, and when the computer program is executed by the processor, the above method is executed.

[0116] The present invention further provides a storage medium 43 on which a computer program is stored. When the computer program is run by the processor 41, the above method is executed.

[0117] Among them, the storage medium 43 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable red-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, disk or optical disk.

[0118] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An early warning method, applicable to a monitoring system, wherein the monitoring system includes a sensor, a camera and a computing power analysis module, characterized in that: The early warning method comprises the following steps: Determine the initial weight and position coordinates of each sensor; When the sensor detects an abnormal event, it calculates the trigger priority according to the preset rules. If the trigger priority meets the threshold condition, an alarm signal is sent; After the camera receives the alarm signal, it captures images according to the calculated capture time sequence and the preset continuous capture strategy; The captured image group is input into the computing power analysis module, and the image with the highest confidence is selected based on the event relevance score of the image; Determine whether the event has reached the warning level based on the image confidence and detection rate, and upload the warning information to the system; According to the analysis results of the alarm events, the weights of the triggered sensors are optimized and adjusted.

2. The early warning method according to claim 1, characterized in that: The initial weight of the sensor is set according to the type of sensor, perception range, historical trigger records and its importance in the target monitoring area. The position coordinates of the sensor are pre-calibrated by physical position measurement during deployment or based on the coordinate system of the monitoring area. The position coordinates are used to calculate the spatial distance between the sensor and the target camera.

3. The early warning method according to claim 1, characterized in that: The calculation of the trigger priority satisfies the following rules: The higher the importance weight of a sensor, the higher the trigger priority; The closer the sensor is to the camera, the higher the trigger priority; The shorter the last trigger interval, the lower the trigger priority.

4. The early warning method according to claim 1, characterized in that: After the camera receives the alarm signal, the camera captures images according to the calculated capture time sequence and a preset continuous capture strategy, including: The camera's capture function is activated according to the alarm signal, and the capture time series is calculated according to the event characteristics; The snapshot time series is dynamically generated based on the priority and temporal relevance of the event status, and each time point is associated with the expected change of the event; Trigger the camera to capture images in time sequence, and adjust the camera's exposure parameters, clarity settings, and lighting compensation during capture to improve capture quality; The capture of multiple continuous images is completed within a preset time range, and the image group is passed as input to the computing power analysis module for subsequent processing.

5. The early warning method according to claim 1, characterized in that: The step of inputting the captured image group into the computing power analysis module and selecting the image with the highest confidence according to the event correlation score of the image comprises: Receive the image group captured by the camera and extract the basic features of the image, including target clarity, object integrity and background interference; Based on the feature extraction results, each image is scored for event relevance, and the score is calculated comprehensively based on the clarity of the target in the image, the completeness of the object features, and the temporal correlation with the alarm event; Sort the event relevance scores of all images and select the image with the highest score as the image with the highest confidence; The image with the highest confidence level is used as the final basis for event determination.

6. The early warning method according to claim 1, characterized in that: The step of determining whether the event reaches the warning level according to the image confidence and detection rate, and uploading the warning information to the system includes: Receive the image with the highest confidence and its corresponding confidence value transmitted by the computing power analysis module; Get the detection rate of the current event, which is based on the overall evaluation result of the event correlation in the image group by the computing power analysis module; Comparing the confidence value with a preset confidence threshold, and at the same time comparing the detection rate with a preset detection rate threshold; When the confidence value is higher than the confidence threshold and the detection rate is higher than the detection rate threshold, it is determined that the current event has reached the warning level and a warning message is generated; The warning information including the warning event type, occurrence time, sensor location and the image with the highest confidence is uploaded to the system for processing or storage.

7. The early warning method according to claim 1, characterized in that: The step of optimizing and adjusting the weight of the triggered sensor according to the analysis result of the alarm event includes: Obtain the actual effect score of the alarm event, which is evaluated based on the accuracy, missed alarm rate and false alarm rate of event judgment; Obtain an expected effectiveness score for a trigger sensor, the expected score being calculated based on the sensor type, historical trigger data, and its importance in event detection; Compare the actual effect score with the expected effect score, and calculate the difference between the two to reflect the degree of deviation of sensor triggering; According to the score difference, adjust the weight of the trigger sensor. When the actual effect score is higher than the expected score, increase the sensor weight; when the actual effect score is lower than the expected score, reduce the sensor weight; Update the sensor weights to the system so that the optimization results are reflected in subsequent trigger priority calculations.

8. An early warning device, applied to the early warning method according to any one of claims 1 to 7, characterized in that: include: Sensor module, used to monitor environmental events in real time and calculate trigger priorities; The camera module is used to receive the alarm signal from the sensor module and execute the image capture task according to the image capture optimization strategy; The computing power analysis module is used to analyze the captured image group, calculate the event confidence and select the image with the highest confidence; An event determination module is used to determine the event level based on confidence and detection rate; The optimization feedback module is used to dynamically adjust the weights of sensors according to the analysis results of alarm events.

9. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, it implements the early warning method according to any one of claims 1 to 7.

10. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the early warning method according to any one of claims 1 to 7 is implemented.

Images