Artwork display cabinet anti-theft and early warning system based on multi-modal sensor data fusion

Abstract

The invention relates to the technical field of anti-theft early warning, in particular to a multi-modal sensor data fusion artwork display cabinet anti-theft and early-warning system, which comprises an extraction module, an early-warning module and an early-warning module, and is characterized in that the extraction module is used for extracting the characteristic and trigger time relation of time sequence data of various sensors; the first matching module is used for matching the relation between the features and the triggering time with the relation between the features and the triggering time of the typical invasion scene, and if matching succeeds, the invasion is typical invasion and corresponding early warning is triggered; the identification and compensation module is used for identifying and compensating an abnormal weak signal from the sensing feature sequence if the matching is unsuccessful; the second matching module is used for constructing a signal triggering time sequence based on the compensated abnormal weak signal, matching the signal triggering time sequence with a standard signal triggering time sequence to obtain a matching distance sequence, and further calculating a matching degree; and the early warning module is used for executing graded early warning based on the matching degree and the historical visitor flow rate state of the exhibition hall. The reliability of the artwork showcase security and protection system can be greatly improved.

Description

Technical Field

[0001] This invention relates to the field of anti-theft and early warning technology, specifically to an anti-theft and early warning system for art display cabinets based on multimodal sensor data fusion. Background Technology

[0002] As the core protective carrier of artworks in exhibition settings, display cases face increasingly stringent requirements for their anti-theft early warning systems. On the one hand, many museum artworks are unique pieces, and damage or theft would cause irreparable losses, thus placing extremely high demands on the accuracy and timeliness of the early warning system. On the other hand, exhibition hall environments exhibit typical dynamic characteristics, with pedestrian traffic fluctuating in a pattern of "high density at peak times and sparseness at troughs," while also facing various environmental disturbances (such as visitor touch, construction vibrations, and airflow changes) and covert intrusion methods (such as laser breaching and slow movement). Traditional security solutions struggle to effectively balance the contradiction between "anti-interference" and "preventing missed detections," necessitating the development of intelligent early warning technologies adapted to the characteristics of exhibition hall environments.

[0003] Currently, anti-theft alarms for art display cases mainly employ a technical solution of "multi-modal sensor acquisition combined with simple data fusion." This involves deploying various sensors such as vibration, ultrasound, and pressure sensors to collect environmental data, and then identifying anomalies based on fixed thresholds or general statistical characteristics (such as mean and variance). Some systems introduce basic temporal constraints or static weight adjustment mechanisms. However, existing technologies still have significant shortcomings: fixed thresholds are difficult to adapt to dynamic interference caused by changes in pedestrian flow; generalized features cannot effectively capture the signal characteristics of concealed intrusion behaviors; rigid temporal rules are prone to missing non-standard intrusion patterns; and the systems generally lack compensation mechanisms for weak signals and the ability to dynamically correlate with pedestrian flow characteristics. These problems result in persistently high false alarm and false negative rates for existing solutions, failing to meet the actual needs of high-reliability security. Summary of the Invention

[0004] To address the technical problem of insufficient reliability in the security systems of existing art display cases, the present invention aims to provide an anti-theft and early warning system for art display cases based on multimodal sensor data fusion. The specific technical solution adopted is as follows: An extraction module is used to extract features from the time-series data of various sensors deployed in the art display case and the trigger time relationship between the features; The first matching module is used to match the features and trigger time relationship with the features and trigger time relationship of typical intrusion scenarios in the preset causal chain library. If the match is successful, it is determined to be a typical intrusion and a corresponding warning is triggered. The identification and compensation module is used to identify abnormal weak signals from the sensing feature sequence if the matching fails, and to calculate their cumulative feature values ​​for compensation. The second matching module is used to sort the compensated abnormal weak signal and other sensing trigger signals in chronological order to construct a signal trigger timing sequence, and match it with a preset standard signal trigger timing sequence to obtain a matching distance sequence. The matching distance sequence is then corrected using the cumulative feature value, and the matching degree is calculated based on the corrected matching distance sequence. The early warning module is used to assess atypical intrusion risks based on the matching degree, divide the traffic flow into intervals based on the historical traffic flow status of the exhibition hall, and execute graded early warnings based on the atypical intrusion risks and the traffic flow intervals.

[0005] Furthermore, the characteristics of the time-series data include: spectral acceleration within a preset frequency band in the vibration signal, micro-variables in the pressure signal, distance change rate or distance change in the ultrasonic signal, opening angle in the door magnetic signal, and temperature gradient in the infrared signal.

[0006] Furthermore, the triggering time relationship includes: When the typical intrusion scenario is a knocking demolition, after the vibration signal is triggered, the pressure signal becomes abnormal within the first preset time interval. After the pressure signal becomes abnormal, the ultrasonic signal changes abruptly within the second preset time interval. The entire triggering process of the above signals is completed within the first preset total time window. When a typical intrusion scenario is the unauthorized opening of a cabinet door, after the door magnetic signal is triggered, the infrared signal becomes abnormal within the third preset time interval. After the infrared signal becomes abnormal, the ultrasonic signal undergoes a sudden change within the fourth preset time interval. The entire triggering process of the above signals is completed within the second preset total time window. When a typical intrusion scenario is the violent movement of a display case, after the pressure signal is triggered, the ultrasonic signal becomes abnormal within the fifth preset time interval. After the ultrasonic signal becomes abnormal, the vibration signal undergoes a sudden change within the sixth preset time interval. The entire triggering process of the above signals is completed within the third preset total time window.

[0007] Furthermore, the process of acquiring the abnormal weak signal includes: Based on the anomaly detection algorithm, candidate weak signals that exceed the normal range are identified from the sensing feature sequence; The average feature value of the candidate weak signal is obtained, and the feature value deviation of the candidate weak signal is calculated by combining the upper quartile and standard deviation of the sensing feature sequence. Calculate the Pearson correlation coefficient between the candidate weak signal and the sensing feature sequences of other modes, construct the cooperability coefficient matrix, and obtain its average cooperability coefficient. The degree of anomaly of the candidate weak signal is calculated based on the eigenvalue deviation and the average cooperability coefficient. When the degree of abnormality is greater than the degree of abnormality threshold, the candidate weak signal is the abnormal weak signal.

[0008] Furthermore, the compensation process for the abnormal weak signal includes: The weight of the abnormally weak signal is determined based on its duration. The cumulative feature value is calculated based on the weights and the feature values ​​of the abnormal weak signal. The cumulative feature value is used as a compensation coefficient to enhance the average feature value of the abnormal weak signal, thereby obtaining the compensated feature value of the abnormal weak signal.

[0009] Furthermore, the process of obtaining the matching distance includes: The matching distance sequence is obtained by matching the signal trigger timing sequence with the standard signal trigger timing sequence using a dynamic time warping algorithm.

[0010] Furthermore, the process of obtaining the matching degree includes: Calculate the average value of the matching distances contained in the matching distance sequence, and perform reverse normalization on the average value to obtain the matching degree.

[0011] Furthermore, assessing atypical intrusion risks based on the matching degree includes: When the matching degree is greater than or equal to the first preset matching value and less than the second preset matching value, the atypical intrusion risk is low risk; When the matching degree is greater than or equal to the second preset matching value and less than the third preset matching value, the atypical intrusion risk is medium risk; When the matching degree is greater than or equal to the third preset matching value, the atypical intrusion risk is high risk.

[0012] Furthermore, the process of dividing the pedestrian flow area includes: Collect historical, cyclical visitor traffic data for the entire day in the exhibition hall at preset time intervals; Calculate the average pedestrian flow for each time period over multiple historical periods to form a mean sequence; The mean sequence is divided using the maximum threshold segmentation method to obtain high-traffic and low-traffic intervals.

[0013] Furthermore, the tiered early warning system based on the aforementioned atypical intrusion risks and the aforementioned pedestrian traffic zones includes: If the current period is a high-traffic area, and the atypical intrusion risk is low, a light warning will be triggered; if the atypical intrusion risk is medium, a light warning will be triggered and an alarm sound will be added; if the atypical intrusion risk is high, a network alarm will be activated and an alarm signal will be sent to a preset security terminal. If the current period is in a low-traffic area, and the atypical intrusion risk is low, a light warning will be triggered and an alarm sound will be added. If the atypical intrusion risk is medium or high, a network alarm will be activated and an alarm signal will be sent to a preset security terminal.

[0014] The present invention has the following beneficial effects: First, the features and trigger time relationships acquired by the extraction module form the data foundation for subsequent analysis. Second, the first matching module matches the features and trigger time relationships with a typical intrusion scenario library, i.e., a causal chain library. If successful, an early warning is triggered directly, achieving efficient response. Next, if no match is found, the identification compensation module is activated, responsible for identifying and amplifying abnormal weak signals that may represent atypical intrusions. Then, the second matching module uses the compensated signal to construct and compare signal trigger time sequences, calculating the corrected matching degree to quantify the probability of atypical intrusion. Finally, the early warning module integrates the matching degree with pedestrian flow information to execute dynamic hierarchical early warnings.

[0015] In summary, this invention significantly improves the accuracy of the art display case security system in identifying typical and atypical intrusions by constructing a causal chain library and using temporal fuzzy matching technology, effectively reducing false alarms under environmental interference and missed alarms under covert attacks. Simultaneously, the system can adaptively adjust its early warning strategy based on the dynamic flow characteristics of the exhibition hall and employs a tiered early warning mechanism. This achieves rapid response and accurate positioning while balancing the viewing experience with security reliability, thus providing a higher level of comprehensive protection for artworks in complex exhibition environments. Attached Figure Description

[0016] To more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0017] Figure 1 This is a schematic diagram of the anti-theft and early warning system for an art display cabinet based on multimodal sensor data fusion provided in the first embodiment of the present invention; Figure 2 A flowchart illustrating the process of acquiring abnormal weak signals provided in the second embodiment of the present invention; Figure 3 A flowchart illustrating the compensation process for abnormally weak signals provided in the third embodiment of the present invention; Figure 4 This is a flowchart illustrating the process of dividing pedestrian flow zones according to the fourth embodiment of the present invention. Detailed Implementation

[0018] To further illustrate the technical means and effects adopted by the present invention to achieve its intended purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details the specific implementation, structure, features, and effects of the multimodal sensor data fusion anti-theft and early warning system for art display cabinets proposed according to the present invention. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. Furthermore, specific features, structures, or characteristics in one or more embodiments can be combined in any suitable form.

[0019] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0020] The following description, in conjunction with the accompanying drawings, details the specific scheme of the anti-theft and early warning system for art display cabinets based on multimodal sensor data fusion provided by this invention.

[0021] Please see Figure 1 The diagram illustrates a multimodal sensor data fusion-based anti-theft and early warning system for art display cases provided in the first embodiment of the present invention. The system includes: The extraction module 101 is used to extract features of time-series data from various sensors deployed in the art display case and the trigger time relationship between the features.

[0022] In security systems for art display cases, a collaborative sensing network composed of multiple sensors is typically deployed to achieve comprehensive coverage and cross-verification of multi-dimensional risk signals. For example, the system can integrate vibration sensors to monitor impacts or abnormal vibrations, pressure sensors to detect subtle changes in the display case load, infrared sensors to detect approach behavior and distance anomalies, and door magnetic sensors to monitor the door opening and closing status. This multi-modal data fusion enhances the completeness and reliability of security decisions.

[0023] Each sensor is installed at a preset position in the art display case, and the data collected is the time-series data. Each type of time-series data has its corresponding characteristics. For example, the vibration sensor is characterized by a spectral acceleration within a preset frequency band that is greater than or equal to a preset spectral acceleration threshold, and the pressure sensor is characterized by a micro-variation that is greater than or equal to a preset micro-variation threshold.

[0024] The micro-variance refers to the minute change in the measured pressure, which can also be understood as the absolute value of the difference between the pressure data at two sampling times. In this embodiment of the invention, the preset micro-variance threshold is set to 0.5. In other embodiments, the implementer may adjust this value according to the actual situation. The current preset micro-variance threshold is determined through long-term observation and analysis of the micro-variance.

[0025] Furthermore, the characteristics of the time-series data include: spectral acceleration within a preset frequency band in the vibration signal, micro-variables in the pressure signal, distance change rate or distance change in the ultrasonic signal, opening angle in the door magnetic signal, and temperature gradient in the infrared signal.

[0026] It should be noted that the temperature gradient in the infrared signal refers to the gradient value in the temperature data sequence collected by the infrared sensor.

[0027] Furthermore, the triggering time relationship includes: When the typical intrusion scenario is a knocking demolition, after the vibration signal is triggered, the pressure signal becomes abnormal within the first preset time interval. After the pressure signal becomes abnormal, the ultrasonic signal changes abruptly within the second preset time interval. The entire triggering process of the above signals is completed within the first preset total time window. When a typical intrusion scenario is the unauthorized opening of a cabinet door, after the door magnetic signal is triggered, the infrared signal becomes abnormal within the third preset time interval. After the infrared signal becomes abnormal, the ultrasonic signal undergoes a sudden change within the fourth preset time interval. The entire triggering process of the above signals is completed within the second preset total time window. When a typical intrusion scenario is the violent movement of a display case, after the pressure signal is triggered, the ultrasonic signal becomes abnormal within the fifth preset time interval. After the ultrasonic signal becomes abnormal, the vibration signal undergoes a sudden change within the sixth preset time interval. The entire triggering process of the above signals is completed within the third preset total time window.

[0028] The first, second, third, fourth, fifth, and sixth preset time intervals can all be set by the user according to their needs, and there are no restrictions here.

[0029] The first, second, and third preset total time windows can all be set by the user according to their needs, and there are no restrictions here.

[0030] Typical intrusion scenarios include, but are not limited to: knocking and breaking down, illegally opening cabinet doors, and forcibly moving display cases.

[0031] The first matching module 102 is used to match the features and trigger time relationships with the features and trigger time relationships of typical intrusion scenarios in the preset causal chain library. If the match is successful, it is determined to be a typical intrusion and a corresponding warning is triggered.

[0032] For security systems of art display cases, different intrusion behaviors exhibit unique characteristic patterns and causal relationships between dimensions at the data level. Therefore, it is necessary to systematically construct a standardized library of typical intrusion causal chains to support high-precision early warning and behavior recognition. The specific process is as follows: Based on the experience of security experts, this study focuses on high-frequency intrusion scenarios such as knocking and breaking, illegal opening of doors, and violent movement, and clarifies their physical processes and expected data performance.

[0033] In a controlled testing environment, specialized equipment is used to simulate various intrusion behaviors, such as impact hammers simulating striking and robotic arms simulating moving. Each scenario is repeated 50–100 times, covering variables such as force and angle to ensure data diversity and representativeness. All sensors are synchronously triggered via a data acquisition card, with a timestamp error ≤1ms. The sampling rate is set according to sensor characteristics: vibration / pressure sensors: ≥1000 Hz, ultrasonic sensors: ≥100 Hz, infrared sensors: ≥1000 Hz. Data is acquired based on a unified clock source. Low-frequency data (such as ultrasound) is linearly interpolated, and high-frequency data (such as vibration) is downsampled to align multimodal data on the time axis, forming a feature sequence with a unified dimension.

[0034] Use tools such as LabelStudio to label each data entry with intrusion scenario tags.

[0035] An anomaly detection algorithm is used to analyze sensor data under various intrusion scenarios. Data is sorted by the degree of anomaly, and data records at or above the 90th percentile are selected as relevant data for that scenario. The anomaly detection algorithm can be an existing algorithm such as the local outlier factor algorithm.

[0036] The extracted data were statistically analyzed, and the quartiles (Q3) on the box plot were used as the characteristic thresholds for each sensor data to determine whether an anomaly was triggered.

[0037] For the same intrusion event, the abnormal peak points (i.e. the points with the highest signal amplitude) in the signals of each sensor are extracted as the key time points of the sensor's response.

[0038] Calculate the time difference between the peak points of the preceding and following sensor signals (the time point at which the sensor is triggered later), and obtain the distribution of this time difference across all events.

[0039] The 95th percentile of the time difference distribution is taken as the upper limit of the maximum allowable time interval between signals in this type of intrusion event, which is used to determine whether the signals belong to the same causal chain.

[0040] The sensors are sorted according to the order in which their abnormal peak values ​​occur; this sorting constitutes a causal order. For example, if the peak value of the vibration sensor occurs earlier than that of the pressure sensor, then vibration is the cause and pressure is the effect, indicating that the intrusion may have started with a knock, which then triggered a pressure change.

[0041] The above analysis results were submitted to security experts for confirmation and correction, and finally a standardized causal chain library was constructed.

[0042] Each causal chain contains three core parameters: Physical parameter order: the types of sensors involved and their causal order; Temporal correlation: the upper limit of the time difference between signals (based on the 95th quartile); Data characteristics: the anomaly threshold of each sensor (based on the upper quartile). The characteristics and triggering time relationships of typical intrusion scenarios in the causal chain library are shown in Table 1. Table 1. Characteristics and Trigger Time Relationships of Typical Intrusion Scenarios in the Causal Chain Library When the characteristics and triggering time relationship match the characteristics (i.e., data characteristics) and triggering time relationship (i.e., time correlation) of typical intrusion scenarios in the causal chain library in Table 1, it is determined to be a typical intrusion. When a typical intrusion occurs, corresponding early warning measures such as activating network alarm and sending alarm signals to preset security terminals can be implemented.

[0043] The identification compensation module 103 is used to identify abnormal weak signals from the sensing feature sequence and calculate their cumulative feature values ​​for compensation if the matching fails.

[0044] If a match fails, atypical intrusions should also be considered. These patterns are typically highly covert, exhibiting weak data characteristics. Their signal features often do not meet the feature thresholds set in typical causal chains, making them easily overlooked by conventional threshold detection methods. However, because atypical intrusions are relatively less destructive, their data often exhibits continuous, weak abnormal fluctuations. To address this characteristic, the system can employ a combination of weak signal accumulation compensation and temporal fuzzy matching, performing multi-granularity comparisons with a causal chain library to effectively identify covert intrusion behaviors.

[0045] Multimodal data is acquired in real time by various sensors, and data synchronization is achieved based on a unified clock source to ensure strict alignment of all signals on the time axis. Before entering the analysis process, the raw data undergoes preprocessing such as filtering and noise reduction to eliminate environmental interference and measurement noise. Subsequently, based on the physical characteristics of different sensor data, corresponding feature extraction methods are used to extract key feature dimensions from the preprocessed signals. For example: vibration signals: extracting the maximum peak acceleration within a preset frequency band to capture the mid-frequency energy characteristics of actions such as knocking and collisions; pressure signals: calculating the micro-motion changes within a certain time window to sense pressure fluctuations caused by slow application of force or covert prying. The above feature extraction methods are all mature technologies in the field of signal processing, and their specific implementation details are not elaborated here.

[0046] The process of identifying the abnormally weak signal will be described in detail in the second embodiment, and will not be repeated here.

[0047] The compensation process for the abnormally weak signal will be described in detail in the third embodiment, and will not be repeated here.

[0048] The second matching module 104 is used to sort the compensated abnormal weak signal and other sensing trigger signals in chronological order to construct a signal trigger timing sequence, and match it with a preset standard signal trigger timing sequence to obtain a matching distance sequence. The matching distance sequence is then corrected using the cumulative feature value, and the matching degree is calculated based on the corrected matching distance sequence.

[0049] All valid signals (including regular trigger signals and compensated and enhanced abnormal weak signals) are sorted according to their timestamps to construct a signal trigger timing sequence, where each element records its trigger time and signal type label.

[0050] The causal chain library pre-sets a corresponding "standard signal trigger timing sequence" for each standard intrusion mode. During the matching process, the Dynamic Time Warping (DTW) algorithm is used to perform temporal fuzzy matching between the test sequence and each standard sequence, obtaining a matching distance sequence that characterizes their correspondence. Each distance value in this sequence represents the matching deviation between the corresponding signal point in the test sequence and the corresponding point in the standard sequence; the smaller the distance, the higher the matching degree.

[0051] To accurately assess the matching contribution of weak signals, the elements from the compensation signal in the matching distance sequence need to be corrected. The correction method is dynamically adjusted based on the cumulative characteristic value of the weak signal: the larger the cumulative characteristic value, the stronger the persistence of the weak signal and the more likely it is to reflect real intrusion behavior. Therefore, its corresponding matching distance should be shortened accordingly to enhance its positive weight in the overall matching degree calculation.

[0052] Elements in the matching distance sequence derived from the compensation signal can be corrected according to the following correction formula, which includes: ; Among them, the This represents the correction value for the matching distance of k elements in the matching distance sequence. This represents the matching distance of k elements in the matching distance sequence. The cumulative feature value corresponding to the k elements in the matching distance sequence represents the cumulative feature value. This represents the normalization function, preferably the maximum-minimum normalization function.

[0053] Finally, based on the corrected matching distance sequence, the overall matching degree between the current test sequence and each standard intrusion pattern is calculated to achieve accurate identification of atypical intrusion behaviors.

[0054] Furthermore, the process of obtaining the matching distance includes: The matching distance sequence is obtained by matching the signal trigger timing sequence with the standard signal trigger timing sequence using a dynamic time warping algorithm.

[0055] Furthermore, the process of obtaining the matching degree includes: Calculate the average value of the matching distances contained in the matching distance sequence, and perform reverse normalization on the average value to obtain the matching degree.

[0056] The early warning module 105 is used to assess atypical intrusion risks based on the matching degree, divide the traffic flow range based on the historical traffic flow status of the exhibition hall, and perform graded early warning based on the atypical intrusion risks and the traffic flow range.

[0057] Furthermore, assessing atypical intrusion risks based on the matching degree includes: When the matching degree is greater than or equal to the first preset matching value and less than the second preset matching value, the atypical intrusion risk is low risk; When the matching degree is greater than or equal to the second preset matching value and less than the third preset matching value, the atypical intrusion risk is medium risk; When the matching degree is greater than or equal to the third preset matching value, the atypical intrusion risk is high risk.

[0058] The first preset matching value is less than the second preset matching value, which is less than the third preset matching value. All of these values ​​can be set by the user according to their needs, and will not be elaborated here.

[0059] The first preset matching value is preferably 0.3, the second preset matching value is preferably 0.5, and the third preset matching value is preferably 0.7.

[0060] Furthermore, the tiered early warning system based on the aforementioned atypical intrusion risks and the aforementioned pedestrian traffic zones includes: If the current period is a high-traffic area, and the atypical intrusion risk is low, a light warning will be triggered; if the atypical intrusion risk is medium, a light warning will be triggered and an alarm sound will be added; if the atypical intrusion risk is high, a network alarm will be activated and an alarm signal will be sent to a preset security terminal. If the current period is in a low-traffic area, and the atypical intrusion risk is low, a light warning will be triggered and an alarm sound will be added. If the atypical intrusion risk is medium or high, a network alarm will be activated and an alarm signal will be sent to a preset security terminal.

[0061] Figure 2 The flowchart below shows the process for acquiring abnormal weak signals according to the second embodiment of the present invention. The process for acquiring abnormal weak signals includes: S201. Based on the anomaly recognition algorithm, identify candidate weak signals that exceed the normal range from the sensing feature sequence.

[0062] The anomaly detection algorithm can be 3 Anomaly detection algorithms are existing technologies.

[0063] S202. Obtain the average feature value of the candidate weak signal, and calculate the feature value deviation of the candidate weak signal by combining the upper quartile and standard deviation of the sensing feature sequence.

[0064] The deviation of the eigenvalue can be expressed by the formula: ; Among them, the The eigenvalue deviation of the candidate weak signal i represents the degree of deviation of the eigenvalue. This represents a normalization function, such as a max-min normalization function. The average eigenvalue of the candidate weak signal is represented by the following: The upper quartile represents the quartile. This represents the standard deviation.

[0065] It should be noted that when the aforementioned When a value of 0 occurs, to prevent the denominator in the formula for obtaining the eigenvalue deviation from 0 from being zero, the denominator can be adjusted accordingly. It can be replaced with , wherein For a very small positive number, such as .

[0066] S203. Calculate the Pearson correlation coefficient between the candidate weak signal and the sensing feature sequences of other modes, construct the cooperability coefficient matrix, and obtain its average cooperability coefficient.

[0067] To effectively identify atypical, low-intensity, and covert intrusions, a cross-modal synergy analysis mechanism is introduced after extracting feature sequences from various dimensions and performing standardization and interpolation alignment. This mechanism calculates the Pearson correlation coefficient matrix between a candidate weak signal and all other modal signals, and then obtains its average synergy coefficient, thereby quantifying the degree of coordination between the signal and the overall system response. If the synergy coefficient is significantly low or negative, it indicates a temporal inconsistency between the weak signal and the system's multimodal behavior, serving as a key criterion for identifying covert anomalies. This, in turn, supports the system in achieving more refined and robust multi-level intrusion detection capabilities.

[0068] S204. Calculate the degree of anomaly of the candidate weak signal based on the eigenvalue deviation and the average coordination coefficient.

[0069] The degree of abnormality can be expressed by the formula: ; Among them, the This indicates the degree of anomalousness of the candidate weak signal i. The eigenvalue deviation of the candidate weak signal i represents the degree of deviation of the eigenvalue. This represents the average synergy coefficient.

[0070] S205. When the degree of abnormality is greater than the degree of abnormality threshold, the candidate weak signal is the abnormal weak signal.

[0071] The abnormality threshold can be set independently according to user needs, and is preferably set to 0.5.

[0072] Because covert signals are typically less destructive, they often manifest as continuous, small-amplitude anomalies over time. Therefore, the system introduces a time-cumulative enhancement mechanism in the feature weighting stage: by integrating the weak signal over a sustained period, its cumulative energy or amplitude is obtained. The longer the duration, the greater the cumulative amount of the signal within the period, indicating that it is more likely to be a continuous, covert intrusion, and its weight in the comprehensive judgment is correspondingly increased. This method transforms temporal continuity into a quantifiable weighting factor, thereby effectively enhancing the sensitivity to identify long-duration, low-intensity anomaly patterns.

[0073] Figure 3 The flowchart illustrates the compensation process for abnormal weak signals provided in the third embodiment of the present invention. The compensation process for abnormal weak signals includes: S301. Determine the weight of the abnormal weak signal based on its duration.

[0074] The duration of the abnormally weak signal i Determine its weight : when hour, ; when hour, ; when hour .

[0075] S302. The cumulative feature value is calculated based on the weight and the feature value of the abnormal weak signal.

[0076] The cumulative eigenvalue can be expressed by the formula: ; Among them, the The cumulative characteristic value of the abnormal weak signal i represents the value of the signal i. Represents the normalization function, the The characteristic value representing the abnormal weak signal i, the and the These represent the start and end times of the duration of the abnormal weak signal i, respectively.

[0077] S303. Using the cumulative feature value as a compensation coefficient, the average feature value of the abnormal weak signal is enhanced to obtain the compensated feature value of the abnormal weak signal.

[0078] The compensation characteristic value can be expressed by the formula: ; Among them, the Represents the compensation feature value, the This represents the cumulative characteristic value of the abnormally weak signal i.

[0079] Figure 4 The flowchart below shows the process of dividing pedestrian flow areas according to the fourth embodiment of the present invention. The process of dividing pedestrian flow areas includes: S401. Collect the historical daily visitor flow data of the exhibition hall at preset time intervals.

[0080] The time interval can be set by the user, preferably 5 minutes, and the historical full day can refer to the preset time period of the previous year or the preset time period of the previous week.

[0081] The pedestrian flow data is obtained in real time through visual sensors (such as cameras) deployed at the entrance of the exhibition hall and key areas, and is identified and counted by a lightweight neural network model, ensuring that the data source is objective and continuous.

[0082] S402. Calculate the average pedestrian flow in each time period over multiple historical periods to form a mean sequence.

[0083] S403. The mean sequence is divided using the maximum threshold segmentation method to obtain high traffic flow intervals and low traffic flow intervals.

[0084] The maximum threshold segmentation method is existing technology and will not be described in detail here.

[0085] The present invention has the following beneficial effects: First, the features and trigger time relationships acquired by the extraction module form the data foundation for subsequent analysis. Second, the first matching module matches the features and trigger time relationships with a typical intrusion scenario library, i.e., a causal chain library. If successful, an early warning is triggered directly, achieving efficient response. Next, if no match is found, the identification compensation module is activated, responsible for identifying and amplifying abnormal weak signals that may represent atypical intrusions. Then, the second matching module uses the compensated signal to construct and compare signal trigger time sequences, calculating the corrected matching degree to quantify the probability of atypical intrusion. Finally, the early warning module integrates the matching degree with pedestrian flow information to execute dynamic hierarchical early warnings.

[0086] In summary, this invention significantly improves the accuracy of the art display case security system in identifying typical and atypical intrusions by constructing a causal chain library and using temporal fuzzy matching technology, effectively reducing false alarms under environmental interference and missed alarms under covert attacks. Simultaneously, the system can adaptively adjust its early warning strategy based on the dynamic flow characteristics of the exhibition hall and employs a tiered early warning mechanism. This achieves rapid response and accurate positioning while balancing the viewing experience with security reliability, thus providing a higher level of comprehensive protection for artworks in complex exhibition environments.

[0087] It should be noted that the order of the above embodiments of the present invention is merely for descriptive purposes and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require a specific or sequential order to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0088] The various embodiments in this specification are described in a progressive manner. The same or similar parts between the various embodiments can be referred to each other. Each embodiment focuses on describing the differences from other embodiments.

Claims

1. A multimodal sensor data fusion-based anti-theft and early warning system for art display cases, characterized in that: The system includes: An extraction module is used to extract features from the time-series data of various sensors deployed in the art display case and the trigger time relationship between the features; The first matching module is used to match the features and trigger time relationship with the features and trigger time relationship of typical intrusion scenarios in the preset causal chain library. If the match is successful, it is determined to be a typical intrusion and a corresponding warning is triggered. The identification and compensation module is used to identify abnormal weak signals from the sensing feature sequence if the matching fails, and to calculate their cumulative feature values ​​for compensation. The second matching module is used to sort the compensated abnormal weak signal and other sensing trigger signals in chronological order to construct a signal trigger timing sequence, and match it with a preset standard signal trigger timing sequence to obtain a matching distance sequence. The matching distance sequence is then corrected using the cumulative feature value, and the matching degree is calculated based on the corrected matching distance sequence. The early warning module is used to assess atypical intrusion risks based on the matching degree, divide the traffic flow into intervals based on the historical traffic flow status of the exhibition hall, and execute graded early warnings based on the atypical intrusion risks and the traffic flow intervals.

2. The anti-theft and early warning system for art display cases based on multimodal sensor data fusion as described in claim 1, characterized in that, The characteristics of the time-series data include: spectral acceleration within a preset frequency band in the vibration signal, micro-variables in the pressure signal, distance change rate or distance change in the ultrasonic signal, opening angle in the door magnetic signal, and temperature gradient in the infrared signal.

3. The anti-theft and early warning system for art display cases based on multimodal sensor data fusion as described in claim 1, characterized in that, The triggering time relationship includes: When the typical intrusion scenario is a knocking demolition, after the vibration signal is triggered, the pressure signal becomes abnormal within the first preset time interval. After the pressure signal becomes abnormal, the ultrasonic signal changes abruptly within the second preset time interval. The entire triggering process of the above signals is completed within the first preset total time window. When a typical intrusion scenario is the unauthorized opening of a cabinet door, after the door magnetic signal is triggered, the infrared signal becomes abnormal within the third preset time interval. After the infrared signal becomes abnormal, the ultrasonic signal undergoes a sudden change within the fourth preset time interval. The entire triggering process of the above signals is completed within the second preset total time window. When a typical intrusion scenario is the violent movement of a display case, after the pressure signal is triggered, the ultrasonic signal becomes abnormal within the fifth preset time interval. After the ultrasonic signal becomes abnormal, the vibration signal undergoes a sudden change within the sixth preset time interval. The entire triggering process of the above signals is completed within the third preset total time window.

4. The anti-theft and early warning system for art display cases based on multimodal sensor data fusion as described in claim 1, characterized in that, The process of acquiring the abnormal weak signal includes: Based on the anomaly detection algorithm, candidate weak signals that exceed the normal range are identified from the sensing feature sequence; The average feature value of the candidate weak signal is obtained, and the feature value deviation of the candidate weak signal is calculated by combining the upper quartile and standard deviation of the sensing feature sequence. Calculate the Pearson correlation coefficient between the candidate weak signal and the sensing feature sequences of other modes, construct the cooperability coefficient matrix, and obtain its average cooperability coefficient. The degree of anomaly of the candidate weak signal is calculated based on the eigenvalue deviation and the average cooperability coefficient. When the degree of abnormality is greater than the degree of abnormality threshold, the candidate weak signal is the abnormal weak signal.

5. The anti-theft and early warning system for art display cases based on multimodal sensor data fusion as described in claim 1, characterized in that, The compensation process for the abnormal weak signal includes: The weight of the abnormally weak signal is determined based on its duration. The cumulative feature value is calculated based on the weights and the feature values ​​of the abnormal weak signal. The cumulative feature value is used as a compensation coefficient to enhance the average feature value of the abnormal weak signal, thereby obtaining the compensated feature value of the abnormal weak signal.

6. The anti-theft and early warning system for art display cases based on multimodal sensor data fusion as described in claim 1, characterized in that, The process of obtaining the matching distance includes: The matching distance sequence is obtained by matching the signal trigger timing sequence with the standard signal trigger timing sequence using a dynamic time warping algorithm.

7. The anti-theft and early warning system for art display cases based on multimodal sensor data fusion as described in claim 1, characterized in that, The process of obtaining the matching degree includes: Calculate the average value of the matching distances contained in the matching distance sequence, and perform reverse normalization on the average value to obtain the matching degree.

8. The anti-theft and early warning system for art display cases based on multimodal sensor data fusion as described in claim 1, characterized in that, The assessment of atypical intrusion risks based on the matching degree includes: When the matching degree is greater than or equal to the first preset matching value and less than the second preset matching value, the atypical intrusion risk is low risk; When the matching degree is greater than or equal to the second preset matching value and less than the third preset matching value, the atypical intrusion risk is medium risk; When the matching degree is greater than or equal to the third preset matching value, the atypical intrusion risk is high risk.

9. The anti-theft and early warning system for art display cases based on multimodal sensor data fusion as described in claim 1, characterized in that, The process of dividing the pedestrian flow area includes: Collect historical, cyclical visitor traffic data for the entire day in the exhibition hall at preset time intervals; Calculate the average pedestrian flow for each time period over multiple historical periods to form a mean sequence; The mean sequence is divided using the maximum threshold segmentation method to obtain high-traffic and low-traffic intervals.

10. The anti-theft and early warning system for art display cabinets based on multimodal sensor data fusion as described in claim 9, characterized in that, Based on the aforementioned atypical intrusion risks and the aforementioned pedestrian flow zones, tiered early warning systems are implemented, including: If the current period is a high-traffic area, and the atypical intrusion risk is low, a light warning will be triggered; if the atypical intrusion risk is medium, a light warning will be triggered and an alarm sound will be added; if the atypical intrusion risk is high, a network alarm will be activated and an alarm signal will be sent to a preset security terminal. If the current period is in a low-traffic area, and the atypical intrusion risk is low, a light warning will be triggered and an alarm sound will be added. If the atypical intrusion risk is medium or high, a network alarm will be activated and an alarm signal will be sent to a preset security terminal.

Images