Video monitoring method and device for diving system, medium and electronic equipment

Abstract

The invention provides a video monitoring method and device for a diving system, a storage medium and electronic equipment. The method comprises the following steps: acquiring sensor data and video data of each cabin in the diving system; the sensor data represents the state of equipment in the cabin; the video data represents a monitoring video in time synchronization with the sensor data; performing anomaly detection on the sensor data of each cabin to generate an abnormal signal; the anomaly detection comprises single-channel detection and multi-channel detection; the single-channel detection is judged based on the deviation degree of sensor data relative to a historical baseline; the multi-channel detection is judged based on the change of the incidence relation of different types of sensor signals in the sensor data; in response to an abnormal signal for the target cabin, positioning a target monitoring fragment of the target cabin based on a time window of the abnormal signal; the target monitoring fragment is determined through timestamp matching and state fingerprint verification; the state fingerprint is a feature identifier calculated based on the sensor data at the abnormal occurrence moment.

Description

Technical Field

[0001] This application relates to the field of submarine technology, and in particular to a video monitoring method, device, storage medium and electronic equipment for a diving system. Background Technology

[0002] In the safety monitoring of diving systems, monitoring cameras configured in multiple cabins continuously collect a large amount of video data, while various sensors monitor the operating status of the equipment in real time. When an abnormal event occurs in the system, it is necessary to search for the monitoring footage corresponding to the abnormal moment from a massive amount of historical video records in order to analyze the cause of the abnormality. However, due to the large amount of video data, the process of retrieving the corresponding video clip is time-consuming and inefficient, making it difficult to meet the timeliness requirements of emergency response. Summary of the Invention

[0003] In view of this, this application provides a video monitoring method, device, storage medium and electronic device for diving systems to overcome the shortcomings of the prior art.

[0004] According to a first aspect of this application, a video monitoring method for a diving system is provided, applied to a central control system, the method comprising: Acquire sensor and video data from each compartment in the diving system; sensor data characterizes the status of equipment in the compartment; video data characterizes multi-channel monitoring video synchronized with sensor data in time. Anomaly detection is performed on the sensor data of each cabin to generate anomaly signals; anomaly detection includes single-channel detection and multi-channel detection; single-channel detection is based on the degree of deviation of sensor data from historical baseline; multi-channel detection is based on the changes in the correlation between different types of sensor signals in the sensor data. In response to anomaly signals targeting the target cabin, the target monitoring segment of the target cabin is located based on the time window of the anomaly signal; wherein, the target monitoring segment is determined by timestamp matching and status fingerprint verification; the status fingerprint is a feature identifier calculated based on sensor data at the time of the anomaly occurrence.

[0005] Another aspect of this application provides a video monitoring device for a diving system, comprising: The acquisition module is used to acquire sensor data and video data from each compartment in the diving system; the sensor data represents the status of the equipment in the compartment; the video data represents multi-channel monitoring video synchronized with the sensor data in time. The detection module is used to detect anomalies in the sensor data of each cabin and generate anomaly signals. Anomaly detection includes single-channel detection and multi-channel detection. Single-channel detection is based on the degree of deviation of sensor data from historical baseline. Multi-channel detection is based on the changes in the correlation between different types of sensor signals in the sensor data. The determination module is used to locate the target monitoring segment of the target cabin based on the time window of the abnormal signal in response to the abnormal signal; wherein, the target monitoring segment is determined by timestamp matching and status fingerprint verification; the status fingerprint is a feature identifier calculated based on sensor data at the time of the abnormality.

[0006] Another aspect of this application provides an electronic device comprising: One or more processors; Memory, used to store one or more programs. When the one or more programs are executed by the one or more processors, the one or more processors implement the method described above.

[0007] Another aspect of this application provides a computer-readable storage medium storing computer-executable instructions that, when executed, are used to implement the method described above.

[0008] By adopting the technical solution of this application, anomaly detection is performed on the sensor data of each cabin, and anomaly signals are generated. This automatically determines the time of anomaly occurrence, avoiding a full-time search through massive historical video. Anomaly detection employs a combination of single-channel and multi-channel detection, improving the accuracy of anomaly identification. Upon detecting an anomaly signal, the target monitoring segment is located based on the time window of the anomaly signal, narrowing the search range from all historical video to the specific time period in which the anomaly occurred, significantly reducing the amount of video data that needs to be processed. Dual verification through timestamp matching and state fingerprint verification ensures that the located target monitoring segment precisely corresponds to the anomaly event. The state fingerprint, as a feature identifier calculated based on sensor data at the time of the anomaly, further guarantees the accuracy of segment location.

[0009] Compared to existing technologies where the large volume of video data leads to time-consuming and inefficient retrieval processes, the technical solution of this application can quickly and accurately locate the video segments corresponding to anomalies, significantly improving retrieval efficiency and meeting the timeliness requirements of emergency response for diving systems.

[0010] It should be understood that the description in this section is not intended to identify key or essential features of the embodiments of this application, nor is it intended to limit the scope of this application. Other features of this application will become readily apparent from the following description. Attached Figure Description

[0011] The above and other objects, features and advantages of this application will become clearer from the following description of embodiments with reference to the accompanying drawings, in which: Figure 1 This schematic diagram illustrates the architecture of a shipboard hoisting system provided in this application; Figure 2 A flowchart illustrating a video monitoring method for a diving system provided in this application is shown schematically. Figure 3 This schematic diagram illustrates a structural block diagram of a video monitoring device for a diving system provided in this application. Figure 4 A schematic block diagram of an electronic device provided in this application is shown. Detailed Implementation

[0012] The embodiments of this application will now be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of this application. In the following detailed description, numerous specific details are set forth to provide a thorough understanding of the embodiments of this application for ease of explanation. However, it will be apparent that one or more embodiments may be implemented without these specific details. Furthermore, descriptions of well-known structures and technologies are omitted in the following description to avoid unnecessarily obscuring the concepts of this application.

[0013] The terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of this application. The terms “comprising,” “including,” etc., as used herein indicate the presence of the stated features, steps, operations, and / or components, but do not exclude the presence or addition of one or more other features, steps, operations, or components.

[0014] This application provides a video monitoring method, device, storage medium, and electronic device for a diving system. The method can be applied to the central control system of the diving system. The diving system described above is described below.

[0015] Figure 1 The schematic diagram illustrates the architecture of a diving system provided in this application.

[0016] like Figure 1 As shown, the 100-dimensional distributed monitoring architecture of the diving system acquires multi-dimensional system operation data by precisely configuring sensor devices in each functional compartment.

[0017] Among them, the living quarters 120 serve as the core living area for divers. The oxygen and carbon dioxide concentration sensors deployed inside are responsible for collecting gas concentration data to ensure that the air quality inside the cabin meets the survival requirements of the personnel. At the same time, the cabin pressure and temperature and humidity sensors provide cabin pressure and environmental data.

[0018] As the core platform for underwater operations, the Diving Bell 121 is equipped with a hydraulic system pressure sensor to collect cabin pressure data of the hydraulic operating system, a motor current sensor to obtain the operating parameters of the electrical equipment of the drive system, and an underwater depth sensor to provide environmental data of the operating location.

[0019] Escape capsule 122 and equipment compartment 123 are responsible for emergency support and equipment maintenance, respectively. The emergency gas supply pressure sensor in escape capsule 122 collects gas concentration data in emergency situations, while the door status sensor monitors the physical status of the escape passage. Meanwhile, the main pump power sensor, cooling water flow sensor, and cable insulation monitoring sensor in equipment compartment 123 comprehensively collect the operating parameters of the electrical equipment of the core equipment.

[0020] The cameras and corresponding sensor devices installed in each cabin use a unified time base to collect data, ensuring that the video data and sensor data are accurately synchronized in the time dimension, thereby forming a multi-channel monitoring video stream with spatiotemporal correspondence.

[0021] Each cabin's edge nodes 120a, 121a, 122a, and 123a establish local communication connections with cameras and various sensors within their respective cabins, enabling unified acquisition, preprocessing, and buffering management of sensor and video data. The edge nodes establish stable communication links with the central control system 110 via standardized network protocols, transmitting locally preprocessed sensor and video data to the central control system 110 in real time, forming a centralized data processing and analysis platform. The central control system 110, through communication connections with each cabin's edge nodes, obtains a complete set of sensor data, including gas concentration data, cabin pressure data, and electrical equipment operating parameters, combined with synchronously acquired multi-channel monitoring video.

[0022] Based on the above system architecture, the video monitoring method for diving systems provided in the embodiments of this application will be described below.

[0023] Figure 2 A flowchart illustrating a video monitoring method for a diving system provided in an embodiment of this application is shown.

[0024] like Figure 2 As shown, the video monitoring method for a diving system includes steps S201 to S203.

[0025] Step S201: Obtain sensor data and video data from each compartment in the diving system; sensor data represents the status of equipment in the compartment; video data represents multi-channel monitoring video synchronized with the sensor data in time. Step S202: Perform anomaly detection on the sensor data of each cabin and generate anomaly signals; anomaly detection includes single-channel detection and multi-channel detection; single-channel detection is based on the degree of deviation of sensor data from historical baseline; multi-channel detection is based on the change in the correlation between different types of sensor signals in the sensor data. Step S203: In response to an abnormal signal for the target cabin, locate the target monitoring segment of the target cabin based on the time window of the abnormal signal; wherein, the target monitoring segment is determined by timestamp matching and status fingerprint verification; the status fingerprint is a feature identifier calculated based on sensor data at the time of the abnormality.

[0026] In step S201, sensor data refers to the quantitative information data collected in real time by various sensors in the diving system to reflect the operating status of the equipment. This data is used to accurately characterize the real-time operating status, performance, and health status of the equipment in the cabin, thereby providing a data basis for equipment status assessment and anomaly identification.

[0027] Sensor data, by reflecting multiple dimensions such as electrical parameters, mechanical parameters, and environmental parameters of the equipment, can comprehensively depict whether the current working status of each piece of equipment in the ship's cabin is normal.

[0028] For example, sensor data includes, but is not limited to, oxygen concentration data and carbon dioxide concentration data reflecting the status of gas handling equipment, cabin pressure data reflecting the status of pressure control equipment, motor current data and main pump power data reflecting the status of electrical system equipment, hydraulic system pressure data reflecting the status of hydraulic system equipment, cooling water flow data reflecting the status of cooling system equipment, cable insulation monitoring data reflecting the status of electrical safety equipment, and cabin door status data reflecting the status of mechanical equipment.

[0029] Correspondingly, video data refers to the real-time image sequence information collected by cameras distributed in various chambers of the diving system. It can be understood as video streams from multiple monitoring channels that are collected synchronously with sensor data using a unified time reference, and are used to provide a visual monitoring screen of equipment operation and personnel activities in the cabin.

[0030] The video data representation and the time synchronization of sensor data in multi-channel monitoring video means that each frame of video image establishes a precise time correspondence with the sensor data collected at the same time, forming a parallel video stream of multiple monitoring channels.

[0031] For example, the video data includes, but is not limited to, videos of personnel living conditions in the living quarters monitoring channel, videos of underwater operations in the diving bell monitoring channel, videos of emergency equipment status in the escape capsule monitoring channel, videos of key equipment operation in the equipment compartment monitoring channel, and videos of personnel movement and equipment status in the monitoring channels of the connecting areas of each compartment.

[0032] In one feasible implementation, the data acquisition timing of sensors and cameras can be uniformly coordinated through edge nodes configured in each cabin. The edge nodes provide a synchronized time reference for sensor data acquisition and video frame capture based on a built-in high-precision clock source, ensuring that sensor readings and video frames acquired at the same timestamp can accurately correspond. Then, the sensor data and video data with time synchronization characteristics are transmitted to the central control system through the network.

[0033] In another feasible implementation, a master-slave time synchronization method can be adopted, in which the central control system acts as the time master node and periodically sends time calibration signals to the edge nodes of each cabin. Each edge node adjusts its local clock according to the received time calibration signal and triggers sensor data acquisition and video frame capture actions according to a unified time reference, thereby achieving precise time synchronization between multi-channel sensor data and multi-channel video data.

[0034] In step S202, anomaly detection refers to the process of identifying deviations in the operating status of the diving system from the normal operating mode by real-time analysis of sensor data. It can be understood as continuously monitoring the sensor data of each chamber by combining single-channel detection and multi-channel detection, in order to promptly detect abnormal situations that may affect the safe operation of the diving system, such as equipment failure, environmental anomalies, or operational errors, and generate corresponding abnormal signals to trigger subsequent emergency response measures.

[0035] Optionally, anomaly detection can include single-channel detection and multi-channel detection. Single-channel detection refers to a detection method that independently analyzes the data of a single sensor channel to identify anomalies in that channel's data. In this embodiment, it can be understood as a detection mechanism that determines whether anomalies exist by comparing and analyzing the current sensor data with the sensor's historical normal operating data, used to identify abnormal states of a single device or a single monitoring parameter.

[0036] For example, single-channel detection can achieve anomaly detection based on the degree of deviation by establishing a historical baseline data model for each sensor.

[0037] Specifically, the system first collects historical data from each sensor under normal operating conditions, and calculates the normal value range, mean, and standard deviation of each sensor through statistical analysis methods to form the historical baseline of that sensor. Then, during real-time monitoring, the system compares the currently collected sensor data with the corresponding historical baseline, calculates the degree of deviation of the current value from the historical baseline, and determines an anomaly when the degree of deviation exceeds a preset threshold and generates an anomaly signal.

[0038] Correspondingly, multi-channel detection refers to a detection method that simultaneously analyzes data from multiple different types of sensor channels to identify abnormal correlations between channels. It can be understood as a detection mechanism that discovers system-level anomalies by monitoring changes in the correlation between signals from different types of sensors, and is used to identify complex anomalies and systemic faults that cannot be detected by single-channel detection.

[0039] For example, multi-channel detection achieves anomaly detection based on changes in correlation by establishing a correlation model between signals from different types of sensors.

[0040] Specifically, the system first analyzes the correlation patterns between different types of sensor signals under normal operating conditions, such as the correlation between the numerical changes of oxygen concentration sensor and carbon dioxide concentration sensor, and the linkage relationship between hydraulic system pressure sensor and motor current sensor, and establishes a normal correlation model between the sensor signals. Then, in real-time monitoring, it calculates the actual correlation between the sensor signals at the current moment, compares it with the normal correlation model, and determines that the correlation is abnormal in multiple channels and generates an abnormal signal when there is a significant deviation in the correlation.

[0041] In one feasible implementation, anomaly detection can be performed by combining threshold comparison and statistical analysis. For single-channel detection, two judgment criteria are set: a fixed threshold and a dynamic threshold. The fixed threshold is determined based on the equipment technical specifications, while the dynamic threshold is determined based on the statistical distribution of historical data. For multi-channel detection, correlation analysis and trend consistency analysis are used to evaluate the changes in the correlation between signals from different sensors.

[0042] In another feasible implementation, a sliding window mechanism can be used for anomaly detection. The system maintains a sliding window containing sensor data within the most recent time period. Within the window, the statistical characteristics of single-channel data and the correlation characteristics of multi-channel data are calculated. Anomalies are determined by comparing the data within the window with historical baselines. The size of the sliding window is dynamically adjusted according to the response characteristics of different types of sensors and the timeliness requirements of anomaly detection.

[0043] It should be noted that single-channel and multi-channel detection employ parallel processing. The abnormal signals generated by the two detection methods have different confidence levels and type identifiers. Single-channel anomalies typically reflect local equipment failures, while multi-channel anomalies typically reflect systemic problems or complex faults. The generation of abnormal signals includes not only the time information of the anomaly occurrence but also detailed information such as the anomaly type, anomaly severity, and the sensor channels involved.

[0044] In step S203, the target monitoring segment refers to a video data segment within a specific cabin corresponding to the time window of the abnormal signal. It can be understood as a video segment extracted from the continuous monitoring video of the cabin, containing a certain time range before and after the time of the abnormality, used to provide a visual record of the abnormal event process so that operators can analyze the cause of the abnormality and assess its impact.

[0045] For example, the target monitoring segments include, but are not limited to, video segments showing the equipment operating status before and after the moment of equipment failure in the equipment compartment, video segments showing the monitoring of personnel behavior before and after the moment of abnormal activity in the living quarters, video segments showing the operation process before and after the moment of abnormal operation in the diving bell, and video segments showing the emergency response before and after the moment of emergency state triggering in the escape capsule.

[0046] Optionally, the target monitoring segment is determined through timestamp matching and status fingerprint verification. Timestamp matching refers to the process of searching for the corresponding video frame in the target cabin's video data based on the time of the anomaly recorded in the anomaly signal. This can be understood as a technical means of accurately locating the video time segment corresponding to the anomaly event in the synchronized video stream based on the time window information of the anomaly signal, ensuring that the acquired video segment and the detected anomaly event correspond accurately in time.

[0047] Correspondingly, state fingerprint verification refers to the process of verifying the accuracy of a video segment by comparing the sensor data characteristics at an abnormal moment with the device state characteristics at the corresponding moment. It can be understood as a verification method that uses state fingerprint as a verification standard to confirm that the located video segment does indeed correspond to the device state reflected by the abnormal signal, in order to eliminate time synchronization errors and ensure the validity of the video segment.

[0048] Among them, state fingerprint refers to the unique identification information formed by extracting and encoding features from sensor data at the moment of an anomaly and the time period before and after it. It can be understood as converting multi-dimensional sensor data at the moment of an anomaly into a digital feature vector that can characterize the equipment status and environmental conditions at that time through a specific algorithm, and using it as a reference benchmark to verify the accuracy of the target monitoring segment.

[0049] Optionally, the state fingerprint is calculated based on sensor data at the moment the anomaly occurred. The system extracts key sensor parameters within a preset time window before and after the anomaly, and converts the multidimensional sensor data into a fixed-length feature identifier code through processing steps such as data normalization, feature selection, and hash encoding.

[0050] In one feasible implementation, a dual verification method can be used to locate the target monitoring segment. First, timestamp matching is performed on the video data of the target cabin based on the time window information of the abnormal signal. A video segment candidate interval is formed by extending a preset time length forward and backward according to the time when the abnormality occurred. Then, the state fingerprint at the time of the abnormality is calculated, and the state fingerprint is compared and verified with the sensor data features in the corresponding time period of the candidate video segment to confirm the validity and accuracy of the video segment.

[0051] In another feasible implementation, a sliding window search method can be used to locate the target monitoring segment. The system sets a search range near the time window of the abnormal signal, and calculates the state fingerprint corresponding to the video segment segment by segment in the sliding window method within the range. The calculated state fingerprint is compared with the standard state fingerprint at the abnormal time, and the video segment with the highest similarity is selected as the target monitoring segment. At the same time, the similarity score is recorded as the segment credibility index.

[0052] It should be noted that timestamp matching and state fingerprint verification are two complementary localization mechanisms. Timestamp matching ensures the correspondence between video clips and anomalous events in the time dimension, while state fingerprint verification ensures the consistency between video clips and anomalous events in the state dimension. The calculation of state fingerprints involves the fusion of data from multiple sensor channels, and different types of anomalous events correspond to different state fingerprint patterns.

[0053] By adopting the technical solution of this application, anomaly detection is performed on the sensor data of each cabin, and anomaly signals are generated. This automatically determines the time of anomaly occurrence, avoiding a full-time search through massive historical video. Anomaly detection employs a combination of single-channel and multi-channel detection, improving the accuracy of anomaly identification. Upon detecting an anomaly signal, the target monitoring segment is located based on the time window of the anomaly signal, narrowing the search range from all historical video to the specific time period in which the anomaly occurred, significantly reducing the amount of video data that needs to be processed. Dual verification through timestamp matching and state fingerprint verification ensures that the located target monitoring segment precisely corresponds to the anomaly event. The state fingerprint, as a feature identifier calculated based on sensor data at the time of the anomaly, further guarantees the accuracy of segment location.

[0054] Compared to existing technologies where the large volume of video data leads to time-consuming and inefficient retrieval processes, the technical solution of this application can quickly and accurately locate the video segments corresponding to anomalies, thereby improving retrieval efficiency and meeting the timeliness requirements of emergency response for diving systems.

[0055] Based on the above embodiments, as an optional embodiment, step S220, which involves single-channel detection of sensor data from each cabin to generate an abnormal signal, may further include the following steps: Step S310: For any cabin, the sensor data of the cabin is standardized within a preset sliding time window to obtain a normalized signal. Step S320: Calculate the deviation of the normalized signal relative to the historical baseline; the historical baseline represents the statistical mean and standard deviation of the corresponding sensor under normal diving operation conditions. Step S330: Calculate the rate of change of sensor data in the cabin; the rate of change of sensor data characterizes the degree of abrupt change in sensor values ​​per unit time. Step S340: Weighted summation of the deviation value corresponding to the cabin and the rate of change of the sensor data to obtain a single-channel anomaly score; Step S350: If the single-channel anomaly score is greater than or equal to the single-channel threshold, the cabin is identified as the target cabin, and an anomaly signal is generated for the target cabin.

[0056] In step S310, the length of the preset sliding time window is set differently according to the response characteristics and anomaly detection requirements of different sensors. Different types of sensors, such as oxygen concentration sensors, cabin pressure sensors, and motor current sensors, adopt corresponding time window length configurations.

[0057] Specifically, the system collects historical data sequences from the corresponding sensors within each sliding time window, calculates the statistical mean and standard deviation of the data within the window, and then subtracts the window mean from the current sensor's raw value and divides it by the window standard deviation to obtain a normalized signal that eliminates the influence of dimensions. The normalized signal facilitates the unified processing and comparative analysis of data from different types of sensors.

[0058] For example, a sliding window can be represented as:

[0059] In the formula, Δ represents the sliding window, t represents the current time, and Δ represents the length of the sliding window.

[0060] The system collects raw sensor sequences within a preset sliding window of length Δ, and standardizes the data by calculating statistics within the window. Specifically, the window statistics can be expressed as:

[0061]

[0062] In the formula, This represents the mean value of channel j within the window; This represents the standard deviation of channel j within the window; Indicates the number of sampling points in the window; This represents the raw sensor data of channel j in cabin i at time p.

[0063] Correspondingly, the calculation of the normalized signal can be expressed as:

[0064] In the formula, This represents the normalized signal for channel j; This represents the original real-time signal value.

[0065] In one feasible implementation, an adaptive window length mechanism can be used for standardization. The system dynamically adjusts the time length of the sliding window based on the historical fluctuation characteristics of the sensor data and the current operating conditions. For sensors with relatively smooth data changes, a longer time window is used to improve statistical stability, while for sensors with more frequent data changes, a shorter time window is used to maintain detection sensitivity.

[0066] In step S320, the historical baseline is established by analyzing long-term historical data of the sensor under normal diving operation conditions, including characteristic parameters such as the statistical mean and standard deviation of the sensor under normal operating conditions. Establishing the historical baseline requires collecting continuous monitoring data of the sensor under normal operating conditions, removing data from abnormal operating conditions and maintenance periods, and determining the data distribution characteristics under normal conditions through statistical analysis methods.

[0067] For example, historical baselines for different types of sensors are statistically calculated based on their actual monitoring data under normal diving operation conditions to form corresponding mean and standard deviation benchmark parameters. The system calculates the relative deviation value as a long-term anomaly assessment indicator by comparing the statistical characteristics of the sensor data within the current window with the corresponding historical baseline characteristics.

[0068] For example, the calculation of the relative baseline deviation value can be expressed as:

[0069] In the formula, This indicates the relative deviation between the current window mean and the historical baseline mean; This represents the long-term historical mean of channel j; This represents the long-term historical standard deviation of channel j.

[0070] In one feasible implementation, a multi-condition historical baseline mechanism can be used to calculate the deviation value. The system establishes multiple sets of historical baselines according to different stages and conditions of diving operations, including the diving stage baseline, the working stage baseline, and the surfacing stage baseline. During the anomaly detection process, the corresponding historical baseline is selected according to the current operation stage to calculate the deviation value, thereby improving the accuracy and adaptability of anomaly detection.

[0071] In step S330, the rate of change refers to the ratio of the magnitude of the change in sensor data between adjacent time points to the time interval. It can be understood as a dynamic index that characterizes the intensity of data fluctuation and abrupt changes by calculating the time derivative of sensor data, and is used to identify instantaneous anomalies and sudden state changes in sensor data.

[0072] For example, the rate of change can be expressed as:

[0073] In the formula, This represents the local rate of change of channel j, used to indicate the intensity of signal fluctuations. This is the sampling interval time; This represents the sensor signal value that is one frame behind in time.

[0074] Specifically, the system can calculate the difference between the sensor value at the current moment and the sensor value at the previous sampling moment, and then divide it by the sampling time interval to obtain the instantaneous rate of change of the sensor. The absolute value of the rate of change reflects the intensity of fluctuation in the sensor data; a large rate of change usually indicates a rapid change in the equipment status or environmental conditions.

[0075] For example, for different types of sensors such as oxygen concentration sensors, cabin pressure sensors, and motor current sensors, the system calculates their corresponding rate of change indicators, and uses the magnitude of the rate of change to assess the degree of abrupt change and the risk of anomalies in the data of each sensor.

[0076] In one feasible implementation, a multi-order rate of change calculation method can be used. The system can calculate not only the first-order rate of change (e.g., velocity) but also the second-order rate of change (e.g., acceleration). By comprehensively analyzing the rates of change of different orders, the dynamic anomaly characteristics of the sensor data can be evaluated more comprehensively.

[0077] In step S340, the single-channel anomaly score refers to the comprehensive anomaly degree quantification value obtained by weighted fusion deviation value and change rate, which is used to measure the intensity of the abnormal state of a single sensor channel.

[0078] Optionally, the system presets deviation value weights and rate of change weights for each sensor channel, and calculates the single-channel anomaly score for that channel through weighted summation. The weight parameters for different types of sensors are configured differently according to their importance in the diving system and the anomaly detection sensitivity requirements.

[0079] The deviation value weight primarily reflects the importance of long-term trend anomalies in sensor data, while the rate of change weight primarily reflects the importance of instantaneous anomalous changes in sensor data. For sensors in critical safety equipment, the system can increase their weight configuration to enhance the sensitivity of anomaly detection.

[0080] In one feasible implementation, an anomaly score calculation can be performed using a dynamic weight adjustment method. The system dynamically adjusts the deviation value weight and the rate of change weight based on the current diving operation stage, environmental conditions, and equipment operating status, thereby optimizing the accuracy and reliability of anomaly detection under different operating conditions.

[0081] In step S350, the single-channel threshold can be set differently according to the technical specifications, environmental conditions, and safety risk levels of different sensor types to ensure the accuracy and timeliness of anomaly detection. When the anomaly score of any sensor channel in the cabin reaches or exceeds the corresponding threshold, the system automatically marks the cabin as the target cabin and generates an anomaly signal containing information such as the time of the anomaly, the type of anomaly, and the degree of anomaly.

[0082] For example, the abnormal signal includes detailed information such as the specific timestamp of the abnormality, the channel identifier of the abnormal sensor, the abnormality score value, and the abnormality type classification, which facilitates subsequent abnormality processing and video segment localization.

[0083] In one feasible implementation, the system can use a multi-level threshold mechanism to generate abnormal signals, setting two levels: a warning threshold and an alarm threshold. When the abnormal score reaches the warning threshold, a warning signal is generated; when the abnormal score reaches the alarm threshold, an emergency abnormal signal is generated. Different levels of abnormal signals correspond to different response strategies and processing priorities.

[0084] By adopting the embodiments of this application, through multi-dimensional anomaly feature extraction such as standardization processing, historical baseline deviation value calculation, rate of change calculation and weighted summation, it is possible to simultaneously capture long-term trend anomalies and instantaneous mutation anomalies in sensor data, thereby achieving comprehensive detection and accurate identification of abnormal states of diving systems.

[0085] Based on the above embodiments, as an optional embodiment, step S220, which involves multi-channel detection of sensor data from each cabin to generate an abnormal signal, may further include the following steps: Step S410: For any cabin, calculate the covariance matrix between different types of sensor signals in the cabin. The covariance matrix represents the coupling relationship between various sensors in the diving system. Step S420: Obtain the benchmark covariance matrix. The benchmark covariance matrix represents the coupling relationship between various sensors under normal saturation diving operation conditions. Step S430: Calculate the difference measure between the current covariance matrix and the benchmark covariance matrix; Step S440: Perform correlation analysis on the sensor signals between various sensors in the cabin to obtain the correlation coefficient; Step S450: Calculate the degree of deviation of the correlation coefficient from the baseline correlation coefficient to obtain the multi-channel anomaly score; Step S460: If the difference metric is greater than or equal to the multi-channel threshold and the multi-channel anomaly score is greater than or equal to the correlation threshold, the cabin is identified as the target cabin, and an anomaly signal for the target cabin is generated.

[0086] In step S410, the covariance matrix refers to the matrix that characterizes the coupling strength between sensors by statistically analyzing and calculating the signals of multiple different types of sensors in the cabin. It can be understood as a mathematical model that reflects the interdependence and linkage change patterns of the signals of various sensors in the diving system. It is used to capture system-level anomalies and multi-parameter coordinated change anomalies that cannot be identified by single-channel detection.

[0087] For example, the coupling relationships between sensors include, but are not limited to, the negative correlation coupling between oxygen concentration sensors and carbon dioxide concentration sensors, where carbon dioxide concentration decreases as oxygen concentration increases; the positive correlation coupling between cabin pressure sensors and underwater depth sensors, where cabin pressure increases as depth increases; the positive correlation coupling between motor current sensors and main pump power sensors, where main pump power increases as motor current increases; the positive correlation coupling between main pump power sensors and cooling water flow sensors, where cooling water flow needs to increase as main pump power increases; the positive correlation coupling between hydraulic system pressure sensors and motor current sensors, where motor current increases as hydraulic load increases; the correlation coupling between internal temperature and humidity of temperature and humidity sensors, where temperature changes affect relative humidity values; the compensating coupling between emergency gas supply pressure sensors and cabin pressure sensors, where cabin pressure changes when emergency gas supply is activated; and the negative correlation coupling between cable insulation monitoring sensors and temperature and humidity sensors, where cable insulation performance decreases as humidity increases.

[0088] Specifically, the above-mentioned coupling relationship exhibits a stable statistical correlation pattern under normal operating conditions. When equipment failure or system anomaly occurs, the original coupling relationship will be disrupted or changed. The changes in the coupling relationship can be captured by the covariance matrix, thereby enabling effective identification of complex anomalies and system-level faults.

[0089] Specifically, the system first constructs a signal matrix for all sensor signals within the cabin within a preset sliding time window, then eliminates the influence of differences in the dimensions of different sensors through standardization processing, and finally calculates the covariance matrix based on the standardized signal matrix to quantify the coupling strength between the sensors.

[0090] For example, the signal matrix can be represented as:

[0091] In the formula, This indicates that within the sliding time window of time t, all of the cabin i... The timing data from each sensor constitutes a signal matrix.

[0092] Correspondingly, the current covariance matrix can be expressed as:

[0093] In the formula, This represents a matrix composed of the standardized signals from each channel; This indicates the number of sampling points within the window.

[0094] For example, in the living quarters, there is a negative correlation between the oxygen concentration sensor and the carbon dioxide concentration sensor; in the equipment compartment, there is a positive correlation between the main pump power sensor and the cooling water flow sensor. The covariance matrix can quantify the changes in the strength of these couplings.

[0095] In one feasible implementation, a sliding update mechanism can be used to calculate the covariance matrix. The system maintains a data buffer of fixed length. When new sensor data arrives, the oldest data is removed and the new data is added. The covariance matrix is ​​updated through incremental calculation, which reduces computational complexity and improves real-time performance.

[0096] In step S420, the benchmark covariance matrix refers to the reference standard matrix established by analyzing the historical coupling relationship between each sensor under normal saturated diving operation conditions. It can be understood as a benchmark model that reflects the coordinated change law of each sensor signal under normal operating conditions of the diving system, and is used as a comparison benchmark to judge whether the current sensor coupling relationship is abnormal.

[0097] In one feasible implementation, a multi-condition benchmark covariance matrix can be used. The system establishes multiple sets of benchmark covariance matrices according to different stages of diving operations and environmental conditions. During multi-channel anomaly detection, the corresponding benchmark matrix is ​​selected for comparison and analysis based on the current operation status, thereby improving the accuracy of anomaly detection and environmental adaptability.

[0098] In step S430, the difference metric refers to a numerical index obtained by calculating the overall deviation between the current covariance matrix and the benchmark covariance matrix. It can be understood as a comprehensive evaluation value that quantifies the degree of deviation of the sensor coupling relationship from the normal state, and is used to identify abnormal changes in the cooperative relationship between sensor groups.

[0099] For example, the covariance variability measure can be expressed as:

[0100] In the formula, Represents the historical baseline covariance matrix of hull i; Let p represent the (p, q)th element of the matrix.

[0101] Specifically, the system calculates the squared difference between the current covariance matrix and the reference covariance matrix element by element, then sums and takes the square root to obtain the overall difference metric. A larger difference metric indicates that the current sensor coupling relationship has deviated from the normal state.

[0102] In one feasible implementation, a weighted difference metric calculation method can be used. The system assigns different weights to different elements of the covariance matrix according to the importance and sensitivity of different sensor pairs, giving greater attention to changes in the coupling relationship of key sensor pairs and improving the targeting and accuracy of anomaly detection.

[0103] In step S440, correlation analysis refers to the analytical process of quantifying the degree of linear correlation between different sensor signals in the cabin by calculating the statistical correlation coefficient. It can be understood as a technical means of evaluating the intensity of coordinated changes between sensor signals using statistical methods such as Pearson correlation coefficient, which is used to supplement covariance analysis from the perspective of correlation and provide a more comprehensive assessment of sensor coupling relationship.

[0104] For example, the correlation coefficient can be calculated as follows:

[0105] In the formula, This represents the Pearson correlation coefficient between channel p and channel q at time t; This represents the covariance between the two channels; These represent the standard deviations of the two channels, respectively.

[0106] Specifically, the system calculates pairwise correlations between all sensor signals within the cabin, forming a correlation coefficient matrix. Anomalies in the correlation relationships between sensors are identified by analyzing changes in the numerical and sign of these correlation coefficients. A positive correlation coefficient indicates that the two sensors are changing in the same direction, while a negative correlation coefficient indicates that they are changing in opposite directions. The absolute value of the correlation coefficient indicates the strength of the correlation.

[0107] In step S450, the system first calculates the correlation coefficient of each sensor pair at the current time, then compares it with the corresponding baseline correlation coefficient, and obtains a multi-channel anomaly score by calculating the average degree of deviation. The baseline correlation coefficient reflects the typical correlation level between sensors under normal operating conditions.

[0108] For example, when the correlation coefficient between the oxygen concentration sensor and the cabin pressure sensor changes from a normal weak positive correlation to a strong negative correlation, it indicates that there may be a gas leak or an abnormality in the pressure regulation system. In this case, the multi-channel anomaly score will increase significantly.

[0109] In one feasible implementation, the system can periodically update the baseline correlation coefficient based on the latest normal operation data, enabling the anomaly detection algorithm to adapt to the slow changes in the operating characteristics of the diving system and the effects of equipment aging, and maintain the long-term stability of detection accuracy.

[0110] In step S460, the system requires both the covariance difference metric and the multi-channel anomaly score to simultaneously meet their respective threshold conditions before determining the cabin as the target cabin. This dual-judgment method can reduce the false alarm rate.

[0111] Specifically, the multi-channel thresholds are set differently based on the sensor configuration of each cabin and historical anomaly data, while the correlation thresholds are dynamically adjusted based on the normal correlation variation range between sensors. When both conditions are met simultaneously, the system generates a multi-channel anomaly signal containing information such as the time of the anomaly, the sensor channels involved, and the degree of the anomaly.

[0112] For example, for the equipment compartment, when the covariance difference metric between the main pump power sensor and the cooling water flow sensor exceeds a threshold, and the deviation of their correlation coefficients exceeds a correlation threshold, the system determines that there may be a cooling system failure and generates a multi-channel abnormal signal for the equipment compartment.

[0113] In one feasible implementation, a hierarchical threshold mechanism can be used to generate abnormal signals. The system sets thresholds for three levels: minor, moderate, and severe abnormalities, and generates different levels of abnormal signals based on different combinations of covariance difference metrics and multi-channel abnormality scores.

[0114] By adopting the embodiments of this application, multi-channel detection based on covariance matrix and correlation analysis can identify system-level anomalies and sensor coupling anomalies that cannot be detected by single-channel detection, and improves the accuracy and reliability of anomaly detection through a dual threshold judgment method.

[0115] Based on the above embodiments, as an optional embodiment, at least one frame of the video data of the cabin includes a timestamp of the acquisition time and a status fingerprint identifier; the status fingerprint identifier is a digital fingerprint obtained by hashing the sensor data corresponding to the acquisition time. The target monitoring segment of the target cabin located based on the time window of the abnormal signal in step S230 may further include the following steps: Step S510: Determine the time interval corresponding to the abnormal signal. The time interval includes a first preset duration before the time of the abnormality and a second preset duration after the time of the abnormality. Step S520: Obtain candidate video frames whose timestamps are within the time interval from the video data of the target cabin; Step S530: Perform a hash operation on the sensor data within the time interval to generate a reference state fingerprint sequence; Step S540: Match and verify the state fingerprint identifier of the candidate video frame with the reference state fingerprint at the corresponding time. Step S550: Candidate video frames whose difference between the status fingerprint identifier and the reference status fingerprint is less than a preset matching threshold are identified as target monitoring segments.

[0116] In step S510, the time interval refers to a continuous time period extending forward and backward from the moment the anomaly occurred, used to limit the time range of the video clip search. The first preset duration is used to capture the device state change process before the anomaly occurred, and the second preset duration is used to record the system response process after the anomaly occurred. By setting a reasonable time interval, it can be ensured that the acquired video clips contain the complete occurrence process of the anomaly event, while avoiding invalid searches in all historical videos.

[0117] Specifically, the first preset duration and the second preset duration can be set differently according to the duration characteristics of different types of abnormal events. Gas concentration anomalies usually require a longer observation time to capture the gradual process, electrical equipment anomalies require a shorter recording time to capture transient phenomena, and hydraulic system anomalies require a medium-length time window to observe the pressure change process.

[0118] In step S520, candidate video frames refer to all video frames whose timestamps fall within the abnormal time interval. A set of video frames related to the abnormal event time is selected from the continuous video stream of the target cabin by comparing timestamps. The system iterates through the video data of the target cabin, compares the timestamp of each frame with the abnormal time interval, and extracts all video frames whose timestamps fall within the interval as candidate frames.

[0119] Specifically, the process of acquiring candidate video frames needs to ensure the accuracy of timestamps and the integrity of video frames. Establishing a timestamp-based index structure can accelerate the acquisition process of candidate video frames and avoid sequential scanning of all video data.

[0120] In step S530, the reference state fingerprint sequence refers to the fingerprint sequence formed by hashing sensor data in chronological order within the abnormal time interval, used to provide a reference for verifying the accuracy of candidate video frames. The system extracts sensor data at each moment within the time interval, performs hashing operations on the multi-dimensional sensor data vector at each moment to generate corresponding state fingerprints, and arranges them in chronological order to form the reference state fingerprint sequence.

[0121] Specifically, hashing converts multidimensional sensor data into fixed-length digital fingerprints, ensuring that the same sensor state generates the same fingerprint, and different sensor states generate different fingerprints. The system standardizes the sensor data before performing hashing to eliminate the impact of differences in the dimensions of different sensors on fingerprint generation.

[0122] In step S540, the matching verification refers to the verification process of determining the validity of a video frame by comparing the similarity between the state fingerprint identifier carried by the candidate video frame and the reference state fingerprint at the corresponding time. For each candidate video frame, the system extracts its state fingerprint identifier, finds the reference fingerprint at the corresponding time in the reference state fingerprint sequence based on the timestamp of the frame, and calculates the similarity between the two fingerprints.

[0123] Specifically, the matching verification uses fingerprint comparison technology to confirm the spatiotemporal consistency between video frames and sensor states, eliminating mismatches between video frames and sensor states caused by time synchronization errors or data transmission delays. Similarity calculation can employ methods such as Hamming distance, edit distance, or cosine similarity, with the appropriate similarity calculation method selected based on the fingerprint encoding method.

[0124] In step S550, the system filters candidate video frames whose state fingerprint similarity is higher than a preset matching threshold to form the final target monitoring segment. The preset matching threshold is used to distinguish the similarity boundary between valid and invalid video frames, ensuring that each frame in the target monitoring segment is highly consistent with the sensor state at the abnormal moment.

[0125] Specifically, the target monitoring segment is composed of all valid video frames that have passed state fingerprint verification, arranged chronologically to form a continuous video segment that precisely corresponds to the abnormal event. Through the dual constraints of timestamp matching and state fingerprint verification, it is ensured that the located video segment accurately reflects the actual situation at the time the abnormality occurred.

[0126] By adopting the embodiments of this application, the state fingerprint identifier embedded in the video data provides a basis for the accurate association between video frames and sensor states. Through the dual constraints of time interval limitation and state fingerprint verification, the rapid and accurate location of video segments corresponding to abnormal events is achieved, avoiding the problem of missing spatiotemporal correlation in traditional video retrieval, and improving the efficiency and accuracy of video retrieval for abnormal events in diving systems.

[0127] Based on the above embodiments, as an optional embodiment, in order to further improve the accuracy and reliability of target monitoring segments and eliminate the influence of visual noise caused by factors such as ambient light fluctuations, underwater suspended object interference, or camera equipment shaking on the anomaly detection results, the above method may further include the following steps: Step S610: Calculate the short-time energy change value of each video frame in the target monitoring segment. The short-time energy change value represents the degree of fluctuation of the brightness of the video frame in the time dimension. Step S620: Calculate the structural similarity index between adjacent video frames in the target monitoring segment. The structural similarity index represents the magnitude of change in the content of the video frame. Step S630: Calculate the visual interference assessment value based on the short-time energy change value and the structural similarity index; Step S640: Determine the visual interference judgment threshold based on the environmental parameters of the target cabin. The environmental parameters include at least one of the cabin lighting status, underwater operating depth, and humidity conditions. Step S650: If the visual interference assessment value is greater than or equal to the visual interference judgment threshold, the corresponding video frame is marked as an environmental interference frame and its weight in the target monitoring segment is adjusted.

[0128] In step S610, the short-time energy change value refers to a numerical index that quantifies the degree of brightness fluctuation of the picture by analyzing the intensity of the change in brightness information of the video frame at continuous time points. It can be understood as a characteristic parameter that reflects the brightness stability of the video picture in the time dimension, and is used to identify visual interference phenomena caused by light flicker, reflection changes or abnormal lighting equipment.

[0129] Specifically, the system first extracts the overall brightness information of each frame in the target monitoring segment, and then assesses the severity of brightness changes by calculating the brightness difference between adjacent video frames. Short-time energy change values ​​can effectively capture abnormal fluctuations in image brightness caused by environmental factors such as cabin lighting system malfunctions, water surface light reflection, bubble obstruction, or lens contamination.

[0130] For example, when the emergency lighting inside the living quarters is turned on or off, there will be a significant change in brightness, and the short-term energy change value will increase significantly. When the diving bell is monitored, the brightness will also fluctuate when the angle of the underwater operation lights is adjusted or when suspended particles in the water block the light.

[0131] In one feasible implementation, a multi-scale energy change analysis method can be used to calculate short-term energy change values. The system simultaneously analyzes the brightness change characteristics at different time scales and comprehensively evaluates the brightness stability of video frames by integrating short-term abrupt changes and medium-term trend changes.

[0132] In step S620, the structural similarity index refers to the similarity measure that quantifies the degree of change in the image content by comparing the image structural features between adjacent video frames. It can be understood as a parameter that evaluates the structural information such as image texture, edge and shape between video frames, and is used to identify abnormalities in the image structure caused by camera shake, lens shift or changes in shooting angle.

[0133] Specifically, the system calculates the structural similarity between two video frames by extracting image structural features from adjacent frames, including edge information, texture patterns, and local feature point distribution. A low structural similarity index indicates significant differences in image content between adjacent frames, which may be caused by abnormal factors.

[0134] For example, when the camera in the equipment compartment shifts position due to vibration or loosening of the fixing, the structural similarity index between consecutive video frames will decrease significantly; when the monitoring screen in the escape pod is obstructed by the rapid movement of people, it will also lead to changes in structural similarity.

[0135] Correspondingly, the system can also analyze the positional changes of key equipment or fixed reference objects in the image to assist in structural similarity calculation, and evaluate the stability of the image by tracking the positional shifts of these stable feature points.

[0136] In one feasible implementation, a regional structural similarity analysis method can be adopted. The system divides the video frame into multiple regions and calculates the structural similarity of each region. By analyzing the similarity differences between different regions, local interference and global interference can be identified, thereby improving the precision of interference detection.

[0137] In step S630, the visual interference evaluation value refers to the comprehensive interference intensity quantification index formed by fusing short-time energy change value and structural similarity index. It can be understood as a composite evaluation parameter that comprehensively evaluates the quality of video frames by simultaneously considering two dimensions: brightness fluctuation and structural change.

[0138] Specifically, the system combines short-time energy change values ​​and structural similarity indicators through weighted fusion to form a single evaluation value that reflects the overall interference level of the video frame. Different weight configurations can be used for different types of cabin environments and monitoring scenarios to adapt to their respective interference characteristics and detection requirements.

[0139] For example, for living quarters where lighting conditions change frequently, the system can increase the weight of short-term energy change values; for equipment quarters with many devices that are prone to shading, the system can increase the weight of structural similarity indicators.

[0140] In one feasible implementation, a dynamic weight adjustment method can be used to calculate the visual interference assessment value. The system dynamically adjusts the fusion weight of the two indicators based on the current working status of the cabin, environmental conditions, and historical interference patterns to achieve adaptive interference assessment.

[0141] In step S640, the visual interference determination threshold refers to the critical value determined based on the specific environmental conditions and working characteristics of the target cabin to distinguish between normal and interfering images.

[0142] Optionally, the system can consider multiple environmental factors such as the type and working status of the cabin lighting equipment, the current underwater operating depth and lighting conditions, and the impact of cabin humidity levels on lens clarity, and determine the appropriate judgment threshold for the current conditions through the mapping relationship between environmental parameters and thresholds.

[0143] For example, during deep-water operations, where natural light is insufficient and artificial lighting is the primary means of illumination, the system will increase its tolerance for brightness variations. In high-humidity environments, where fogging can easily affect image clarity, the system will appropriately reduce the requirements for determining structural similarity.

[0144] Correspondingly, the environmental characteristics of different types of cabins vary greatly. Living cabins usually have relatively stable lighting conditions, while equipment cabins may experience changes in light due to equipment operation. Diving bells are more significantly affected by the underwater environment, and the lighting conditions of escape cabins can change drastically in emergency situations.

[0145] In one feasible implementation, a dynamic mapping table between environmental parameters and judgment thresholds can be established. The system can automatically query the corresponding threshold configuration based on the environmental parameters monitored in real time, and supports optimization and adjustment of the mapping relationship based on historical data and experience feedback.

[0146] In step S650, an environmental interference frame refers to a video frame that is identified as being affected by environmental factors and may affect the accuracy of anomaly detection. It can be understood as a video frame whose picture quality is degraded due to factors other than equipment failure or abnormal operation.

[0147] Specifically, when the visual interference assessment value exceeds the judgment threshold, the system marks the corresponding video frame as an environmental interference frame and adjusts the analysis weight of the frame in the target monitoring segment according to the degree of interference. The weight adjustment can adopt different methods such as linear decay, exponential decay, or step adjustment to ensure that the interference frame will not mislead the accurate identification of abnormal events.

[0148] Optionally, weight adjustments not only affect the contribution of video frames to anomaly analysis, but also the assessment of the continuity and integrity of video segments. The system needs to comprehensively consider both the quality of individual frames and the temporal coherence of the overall segment when adjusting weights.

[0149] In one feasible implementation, a hierarchical weight adjustment strategy can be adopted. The system sets different levels of weight adjustment range according to the type and intensity of interference. For minor interference that can be recovered, a smaller weight adjustment is used, while for severe interference that cannot be recovered, a larger weight adjustment is used or it is completely eliminated.

[0150] By employing the embodiments of this application, environmental interference frames in target monitoring segments can be effectively identified and processed. Through dual analysis of short-time energy change values ​​and structural similarity indicators, accurate detection of various interference types, such as illumination fluctuations and equipment jitter, can be achieved. Combined with adaptive threshold adjustment of environmental parameters, the interference detection algorithm maintains stable performance under different cabin environments and operating conditions.

[0151] Based on the above embodiments, as an optional embodiment, in order to reduce duplicate alarms, improve the systematicness and accuracy of abnormal event identification, and avoid misidentifying multiple anomalies caused by the same root cause as multiple independent events, the above method may further include the following steps: Step S710: Perform spatiotemporal correlation analysis on the detected multiple abnormal signals, and calculate the causal correlation degree based on the occurrence time interval of the abnormal signals and the similarity of sensor data features; Step S720: If the causal correlation is greater than a preset correlation threshold, the associated abnormal signals are merged into the same abnormal event.

[0152] In step S710, spatiotemporal correlation analysis refers to the analysis process of identifying potential causal relationships between signals by comprehensively considering the distribution characteristics of abnormal signals in the time and space dimensions. It can be understood as judging whether multiple abnormal signals originate from the same root cause based on the timing characteristics of the occurrence of anomalies and the location characteristics of the cabin, and is used to discover cascading faults across cabins.

[0153] Optionally, temporal correlation identifies possible causal propagation paths by analyzing the occurrence times of different anomalous signals. When multiple anomalous signals occur successively within a short time interval, it often indicates the existence of fault propagation or a chain reaction. Spatial correlation identifies physically or functionally related anomalous patterns by analyzing the cabin locations and system connections involved in the anomalous signals.

[0154] For example, when the main pump power sensor in the equipment compartment detects an anomaly first, and then the cabin pressure sensor and oxygen concentration sensor in the living quarters alarm in succession, the timing relationship and functional correlation indicate that there may be a chain of anomalies caused by the main pump failure.

[0155] Correspondingly, sensor data feature similarity refers to a quantitative index that assesses the similarity of anomaly types by comparing sensor data patterns at different times corresponding to different anomaly signals. It can be understood as the degree of consistency in the trend and amplitude characteristics of the changes in sensor data when an anomaly occurs, and is used to identify signal groups with the same anomaly feature patterns.

[0156] Optionally, the system extracts sensor data change characteristics before and after each abnormal signal occurs, including characteristic parameters such as numerical change trend, change amplitude, and duration. The consistency of the abnormality type is evaluated by calculating the similarity of these characteristic parameters between different abnormal signals.

[0157] For example, when the hydraulic system pressure sensor and the motor current sensor in the diving bell simultaneously detect an abnormal upward trend, their sensor data characteristics are highly similar, indicating that they may belong to the same hydraulic system failure event.

[0158] Among them, the causal correlation degree refers to the index of the strength of the causal relationship between anomalous signals formed by combining time interval and feature similarity information. It can be understood as an assessment parameter reflecting the probability that multiple anomalous signals belong to the same event.

[0159] Specifically, the system calculates causal correlation by weighting time correlation and feature similarity. Shorter time intervals indicate a weaker causal relationship, while higher feature similarity indicates a more consistent anomaly type. Different types of system faults have different propagation time characteristics, and the system can adjust the time weights according to the fault type.

[0160] In one feasible implementation, a multi-level correlation analysis method can be used to calculate the degree of causal correlation. The system first performs direct correlation analysis to identify obvious causal relationships, then performs indirect correlation analysis to discover correlations propagated through intermediate links, and finally combines direct and indirect correlations to form the final degree of causal correlation.

[0161] In step S720, abnormal event merging refers to the process of merging multiple abnormal signals with high causal correlation into a single abnormal event record. It can be understood as a technical means of reorganizing scattered abnormal signals into logically unified event units based on the correlation analysis results, which is used to simplify abnormal event management and improve fault analysis efficiency.

[0162] Specifically, when the causal correlation calculated by the system reaches or exceeds a preset correlation threshold, it indicates that the corresponding abnormal signals have a sufficiently strong correlation and should be merged into the same abnormal event. The merged abnormal event includes the time information, cabin information, sensor information, and anomaly severity information of all relevant abnormal signals.

[0163] For example, when abnormal oxygen concentration, abnormal carbon dioxide concentration, and abnormal cabin pressure occur successively within a short period of time and have similar change patterns, the system will classify these three abnormal signals into "habitat gas system abnormal event" to facilitate unified processing and analysis by operators.

[0164] Correspondingly, the preset association threshold can be set differently according to the system complexity, abnormal history patterns and safety requirements of different cabins. For critical systems with high safety requirements, a lower association threshold can be set to improve the sensitivity of abnormal event merging, while for systems with strong independence, a higher association threshold can be set to avoid excessive merging.

[0165] After merging abnormal events, the system will regenerate a unified event identifier, determine the start and end times of the event, summarize the information of the involved cabins and sensors, assess the overall severity of the event, and generate corresponding target monitoring segments for the merged abnormal events.

[0166] In one feasible implementation, a hierarchical merging strategy can also be used for abnormal event handling. The system first merges abnormal signals within a single cabin to identify system-level abnormalities within a single cabin; then it merges abnormal signals across cabins to identify global abnormalities affecting multiple cabins; finally, a hierarchical abnormal event structure is formed to facilitate management and response at different levels.

[0167] By employing the embodiments of this application, multiple related abnormal signals can be effectively identified and integrated, avoiding information interference to operators caused by repeated alarms due to the same fault cause. Through spatiotemporal correlation analysis and causal correlation calculation, automatic identification and tracking of fault propagation paths in complex systems are achieved.

[0168] Figure 3 This schematic diagram illustrates a structural block diagram of a video monitoring device for a diving system provided in this application, which may include: The acquisition module is used to acquire sensor data and video data from each compartment in the diving system; the sensor data represents the status of the equipment in the compartment; the video data represents multi-channel monitoring video synchronized with the sensor data in time. The detection module is used to detect anomalies in the sensor data of each cabin and generate anomaly signals. Anomaly detection includes single-channel detection and multi-channel detection. Single-channel detection is based on the degree of deviation of sensor data from historical baseline. Multi-channel detection is based on the changes in the correlation between different types of sensor signals in the sensor data. The determination module is used to locate the target monitoring segment of the target cabin based on the time window of the abnormal signal in response to the abnormal signal; wherein, the target monitoring segment is determined by timestamp matching and status fingerprint verification; the status fingerprint is a feature identifier calculated based on sensor data at the time of the abnormality.

[0169] Based on the above embodiments, as an optional embodiment, the detection module is further configured to: standardize the sensor data of any cabin within a preset sliding time window to obtain a normalized signal; calculate the deviation of the normalized signal relative to the historical baseline; the historical baseline represents the statistical mean and standard deviation of the corresponding sensor under normal diving operation conditions; calculate the rate of change of the sensor data of the cabin; the rate of change of the sensor data represents the degree of abrupt change in the sensor value per unit time; perform a weighted summation of the deviation value corresponding to the cabin and the rate of change of the sensor data to obtain a single-channel anomaly score; if the single-channel anomaly score is greater than or equal to the single-channel threshold, determine the cabin as the target cabin, and generate an anomaly signal for the target cabin.

[0170] Based on the above embodiments, as an optional embodiment, the detection module is further configured to, for any cabin, calculate the covariance matrix between different types of sensor signals in the cabin, the covariance matrix representing the coupling relationship between various sensors in the diving system; obtain a benchmark covariance matrix, the benchmark covariance matrix representing the coupling relationship between various sensors under normal saturation diving operation conditions; calculate the difference measure between the current covariance matrix and the benchmark covariance matrix; perform correlation analysis on the sensor signals between various sensors in the cabin to obtain the correlation coefficient; calculate the degree of deviation of the correlation coefficient relative to the baseline correlation coefficient to obtain a multi-channel anomaly score; and determine the cabin as the target cabin if the difference measure is greater than or equal to the multi-channel threshold and the multi-channel anomaly score is greater than or equal to the correlation threshold, and generate anomaly signals for the target cabin.

[0171] Based on the above embodiments, as an optional embodiment, at least one frame in the video data of the cabin includes a timestamp of the acquisition time and a status fingerprint identifier; the status fingerprint identifier is a digital fingerprint obtained by hashing the sensor data corresponding to the acquisition time; the determining module is further configured to determine the time interval corresponding to the abnormal signal, the time interval including a first preset duration before the time of the abnormality and a second preset duration after the time of the abnormality; acquire candidate video frames whose timestamps are located within the time interval from the video data of the target cabin; perform a hash operation on the sensor data within the time interval to generate a reference status fingerprint sequence; match and verify the status fingerprint identifier of the candidate video frame with the reference status fingerprint at the corresponding time; and determine the candidate video frame whose difference value between the status fingerprint identifier and the reference status fingerprint is less than a preset matching threshold as the target monitoring segment.

[0172] Based on the above embodiments, as an optional embodiment, the determining module is further configured to calculate the short-time energy change value of each video frame in the target monitoring segment, wherein the short-time energy change value characterizes the degree of fluctuation of the video frame brightness in the time dimension; calculate the structural similarity index between adjacent video frames in the target monitoring segment, wherein the structural similarity index characterizes the change range of the video frame content; calculate the visual interference assessment value based on the short-time energy change value and the structural similarity index; determine the visual interference judgment threshold according to the environmental parameters of the target cabin, wherein the environmental parameters include at least one of the cabin lighting status, underwater operating depth, and humidity conditions; and, if the visual interference assessment value is greater than or equal to the visual interference judgment threshold, mark the corresponding video frame as an environmental interference frame and adjust its weight in the target monitoring segment.

[0173] Based on the above embodiments, as an optional embodiment, the video monitoring device of the diving system further includes an adjustment module, which is used to perform spatiotemporal correlation analysis on multiple detected abnormal signals, calculate the causal correlation degree based on the occurrence time interval of the abnormal signals and the similarity of sensor data features; and merge the associated abnormal signals into the same abnormal event when the causal correlation degree is greater than a preset correlation threshold.

[0174] Figure 4 The diagram illustrates a structural block diagram of an electronic device suitable for implementing the methods described above, according to an embodiment of this application. Figure 4 The electronic device shown is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of this application.

[0175] like Figure 4As shown, an electronic device according to an embodiment of this application includes a processor 401, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 402 or a program loaded from a storage portion 408 into a random access memory (RAM) 403. The processor 401 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or an associated chipset and / or a special-purpose microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 401 may also include onboard memory for caching purposes. The processor 401 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of this application.

[0176] RAM 403 stores various programs and data required for the operation of the electronic device. Processor 401, ROM 402, and RAM 403 are interconnected via bus 404. Processor 401 executes various operations of the method flow according to embodiments of this application by executing programs in ROM 402 and / or RAM 403. It should be noted that the programs may also be stored in one or more memories other than ROM 402 and RAM 403. Processor 401 may also execute various operations of the method flow according to embodiments of this application by executing programs stored in said one or more memories.

[0177] According to embodiments of this application, the electronic device may further include an input / output (I / O) interface 405, which is also connected to a bus 404. The electronic device may also include one or more of the following components connected to the input / output (I / O) interface 405: an input section 406 including a keyboard, mouse, etc.; an output section 407 including a cathode ray tube (CRT), liquid crystal display (LCD), etc., and a speaker, etc.; a storage section 408 including a hard disk, etc.; and a communication section 409 including a network interface card such as a LAN card, modem, etc. The communication section 409 performs communication processing via a network such as the Internet. A drive 410 is also connected to the input / output (I / O) interface 405 as needed. A removable medium 411, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on the drive 410 as needed so that computer programs read from it can be installed into the storage section 408 as needed.

[0178] This application also provides a computer-readable storage medium, which may be included in the device / system / system described in the above embodiments; or it may exist independently and not assembled into the device / system / system. The aforementioned computer-readable storage medium carries one or more programs, which, when executed, implement the method according to the embodiments of this application.

[0179] According to embodiments of this application, the computer-readable storage medium can be a non-volatile computer-readable storage medium. Examples include, but are not limited to: portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, the computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device.

[0180] For example, according to embodiments of this application, a computer-readable storage medium may include the ROM 402 and / or RAM 403 described above and / or one or more memories other than ROM 402 and RAM 403.

[0181] Embodiments of this application also include a computer program product comprising a computer program containing program code for performing the methods provided in the embodiments of this application. When the computer program product is run on an electronic device, the program code is used to enable the electronic device to implement the methods provided in the embodiments of this application.

[0182] When the computer program is executed by the processor 401, it performs the functions defined in the system / system of the embodiments of this application. According to the embodiments of this application, the systems, modules, units, etc., described above can be determined by computer program modules.

[0183] In one embodiment, the computer program may rely on a tangible storage medium such as an optical storage device or a magnetic storage device. In another embodiment, the computer program may also be transmitted and distributed in the form of signals over a network medium, and downloaded and installed via communication section 409, and / or installed from removable medium 411. The program code contained in the computer program can be transmitted using any suitable network medium, including but not limited to: wireless, wired, etc., or any suitable combination thereof.

[0184] The embodiments of this application have been described above. However, these embodiments are merely illustrative and not intended to limit the scope of this application. Although various embodiments have been described above, this does not mean that the measures in the various embodiments cannot be used advantageously in combination. The scope of this application is defined by the appended claims and their equivalents. Various substitutions and modifications can be made by those skilled in the art without departing from the scope of this application, and all such substitutions and modifications should fall within the scope of this application.

Claims

1. A video monitoring method for a diving system, characterized in that, Applied to a central control system, the method includes: The system acquires sensor data and video data from each compartment of the diving system; the sensor data represents the status of equipment in the compartment; the video data represents multi-channel monitoring video synchronized with the sensor data in time. Anomaly detection is performed on the sensor data of each of the aforementioned cabins to generate anomaly signals; the anomaly detection includes single-channel detection and multi-channel detection; the single-channel detection is based on the degree of deviation of the sensor data from the historical baseline; the multi-channel detection is based on the changes in the correlation between different types of sensor signals in the sensor data. In response to an abnormal signal for the target cabin, a target monitoring segment of the target cabin is located based on the time window of the abnormal signal; wherein, the target monitoring segment is determined by timestamp matching and status fingerprint verification; the status fingerprint is a feature identifier calculated based on sensor data at the time of the abnormality.

2. The method according to claim 1, characterized in that, The sensor data includes at least one of gas concentration data, chamber pressure data, electrical equipment operating parameters, and diver physiological monitoring data; the diving system's chamber includes: The living quarters are equipped with oxygen concentration sensors, carbon dioxide concentration sensors, cabin pressure sensors, and temperature and humidity sensors. The diving bell is equipped with a hydraulic system pressure sensor, a motor current sensor, and an underwater depth sensor. The escape capsule is equipped with an emergency gas supply pressure sensor and a door status sensor. The equipment compartment is equipped with a main pump power sensor, a cooling water flow sensor, and a cable insulation monitoring sensor. Each of the aforementioned cabins is equipped with a camera and an edge node. The camera is used to collect video from the corresponding cabin to form the multi-channel monitoring video. The edge node is communicatively connected to the camera and sensors of the corresponding cabin. The central control system is communicatively connected to the edge nodes of each of the cabins.

3. The method according to claim 2, characterized in that, The process of performing single-channel detection on sensor data from each cabin to generate abnormal signals includes: For any of the aforementioned cabins, the sensor data of the cabins are standardized within a preset sliding time window to obtain a normalized signal. Calculate the deviation of the normalized signal relative to the historical baseline; the historical baseline represents the statistical mean and standard deviation of the corresponding sensor under normal diving operation conditions; Calculate the rate of change of the sensor data of the cabin; the rate of change of the sensor data characterizes the degree of abrupt change in the sensor values ​​per unit time. The deviation value corresponding to the cabin and the rate of change of the sensor data are weighted and summed to obtain a single-channel anomaly score; If the single-channel anomaly score is greater than or equal to the single-channel threshold, the cabin is determined to be the target cabin, and an anomaly signal is generated for the target cabin.

4. The method according to claim 2, characterized in that, The process of performing multi-channel detection on sensor data from each cabin to generate abnormal signals includes: For any of the aforementioned cabins, calculate the covariance matrix between different types of sensor signals within the cabin, whereby the covariance matrix characterizes the coupling relationship between various sensors in the diving system. Obtain the benchmark covariance matrix, which characterizes the coupling relationship between various sensors under normal saturation diving operation conditions; Calculate the difference measure between the current covariance matrix and the reference covariance matrix; Correlation analysis was performed on the sensor signals of the various sensors in the cabin to obtain the correlation coefficients; The degree of deviation of the correlation coefficient from the baseline correlation coefficient is calculated to obtain a multi-channel anomaly score; If the difference metric is greater than or equal to the multi-channel threshold and the multi-channel anomaly score is greater than or equal to the correlation threshold, the cabin is determined to be the target cabin, and an anomaly signal for the target cabin is generated.

5. The method according to claim 2, characterized in that, At least one frame of the video data of the cabin includes a timestamp of the acquisition time and a status fingerprint identifier; the status fingerprint identifier is a digital fingerprint obtained by hashing the sensor data corresponding to the acquisition time. The target monitoring segment for locating the target cabin based on the time window of the abnormal signal includes: Determine the time interval corresponding to the abnormal signal, wherein the time interval includes a first preset duration before the time of the abnormality and a second preset duration after the time of the abnormality; Candidate video frames whose timestamps are located within the time interval are obtained from the video data of the target cabin; Perform a hash operation on the sensor data within the time interval to generate a reference state fingerprint sequence; The state fingerprint identifier of the candidate video frame is matched and verified with the reference state fingerprint at the corresponding time. Candidate video frames whose difference between the status fingerprint identifier and the reference status fingerprint is less than a preset matching threshold are identified as the target monitoring segments.

6. The method according to claim 5, characterized in that, The method further includes: Calculate the short-time energy change value of each video frame in the target monitoring segment, whereby the short-time energy change value characterizes the degree of fluctuation in the brightness of the video frame over time. Calculate the structural similarity index between adjacent video frames in the target monitoring segment, whereby the structural similarity index characterizes the magnitude of change in the content of the video frame. The visual interference assessment value is calculated based on the short-time energy change value and the structural similarity index; The visual interference threshold is determined based on the environmental parameters of the target cabin, wherein the environmental parameters include at least one of the cabin lighting status, underwater operating depth, and humidity conditions. If the visual interference assessment value is greater than or equal to the visual interference determination threshold, the corresponding video frame is marked as an environmental interference frame and its weight in the target monitoring segment is adjusted.

7. The method according to claim 1, characterized in that, The method further includes: Spatiotemporal correlation analysis was performed on multiple detected abnormal signals, and the causal correlation degree was calculated based on the time interval between the occurrence of abnormal signals and the similarity of sensor data features. If the causal correlation is greater than a preset correlation threshold, the associated abnormal signals are merged into the same abnormal event.

8. A video monitoring device for a diving system, characterized in that, include: The acquisition module is used to acquire sensor data and video data from each chamber in the diving system. The sensor data characterizes the status of equipment in the ship's cabin; The video data represents a multi-channel monitoring video that is time-synchronized with the sensor data; The detection module is used to detect anomalies in the sensor data of each of the cabins and generate anomaly signals. The anomaly detection includes single-channel detection and multi-channel detection. The single-channel detection is based on the degree of deviation of the sensor data from the historical baseline. The multi-channel detection is based on the changes in the correlation between different types of sensor signals in the sensor data. The determination module is used to locate the target monitoring segment of the target cabin based on the time window of the abnormal signal in response to the abnormal signal; wherein the target monitoring segment is determined by timestamp matching and status fingerprint verification; the status fingerprint is a feature identifier calculated based on sensor data at the time of the abnormality.

9. An electronic device, comprising: One or more processors; Memory, used to store one or more programs. Wherein, when the one or more programs are executed by the one or more processors, the one or more processors implement the method of any one of claims 1 to 7.

10. A computer-readable storage medium having executable instructions stored thereon, which, when executed by a processor, cause the processor to perform the method of any one of claims 1 to 7.

Images