Family security prediction method and device, electronic equipment and storage medium

Abstract

The invention discloses a home security prediction method and device, electronic equipment and a storage medium, and relates to the technical field of data processing. Visual and non-visual data can be fused to generate perceptual event features, and a dynamic causal knowledge graph including home entities, events and situational states is constructed; deep modeling of association between the overall state of the family and the event causality is realized; complex scene anomalies can be captured through causal chain template matching and anomaly marking, multiple groups of causal hypotheses are generated, an optimal result is screened, explainable early warning containing causal logic is generated in combination with risk scores, and the limitation of a traditional static rule or a single sensor is avoided, so that the method is suitable for popularization and application. The technical problem that an existing traditional scheme lacks deep modeling capability of family overall state, member behavior mode and causal association between events and is difficult to adapt to complex dynamic scenes can be solved. The technical effects of improving the adaptability of the home security system to complex dynamic scenes, improving the early warning accuracy and interpretability and enhancing the intelligent level of the system are achieved.

Description

Technical Field

[0001] This disclosure relates to the field of data processing technology, and in particular to a predictive method and apparatus for home security, electronic equipment, and storage medium. Background Technology

[0002] Home security systems, as an important component of smart homes, are widely used in security monitoring, behavior recognition, and environmental sensing. With the development of computer vision and IoT technologies, existing systems are gradually integrating video analytics functions such as motion detection and personnel recognition, and introducing specific behavior recognition technologies such as fall detection. Specifically, this technology system covers the entire process from sensor data acquisition to event triggering, including key aspects such as passive infrared detection, door magnetic sensing, and video stream processing. Traditional solutions often employ static rules or single-sensor triggering mechanisms, directly generating warning signals based on preset conditions (such as changes in door and window status), lacking the ability to deeply model the overall state of the home environment, member behavior patterns, and causal relationships between events, making the system difficult to adapt to complex dynamic scenarios. Summary of the Invention

[0003] This disclosure provides a predictive method, device, electronic device, and storage medium for home security. Its main purpose is to address the problem that the lack of deep modeling capabilities for the overall state of the home environment, member behavior patterns, and causal relationships between events makes it difficult for the system to adapt to complex and dynamic scenarios.

[0004] According to a first aspect of this disclosure, a method for predicting home security is provided, comprising: Visual data and non-visual data are collected, target detection and behavior analysis are performed on the visual data, the non-visual data are standardized, and the processed visual data and non-visual data are fused to generate a first perceptual event feature. A dynamic causal knowledge graph is constructed based on the features of the first perceived event; wherein, the knowledge graph includes nodes representing family entities, events and situational states; and edges representing the relationships between nodes; Receive the second perceived event features and match the second perceived event features with the preset causal chain template in the knowledge graph; If the match fails, the second perceived event feature is marked as a causal abnormal event and an associated node is generated; Based on the dynamic causal knowledge graph, multiple sets of causal explanation hypotheses are generated, the posterior probability of each hypothesis is calculated, and the hypothesis with the highest posterior probability is selected as the causal explanation result. Based on the posterior probability of the causal explanation results, an uncertainty measure and risk score are calculated, and interpretable early warning information containing causal logic is generated to issue warnings to users.

[0005] Optionally, the process of collecting visual and non-visual data, performing target detection and behavior analysis on the visual data, standardizing the non-visual data, and fusing the processed visual and non-visual data to generate the first perceptual event feature includes: The object detection algorithm is used to identify people and objects in the image and output category labels and location coordinates; the pose estimation algorithm is used to extract the coordinates of key points of the human skeleton, and the pose changes in continuous frames are analyzed by combining a time series model to identify human behavior patterns and generate behavior labels. The extreme value normalization method is used to map environmental parameters and device operation data in non-visual data to a preset value range, and a timestamp accurate to milliseconds is added to each processed data, which is then encapsulated as environmental status information in key-value pair format. By aggregating personnel information, behavioral tags, and environmental status information within the same time window, structured perception event features with multidimensional information are generated.

[0006] Optionally, the step of generating multiple sets of causal explanation hypotheses based on the dynamic causal knowledge graph, calculating the posterior probability of each hypothesis, and selecting the hypothesis with the highest posterior probability as the causal explanation result includes: The dynamic time warping algorithm is used to calculate the similarity between the event sequence corresponding to the features of newly perceived events and the preset causal chain template; If the similarity is lower than a preset threshold, it is determined to be a causal anomaly, a causal anomaly event node is generated, and an association edge is established between the node and the event node that triggered the anomaly. Based on historical event records, current situational state probability distribution, and entity attributes in the knowledge graph, multiple mutually exclusive causal explanation hypotheses are generated. The prior probability of each hypothesis is calculated using the frequency of historical events in the knowledge graph as the hypothesis, and the posterior probability of each hypothesis is calculated using the characteristics of the current perceived event and the probability of the situational state as evidence. The hypothesis with the highest posterior probability is selected as the final causal explanation, and the relationship between the hypothesis and the evidence is recorded and stored in the knowledge graph.

[0007] Optionally, the step of calculating an uncertainty measure and a risk score based on the posterior probability of the causal explanation result, generating interpretable early warning information containing causal logic, and executing an early warning for the user includes: Warning levels are classified according to risk scores and corresponding push notification methods are assigned: different warning levels correspond to different push notification methods. Obtain user feedback and update the knowledge graph based on the user feedback.

[0008] Optionally, the method further includes: When no event is detected within a preset number of consecutive time windows, non-visual data for the current time period is collected as the environmental baseline value, including temperature, humidity, gas concentration, and equipment standby power parameters. Calculate the drift coefficients of real-time environmental parameters relative to baseline values; If the drift coefficient exceeds the preset threshold, the environmental parameters are determined to be abnormal, an environmental parameter drift event node is generated, the attributes of the corresponding environmental state node in the knowledge graph are updated, and a low-level warning is triggered to prompt the user to check the environment or equipment.

[0009] According to a second aspect of this disclosure, a predictive device for home security is provided, comprising: The acquisition unit is used to acquire visual data and non-visual data, perform target detection and behavior analysis on the visual data, perform standardization processing on the non-visual data, and fuse the processed visual data and non-visual data to generate a first perceptual event feature. A construction unit is used to construct a dynamic causal knowledge graph based on the features of the first perceived event; wherein, the knowledge graph includes nodes representing family entities, events and situational states; and edges representing the relationships between nodes; The matching unit is used to receive the second perceived event features and match the second perceived event features with the preset causal chain template in the knowledge graph; A generation unit is used to mark the second perceived event feature as a causal abnormal event and generate an associated node if the matching fails. The computing unit is used to generate multiple sets of causal explanation hypotheses based on the dynamic causal knowledge graph, calculate the posterior probability of each hypothesis, and select the hypothesis with the highest posterior probability as the causal explanation result. The early warning unit is used to calculate uncertainty measure and risk score based on the posterior probability of the causal explanation result, generate interpretable early warning information containing causal logic, and execute early warning for the user.

[0010] Optionally, the acquisition unit is further configured to: The object detection algorithm is used to identify people and objects in the image and output category labels and location coordinates; the pose estimation algorithm is used to extract the coordinates of key points of the human skeleton, and the pose changes in continuous frames are analyzed by combining a time series model to identify human behavior patterns and generate behavior labels. An extreme value normalization device is used to map environmental parameters and equipment operation data in non-visual data to a preset value range, and a timestamp accurate to milliseconds is added to each processed data, which is then encapsulated as environmental status information in key-value pair format. By aggregating personnel information, behavioral tags, and environmental status information within the same time window, structured perception event features with multidimensional information are generated.

[0011] Optionally, the computing unit is further configured to: The dynamic time warping algorithm is used to calculate the similarity between the event sequence corresponding to the features of newly perceived events and the preset causal chain template; If the similarity is lower than a preset threshold, it is determined to be a causal anomaly, a causal anomaly event node is generated, and an association edge is established between the node and the event node that triggered the anomaly. Based on historical event records, current situational state probability distribution, and entity attributes in the knowledge graph, multiple mutually exclusive causal explanation hypotheses are generated. The prior probability of each hypothesis is calculated using the frequency of historical events in the knowledge graph as the hypothesis, and the posterior probability of each hypothesis is calculated using the characteristics of the current perceived event and the probability of the situational state as evidence. The hypothesis with the highest posterior probability is selected as the final causal explanation, and the relationship between the hypothesis and the evidence is recorded and stored in the knowledge graph.

[0012] Optionally, the early warning unit is further configured to: Warning levels are classified according to risk scores and corresponding push notification methods are assigned: different warning levels correspond to different push notification methods. Obtain user feedback and update the knowledge graph based on the user feedback.

[0013] Optionally, the device further includes: The data acquisition unit is also used to acquire non-visual data for the current time period as an environmental baseline value when no event is detected within a consecutive preset number of time windows, including temperature and humidity, gas concentration, and equipment standby power parameters. The computing unit is also used to calculate the drift coefficient between real-time environmental parameters and baseline values; The early warning unit is also used to determine that the environmental parameters are abnormal if the drift coefficient exceeds a preset threshold, generate an environmental parameter drift event node, update the attributes of the corresponding environmental state node in the knowledge graph, and trigger a low-level early warning to prompt the user to check the environment or equipment.

[0014] According to a third aspect of this disclosure, an electronic device is provided, comprising: At least one processor; and A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method described in the first aspect above.

[0015] According to a fourth aspect of this disclosure, a non-transitory computer-readable storage medium is provided storing computer instructions, wherein the computer instructions are configured to cause the computer to perform the method described in the first aspect above.

[0016] According to a fifth aspect of this disclosure, a computer program product is provided, comprising a computer program that, when executed by a processor, implements the method described in the first aspect above.

[0017] The home security prediction method, device, electronic equipment, and storage medium disclosed herein, through this application, can integrate visual and non-visual data to generate perceived event features, construct a dynamic causal knowledge graph containing home entities, events, and situational states, and achieve deep modeling of the overall state of the home and the causal relationship between events. Furthermore, it can capture anomalies in complex scenarios through causal chain template matching and anomaly labeling, generate multiple sets of causal hypotheses and select the optimal results, and combine risk scoring to generate interpretable warnings with causal logic. This avoids the limitations of traditional static rules or single sensors. Therefore, it can solve the technical problem that existing traditional solutions lack the ability to deeply model the overall state of the home, member behavior patterns, and causal relationships between events, making it difficult to adapt to complex dynamic scenarios. This achieves the technical effects of improving the adaptability of home security systems to complex dynamic scenarios, increasing the accuracy and interpretability of warnings, and enhancing the system's intelligence level.

[0018] 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

[0019] The accompanying drawings are provided to better understand this solution and do not constitute a limitation of this disclosure. Wherein: Figure 1 This is a flowchart illustrating a home security prediction method provided in an embodiment of the present disclosure. Figure 2 A system architecture diagram of a home security system provided in this application embodiment; Figure 3 A schematic diagram of data flow for a home security prediction method provided in an embodiment of this application; Figure 4 A flowchart illustrating another predictive method for home security provided in an embodiment of this application; Figure 5 A flowchart illustrating another predictive method for home security provided in an embodiment of this application; Figure 6 A flowchart illustrating another predictive method for home security provided in an embodiment of this application; Figure 7 A schematic diagram of the structure of a predictive device for home security provided in an embodiment of this disclosure; Figure 8 A schematic block diagram of an example electronic device provided for embodiments of this disclosure. Detailed Implementation

[0020] The exemplary embodiments of this disclosure are described below with reference to the accompanying drawings, including various details of the embodiments to aid understanding, and should be considered merely exemplary. Therefore, those skilled in the art will recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of this disclosure. Similarly, for clarity and brevity, descriptions of well-known functions and structures are omitted in the following description.

[0021] The following description, with reference to the accompanying drawings, outlines a method, apparatus, electronic device, and storage medium for predicting home security according to embodiments of the present disclosure.

[0022] Figure 1 This is a flowchart illustrating a home security prediction method provided in an embodiment of the present disclosure.

[0023] like Figure 1 As shown, the method includes the following steps: Step 101: Collect visual data and non-visual data, perform target detection and behavior analysis on the visual data, standardize the non-visual data, and fuse the processed visual data and non-visual data to generate the first perceptual event feature. It collects visual and non-visual data from the home environment. The visual data comes from one or more cameras deployed in different areas of the home, which can capture dynamic information such as the activities of people and the status of objects within the home. The non-visual data comes from various non-visual sensors within the home, including door and window sensors, smart sockets, temperature and humidity sensors, and passive infrared sensors. Door and window sensors can obtain the opening and closing status of doors and windows, smart sockets can collect real-time power consumption values, temperature and humidity sensors are used to obtain environmental temperature and humidity parameters, and passive infrared sensors can detect motion signals in the area.

[0024] The acquired visual data is first processed for target detection using deep learning-based target detection algorithms, such as the YOLO series of single-stage detectors, SSD, or the Faster R-CNN series of two-stage detectors. This process identifies people and objects in each frame of visual data and outputs information including the category label of the detected object, its position coordinates in the image, and temporary identification information for continuously tracking the same target. Next, the visual data after target detection is further analyzed for behavior. The key point coordinates of people are extracted using pose estimation algorithms such as OpenPose and HRNet. Then, the key point coordinate sequence of the same person in consecutive time frames is input into a temporal model such as a Long Short-Term Memory Network (LSTM) or a Gated Recurrent Unit (GRU) to learn the pattern of pose sequence changes in order to identify the behavior patterns of people and obtain the association information between the corresponding person identification and behavior pattern label.

[0025] The collected non-visual data is standardized by adding a precise collection timestamp to each piece of non-visual data. Analog data such as power consumption, temperature and humidity of smart sockets are normalized to a preset value range. At the same time, all non-visual data is uniformly packaged into a standard format such as key-value pair format to form structured environmental status information.

[0026] Finally, based on the timestamp, the target recognition information and behavior analysis results in the processed visual data are aligned and integrated with the environmental state information corresponding to the standardized non-visual data to form a comprehensive data record. This data record is the first perception event feature, which can completely describe the atomic events that occur in the home environment at a specific moment, providing basic data support for subsequent system processing.

[0027] Step 102: Construct a dynamic causal knowledge graph based on the features of the first perceived event; wherein the knowledge graph includes nodes representing family entities, events, and situational states; and edges representing the relationships between nodes; Based on the characteristics of the first perceived event, the construction process of a dynamic causal knowledge graph is initiated. This dynamic causal knowledge graph is a knowledge model that can dynamically record and update information related to the family environment and reflect the associations and causal relationships between nodes. Its core components include nodes that represent family entities, events, and situational states, as well as edges that represent the relationships between nodes.

[0028] Nodes representing family entities are used to record specific physical objects or people in the family environment, such as family members, doors and windows, smart appliances, and sensors. Each entity node is accompanied by corresponding attribute information, such as the identity of the person, the type of device, and its installation location. Nodes representing events are used to record various events that occur dynamically in the family environment. This event information is derived from the characteristics of first-sensory events, such as people opening doors, devices turning on, and abnormal environmental parameters. Each event node is marked with a precise timestamp and key attributes related to the event, such as the duration of the event and the location of the event. Nodes representing contextual states are used to record the macro-background state of the family at a specific time scale, such as the family's daily routine and security mode status. These nodes can reflect the overall environmental atmosphere and operating mode of the family, and their status will change dynamically with the actual situation of the family.

[0029] Edges representing relationships between nodes are used to connect different types of nodes to reflect semantic associations or causal connections. For example, a person entity node and a door opening event node can be associated through an edge to indicate that the person is the executor of the door opening event; a door opening event node and a door entity node can be associated through an edge to indicate that the event occurred at the location corresponding to the door; two event nodes can also be associated through edges to reflect causal relationships between events, such as an edge between an abnormal equipment power event node and an equipment failure event node, which can represent that the former is one of the causes of the latter. During the construction process, the system parses various types of information in the features of the first perceived event, maps them to corresponding nodes and edges, and gradually builds a complete dynamic causal knowledge graph, providing knowledge support for subsequent system reasoning and analysis.

[0030] Step 103: Receive the second perceived event features and match the second perceived event features with the preset causal chain template in the knowledge graph; The system receives a second perceptual event feature. The generation method of the second perceptual event feature is consistent with the generation logic of the first perceptual event feature. Both are structured data records formed by collecting visual and non-visual data in the home environment, performing target detection, behavior analysis, and non-visual data standardization processing, and then merging them according to timestamps. This data record can fully reflect the atomic events that occur in the home environment at a specific moment, including key information such as the people or objects involved in the event, corresponding behavior patterns, environmental state parameters, and precise timestamps.

[0031] The system retrieves preset causal chain templates from a dynamic causal knowledge graph. These causal chain templates are typical event sequences summarized from the regularities of daily family life. They cover the event association logic corresponding to common normal behavior processes in family scenarios. For example, event sequences that conform to the routine homecoming process of family members, such as "garage door opening event - people entering the garage event - door opening event", or event sequences that conform to the daily cooking preparation process, such as "people walking to the kitchen event - kitchen light turning on event - smart socket power increasing event". Each causal chain template clearly defines the order and relationship of events in a normal scenario.

[0032] The system compares and matches the event sequence corresponding to the received second perceived event features with the retrieved preset causal chain template one by one. By analyzing the occurrence order of events, the entity associations involved, and the environmental state associations in the second perceived event features, it determines whether the event sequence conforms to the conventional causal logic defined by a preset causal chain template, providing a basis for further identification of whether there are anomalies in the event.

[0033] Step 104: If the matching fails, mark the second perceived event feature as a causal abnormal event and generate an associated node; Based on the matching results, if the event sequence corresponding to the second perceived event feature cannot be matched with any of the preset causal chain templates in the dynamic causal knowledge graph, it indicates that the event sequence violates the normal causal logic under the rules of daily family life. For example, the "door opening event" occurs first, but no preceding related events such as "garage door opening event" or "person entering garage event" are detected, or the "window opening event" occurs without reasonable preceding contextual events such as "person approaching window event" or "abnormal environmental parameters (such as air quality decline) event". In this case, the system determines that the matching result is a failure and marks the second perceived event feature as a causal abnormal event.

[0034] The system initiates a process to generate associated nodes, creating a new "causal anomaly event" node in the dynamic causal knowledge graph. This node must contain key information related to the causal anomaly, including the timestamp of the event, involved household entities (such as open doors or windows, sensors in relevant areas), and core content of the second-sensory event characteristics (such as environmental parameters at the time of the event, and whether any individuals are involved). Simultaneously, the system establishes an association edge between this "causal anomaly event" node and the event node corresponding to the second-sensory event characteristics that triggered the anomaly, clearly marking their "triggered and triggered" relationship. This allows the dynamic causal knowledge graph to clearly present the source and specific associated objects of the causal anomaly, providing clear anomaly identification and associated information support for subsequent causal reasoning and risk assessment.

[0035] Step 105: Based on the dynamic causal knowledge graph, generate multiple sets of causal explanation hypotheses, calculate the posterior probability of each hypothesis, and select the hypothesis with the highest posterior probability as the causal explanation result. The system operates based on a pre-constructed dynamic causal knowledge graph. This knowledge graph contains family entity nodes, event nodes, situational state nodes, and edges connecting these nodes. It stores information on family members, equipment, historical events, and the probability distribution of the family's macro-state, providing knowledge support for generating causal explanation hypotheses. The system generates multiple sets of causal explanation hypotheses from a pre-set hypothesis library, combined with node association information within the knowledge graph. These hypotheses must cover different logical inferences about the possible causes of the current event. For example, for the scenario of "abnormal window opening," multiple mutually exclusive hypotheses can be generated, such as "family members opening windows for ventilation," "illegal intrusion by outsiders," and "windows opening on their own due to environmental factors or malfunctions." Each set of hypotheses is associated with corresponding entity or event information in the knowledge graph. For instance, "family members opening windows for ventilation" would be associated with the "family member" entity node and the "daily ventilation" historical event node.

[0036] The system employs Bayesian inference to calculate the posterior probability of each hypothesis. During the calculation, it integrates multi-source evidence, including direct information from the second-perception event features (such as whether the event was accompanied by abnormal signals) and background information from the knowledge graph (such as the existence of a "family member going out" planned event node, and the probability of "nighttime sleep" displayed in the current family situation status node). The evidence is combined with the prior probabilities of the hypotheses using Bayes' theorem to derive the posterior probability of each hypothesis. Finally, the system compares the posterior probabilities of all hypotheses, selects the hypothesis with the highest probability, and identifies it as the causal explanation for the current event, providing a decision-making basis for subsequent early warning generation.

[0037] Step 106: Calculate the uncertainty measure and risk score based on the posterior probability of the causal explanation result, generate interpretable early warning information containing causal logic, and issue an early warning to the user.

[0038] The system operates based on the causal explanation and its corresponding posterior probability. First, it calculates an uncertainty metric, which reflects the likelihood that the causal explanation is invalid. This metric is calculated by subtracting the posterior probability of the causal explanation from 1. For example, if the posterior probability of the causal explanation is 92%, the uncertainty metric is 8%, intuitively reflecting the reliability of the system's causal explanation.

[0039] The system calculates a risk score, which requires consideration of two key parameters: first, a pre-defined inherent severity level coefficient associated with the type of causal explanation result. This coefficient is set according to the potential risk level of the event; for example, the coefficient for "illegal intrusion by external personnel" is higher than that for "minor fluctuations in equipment parameters"; second, the posterior probability (i.e., confidence level) of the causal explanation result. The risk score is obtained by multiplying the inherent severity level coefficient by the posterior probability. For example, when the inherent severity level coefficient is 0.9 and the posterior probability is 92%, the risk score is 0.828.

[0040] The system generates interpretable warning information containing causal logic. The information not only covers the event itself (such as "the backyard window was opened at 3 a.m."), but also clearly includes the causal explanation (such as "inferred to be illegal intrusion by outsiders"), the corresponding posterior probability and uncertainty measure, fully presenting the system's reasoning logic and helping users quickly understand the source of risk and its reliability.

[0041] The system issues warnings to users based on risk scores, classifying them into different levels according to preset thresholds: high-risk scores (e.g., scores ≥ 0.8) trigger the highest priority warning, including home sound alarms and calls to the user's emergency contacts; medium-risk scores (e.g., 0.5 ≤ scores < 0.8) trigger push notifications to mobile applications with prominent alert sounds; low-risk scores (e.g., scores < 0.5) only generate silent log records within the application, ensuring that warnings effectively convey risk information while avoiding excessive disturbance to users.

[0042] In some embodiments, the steps of collecting visual data and non-visual data, performing target detection and behavior analysis on the visual data, standardizing the non-visual data, and fusing the processed visual data and non-visual data to generate a first perceptual event feature include: The object detection algorithm is used to identify people and objects in the image and output category labels and location coordinates; the pose estimation algorithm is used to extract the coordinates of key points of the human skeleton, and the pose changes in continuous frames are analyzed by combining a time series model to identify human behavior patterns and generate behavior labels. The extreme value normalization method is used to map environmental parameters and device operation data in non-visual data to a preset value range, and a timestamp accurate to milliseconds is added to each processed data, which is then encapsulated as environmental status information in key-value pair format. By aggregating personnel information, behavioral tags, and environmental status information within the same time window, structured perception event features with multidimensional information are generated.

[0043] In the process of collecting visual and non-visual data and generating the first perceptual event features, the visual data is processed first. Deep learning-based object detection algorithms, such as the YOLO series and Faster R-CNN series, are used to accurately identify people and objects in each frame of images captured by home cameras. The output includes the category label of the detected object and its position coordinates in the image, providing clear target data for subsequent behavior analysis. Next, pose estimation algorithms, such as OpenPose and HRNet, are used to extract the skeletal keypoint coordinates of the identified people. The keypoint coordinate sequence of the same person within consecutive time frames is input into a temporal model, such as a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU). By learning the dynamic changes in the pose sequence, the model accurately identifies the behavior patterns of people and generates corresponding behavior labels, establishing a direct correlation between people and behavior.

[0044] For non-visual data, an extreme value normalization method is used to map environmental parameters such as the on / off status of door and window sensors, the real-time power consumption of smart sockets, the ambient temperature and humidity parameters of temperature and humidity sensors, and the motion detection signals of passive infrared sensors, along with equipment operating data, to a preset numerical range. This eliminates analytical biases caused by differences in the units of measurement between different types of data. Simultaneously, a timestamp accurate to the millisecond level is added to each normalized non-visual data point to ensure the data's temporal accuracy. Finally, all processed non-visual data is encapsulated into key-value pair format environmental status information, forming a structured non-visual data output.

[0045] Using time as the core benchmark, the same time window is selected, and the personnel information within the window, including category labels and location coordinates, behavior labels, and environmental status information, is aggregated and integrated to form an organic relationship between multi-dimensional data. Finally, a structured perception event feature containing multi-dimensional information of personnel, behavior, and environment is generated, providing standardized and high-quality basic data support for the subsequent construction of dynamic causal knowledge graphs.

[0046] In some embodiments, generating multiple sets of causal explanation hypotheses based on the dynamic causal knowledge graph, calculating the posterior probability of each hypothesis, and selecting the hypothesis with the highest posterior probability as the causal explanation result includes: The dynamic time warping algorithm is used to calculate the similarity between the event sequence corresponding to the features of newly perceived events and the preset causal chain template; If the similarity is lower than a preset threshold, it is determined to be a causal anomaly, a causal anomaly event node is generated, and an association edge is established between the node and the event node that triggered the anomaly. Based on historical event records, current situational state probability distribution, and entity attributes in the knowledge graph, multiple mutually exclusive causal explanation hypotheses are generated. The prior probability of each hypothesis is calculated using the frequency of historical events in the knowledge graph as the hypothesis, and the posterior probability of each hypothesis is calculated using the characteristics of the current perceived event and the probability of the situational state as evidence. The hypothesis with the highest posterior probability is selected as the final causal explanation, and the relationship between the hypothesis and the evidence is recorded and stored in the knowledge graph.

[0047] In the process of generating causal explanation results based on dynamic causal knowledge graph, the dynamic time warping algorithm is first used to calculate the similarity between the event sequence corresponding to the features of the newly perceived event and the preset causal chain template. This algorithm can effectively handle the length difference and time offset that may exist in the event sequence in the time dimension, accurately quantify the degree of matching between the two in the order of event occurrence and the logical association, and adapt to the dynamic changes in the rhythm of event occurrence in the family scene.

[0048] If the calculated similarity is lower than a preset threshold, the newly perceived event sequence is determined to violate conventional causal logic and belong to causal anomalies. At this time, a causal anomaly event node is generated in the dynamic causal knowledge graph. This node contains attributes such as the timestamp of the anomaly occurrence and the entity information involved. An association edge is established between this causal anomaly event node and the event node that triggered the anomaly, clearly marking the triggering relationship between the two and improving the anomaly event record of the knowledge graph.

[0049] Subsequently, relying on the data resources in the dynamic causal knowledge graph, and combining the occurrence context of similar events in historical event records, the probability distribution of the current situation state (such as the probability value of family routine status), and entity attributes (such as personnel identity and equipment operating parameters), multiple sets of logically mutually exclusive causal explanation hypotheses are generated to ensure that the hypotheses can cover different possible causes of the event.

[0050] The prior probability of each hypothesis is determined based on the frequency of occurrence of historical events in the knowledge graph. Then, the specific information in the characteristics of the current perceived event and the probability of the current situation state are used as evidence. The posterior probability of each hypothesis is calculated through Bayesian inference, thereby achieving a quantitative assessment of the credibility of the hypothesis.

[0051] The hypothesis with the highest posterior probability is selected as the final causal explanation. At the same time, the relationship between the hypothesis and the evidence used is recorded and stored in a dynamic causal knowledge graph to provide data support for subsequent system reasoning and knowledge updates.

[0052] In some embodiments, the step of calculating an uncertainty measure and a risk score based on the posterior probability of the causal explanation result, generating interpretable early warning information containing causal logic, and executing an early warning for the user includes: Warning levels are classified according to risk scores and corresponding push notification methods are assigned: different warning levels correspond to different push notification methods. Obtain user feedback and update the knowledge graph based on the user feedback.

[0053] In the process of generating alerts and processing user feedback based on causal explanations, the uncertainty measure and risk score are first calculated based on the posterior probability of the causal explanation. The uncertainty measure is obtained by subtracting the posterior probability from 1, intuitively reflecting the possibility that the causal explanation is invalid, and presenting the user with the reliability of the system's judgment. The risk score is obtained by multiplying the predefined inherent severity level coefficient corresponding to the type of causal explanation with the posterior probability (i.e., confidence level) of the causal explanation, thus quantifying the risk level of the event.

[0054] The system classifies warning levels based on calculated risk scores, with different notification methods for each level: A high-priority warning triggers a sound alarm combined with a call to the user's emergency contacts to ensure timely communication of emergency risks; a medium-priority warning triggers a notification with a prominent audible alert to balance risk transmission with user disruption; and a low-priority warning triggers a silent log entry within the app to avoid unnecessary disturbance. Simultaneously, the system generates interpretable warning information containing causal logic. This information includes not only the event itself but also the causal explanation, corresponding posterior probability, and uncertainty measure, clearly presenting the system's reasoning logic and helping users quickly understand the root cause of the risk.

[0055] By obtaining user feedback, users can input feedback information through preset interactive buttons within the application or natural language. The system parses the feedback content, extracts the user's core intent and key information, and directly operates on the dynamic causal knowledge graph based on the parsing results, such as correcting node attributes, supplementing related edges, or adding entity nodes. This transforms user feedback into updated data for the knowledge graph, aligning the system with the user's cognition and continuously improving the accuracy of subsequent judgments.

[0056] In some embodiments, the method further includes: When no event is detected within a preset number of consecutive time windows, non-visual data for the current time period is collected as the environmental baseline value, including temperature, humidity, gas concentration, and equipment standby power parameters. Calculate the drift coefficients of real-time environmental parameters relative to baseline values; If the drift coefficient exceeds the preset threshold, the environmental parameters are determined to be abnormal, an environmental parameter drift event node is generated, the attributes of the corresponding environmental state node in the knowledge graph are updated, and a low-level warning is triggered to prompt the user to check the environment or equipment.

[0057] First, the system monitors events occurring in the home environment in real time. When no events occur for a preset number of consecutive time windows, indicating a prolonged period of inactivity or equipment idling, the system automatically initiates the environmental baseline data collection process. At this point, the system focuses on collecting non-visual data for the current time period. This data covers core basic parameters of the home environment, including air temperature and relative humidity data from temperature and humidity sensors, gas concentration data from gas sensors, and standby power data of various appliances recorded by smart sockets or device-based monitoring modules. This data is integrated to determine the current environmental baseline value, which serves as a benchmark for judging whether the environment is normal in the future.

[0058] During subsequent operation, the system continuously collects real-time environmental parameters and calculates the drift coefficient between the real-time environmental parameters and the determined baseline value through a preset algorithm. This drift coefficient is used to quantify the degree of deviation of the real-time parameters from the baseline value. For example, it can be obtained by calculating the percentage of the difference between the real-time temperature and humidity and the baseline temperature and humidity relative to the baseline value, or by calculating the absolute deviation of the real-time gas concentration from the baseline gas concentration, etc., so as to intuitively reflect the change range of environmental parameters.

[0059] If the calculated drift coefficient exceeds a preset threshold, it indicates that the real-time environmental parameters have deviated from the normal static level, and the system determines that the environmental parameters are abnormal. At this time, the system generates an environmental parameter drift event node in the dynamic causal knowledge graph. This node contains attributes such as the timestamp of the abnormality, the specific parameter type of the deviation, and the drift coefficient value. At the same time, the system updates the attributes of the corresponding environmental state node in the knowledge graph, such as modifying the current parameter values ​​and abnormal markers of nodes such as "temperature and humidity status" and "gas concentration status". Finally, a low-level warning is triggered. This warning is pushed in the form of a silent notification in the mobile application, prompting the user to check the home environment or the operating status of related equipment in a timely manner, which ensures that environmental abnormalities are noticed, while avoiding excessive disturbance to the user due to non-urgent risks.

[0060] The following example illustrates the home security prediction method provided in the embodiments of this application.

[0061] In one specific embodiment, please refer to Figure 2 , Figure 2 A system architecture diagram of a home security system provided in this application embodiment is shown below. Figure 2 The system shown includes the following modules: data perception and feature extraction module, knowledge graph construction and maintenance module, causal reasoning and risk assessment module, and human-machine collaboration and knowledge alignment module.

[0062] I. Data Perception and Feature Extraction Module Please participate Figure 3 , Figure 3 This is a data flow diagram illustrating a home security prediction method provided in an embodiment of this application, as shown below. Figure 3 As shown, the function of this module is to serve as the data input layer of the system, collecting raw data from various sensors deployed in the home physical environment, and converting these heterogeneous and unstructured data streams into unified and structured sensing event features for subsequent modules to process.

[0063] In one embodiment, the module includes a target detection unit, a behavior analysis unit, and an environmental data processing unit.

[0064] Target Detection Unit: This unit is dedicated to processing visual data streams from one or more cameras. Internally, it can employ deep learning-based target detection algorithm models, such as single-stage detectors (e.g., YOLO series, SSD) or two-stage detectors (e.g., Faster R-CNN series). This unit receives video frames as input, processes each frame, and outputs category labels for all detected people and objects, their position coordinates in the image (usually bounding box coordinates), and a temporary identifier (ID) for tracking the same target in consecutive video frames. The output data is a structured list of target information.

[0065] Behavior Analysis Unit: This unit receives person target information (especially their position coordinates and tracking ID) from the target detection unit. Internally, it can integrate a pose estimation algorithm (such as OpenPose or HRNet) to extract keypoint coordinates of the identified human target from video images. Subsequently, this unit inputs the keypoint coordinate sequence of the same person (determined by the tracking ID) within consecutive time frames into a temporal model. This temporal model can be a Long Short-Term Memory (LSTM) network or a Gated Recurrent Unit (GRU), which learns the changing patterns of the pose sequence to identify specific behavioral patterns. The unit ultimately outputs the identified person ID and its corresponding behavioral pattern label.

[0066] Environmental Data Processing Unit: This unit processes all data streams collected by non-visual sensors. These data may include: binary switch status (0 or 1) returned by door and window sensors, real-time power consumption values ​​returned by smart sockets, environmental parameters returned by temperature and humidity sensors, and motion detection signals returned by passive infrared (PIR) sensors. The function of this unit is to standardize these raw data, including adding a precise timestamp to each data point, normalizing analog data to a preset range, and uniformly encapsulating all data into a standard format (e.g., key-value pair format) to form environmental status information.

[0067] The final output of the module, namely the structured perceived event feature, is the result of combining and aligning the outputs of the three units mentioned above. Specifically, the system uses timestamps as a reference to aggregate the people and objects identified by the target detection unit at the same moment, the behavioral patterns of the corresponding people analyzed by the behavior analysis unit, and the environmental state information obtained by the environmental data processing unit into a single, comprehensive data record. This record is a structured perceived event feature, which completely describes an atomic event that occurs within the home environment at a specific instant.

[0068] II. Knowledge Graph Construction and Maintenance Module This module is the core knowledge base of the system, responsible for building and continuously updating a dynamic causal knowledge graph describing the family environment. This knowledge graph is stored in the form of a graph database, and the nodes and edges together constitute a digital twin of the family environment. In one embodiment, this module includes the following five units: Knowledge Graph Storage Unit: Responsible for the physical storage and access of the knowledge graph. The graph structure is defined as follows: Entity Nodes: Represent physical objects in the family; Event Nodes: Represent specific events that occur, each event node has a timestamp attribute; Context State Nodes: Represent the macroscopic state of the family during a specific time period; Edges: Represent the relationships between nodes.

[0069] Knowledge Graph Update Unit: This unit is responsible for writing information into the knowledge graph. It receives structured event features from the data perception and feature extraction module and parses them into nodes and edges in the graph. For example, if a feature indicates "Person A opened door B at time T," this unit will create a new "Door Opening Event" node in the graph and establish the relationship between this event node and the existing "Person: A" entity node and "Device: Door B" entity node. Simultaneously, this unit is also responsible for pre-creating future plans input by the user through the application (such as "Visitor C will visit on Friday afternoon") as event nodes and adding them to the graph.

[0070] Please see Figure 4 , Figure 4 This is a flowchart illustrating another predictive method for home security provided in an embodiment of this application, as shown below. Figure 4 As shown, the time-probability state prediction unit: The function of this unit is to assign dynamic probabilities to situational state nodes. It can be configured with a Hidden Markov Model (HMM). This model learns from historical event sequence data to automatically identify several typical macro states of the family (as hidden states of the HMM) and learns the transition probability matrix between these states and the emission probability matrix of various events occurring in specific states. When the system is running, this unit calculates the probability distribution of the current macro state based on the latest event sequence using the forward algorithm of the HMM and updates the probability attributes of situational state nodes such as "family state".

[0071] Causal Chain Integrity Monitoring Unit: This unit is responsible for detecting anomalous events. It pre-stores a series of standard causal chain templates, which describe typical event sequences in daily life. This unit continuously monitors event sequences added to the knowledge graph in real time and matches them against all templates. If a real-time event sequence shows a break from a standard template, the unit determines this as a causal anomaly, generates a "causal anomaly event" node, associates it with the event that caused the anomaly, and injects it into the knowledge graph.

[0072] User Intent Hypothesis Generation Unit: This unit is used to generate predictive event nodes. It infers the user's potential intent by analyzing the combination of multi-dimensional data. For example, when the system detects multiple sub-events such as "person walks towards air conditioner", "ambient temperature is higher than set temperature", and "remote control is picked up", this unit can generate a high-confidence "user intent hypothesis" event node with the content "intent: adjust air conditioner temperature" and add it to the knowledge graph to provide richer context for subsequent reasoning.

[0073] III. Causal Reasoning and Risk Assessment Module Please see Figure 5 , Figure 5 This is a flowchart illustrating another predictive method for home security provided in an embodiment of this application, as shown below. Figure 5 As shown, this module is the decision-making core of the system. It receives real-time perceived events and uses the rich information in the knowledge graph to perform in-depth analysis of the nature of the events and risk level assessment. In one embodiment, the module includes a hypothesis generation unit and a Bayesian inference unit.

[0074] Hypothesis generation unit: When a new perceived event feature is received (e.g., "the backyard window was opened at 3 a.m."), this unit will match and generate a set of mutually exclusive causal explanation hypotheses from a pre-set hypothesis library. For the above event, the possible hypothesis set generated is: {H1: "family members opened the window for ventilation", H2: "illegal intrusion by outsiders", H3: "the window opened by itself due to strong wind or malfunction"}.

[0075] Bayesian Inference Unit: This unit quantifies and evaluates each generated hypothesis; its core is the Bayesian formula. ,in Represents a certain assumption, Representing all available evidence, this unit calculates the posterior probability of each hypothesis, and the combined evidence E includes: Direct evidence derived from the characteristics of the perceived event: such as whether the event itself was accompanied by the sound of breaking glass; Information from the family dynamic causal knowledge graph includes: user-planned events: whether the event node "the homeowner is on vacation" exists in the knowledge graph; entity relationships and attributes: whether the person opening the window is a known family member; probability distribution of situational state nodes: the current macro-state of the family calculated by the time-probability state prediction unit, which uses this evidence as conditions to calculate... And combined with the prior probability of the hypothesis. Finally, the posterior probability of each hypothesis is obtained, and the hypothesis with the highest probability will be determined as the system's final inference result for the event, i.e., the causal explanation hypothesis.

[0076] IV. Human-Computer Collaboration and Knowledge Alignment Module This module serves as the interface between the system and the user, responsible for conveying the system's analysis results to the user in an understandable way and using user feedback to optimize the system itself.

[0077] In one embodiment, the module includes an interpretable warning generation unit and a feedback processing and knowledge update unit.

[0078] Explainable Early Warning Generation Unit: This unit receives the causal explanation hypothesis and its calculated uncertainty determined by the causal reasoning and risk assessment module. It is responsible for generating early warning notifications. The notification content includes not only the event itself, but also the inferred cause and its confidence level. In addition, this unit can also classify and push notifications according to the risk level and confidence level of the hypothesis. High-risk, high-confidence events trigger the highest priority alarms (such as sound alarms and emergency calls), while low-risk or low-confidence events may only trigger a normal mobile application message.

[0079] Feedback Processing and Knowledge Update Unit: This unit is responsible for processing user responses. It can parse user feedback information through preset interactive buttons or basic natural language processing technology. Based on the parsing results, this unit directly manipulates the knowledge graph. In this way, every valid user feedback is directly used to correct and enrich the knowledge graph, forming a closed-loop learning and alignment process that continuously improves the accuracy of the system's future judgments.

[0080] In one specific embodiment, the data perception and feature extraction module serves as the initial information processing layer of the predictive home security system of the present invention. Its core function is to systematically process and convert the raw and unstructured multidimensional data streams originating from heterogeneous sensor networks in the home environment into highly structured perception event features that can be directly utilized by the subsequent knowledge graph construction and maintenance module.

[0081] To achieve the above functions, the internal technical solution of the data perception and feature extraction module can be specifically set up to include a target detection unit, a behavior analysis unit, and an environmental data processing unit. These three units process data of different dimensions in parallel or serially, and finally fuse their outputs.

[0082] Specifically, the target detection unit is dedicated to processing visual data streams from one or more cameras in a home surveillance network. In a preferred implementation, this unit can integrate one or more pre-trained deep learning target detection models. To balance detection speed and accuracy, or in scenarios with higher accuracy requirements, a Faster or R-CNN two-stage detector can be used. This unit receives continuous video frame images as input and identifies and locates people and key objects in each frame through forward propagation calculations of the model. Its output is not a simple recognition result, but a set of structured target information lists. Each piece of information includes the target's category label, its bounding box position coordinates in the current frame image, and a temporary identity identifier (ID) assigned by an inter-frame tracking algorithm. This ID is used to ensure that the same physical target has a unique identity in a continuous time series.

[0083] Based on the identification and tracking of personnel completed by the target detection unit, the behavior analysis unit further performs in-depth analysis of the dynamic behavior of personnel. The input of this unit comes directly from the personnel target information output by the target detection unit, especially its unique tracking ID and continuous bounding box coordinates. Its internal processing flow can be divided into two stages: the first is the pose estimation stage, which can use pose estimation algorithms such as OpenPose and HRNet to accurately extract the two-dimensional or three-dimensional coordinates of human skeleton key points from the detected personnel image region.

[0084] The process then proceeds to the temporal analysis phase. This phase processes the sequence of pose keypoint coordinates for individuals belonging to the same tracking ID across consecutive time frames as a time series. To identify semantically meaningful behavioral patterns from this sequence, this unit can deploy a temporal analysis model, such as a Long Short-Term Memory network or a gated recurrent unit. This model, by learning from a large amount of labeled behavioral data, can capture the dynamic changes in the pose sequence. The processing can be formally represented as follows: ; in, Identification mark representing a specific person This is the sequence of attitude key point coordinates. This is a time series analysis model, and its output is... That is to be in Behavioral pattern tags identified in real time.

[0085] Meanwhile, to achieve comprehensive environmental perception, the environmental data processing unit is responsible for processing all data collected by non-visual sensors. These data sources may include, but are not limited to: magnetic contact sensors installed on doors and windows to return binary on / off states; power meters integrated into smart home appliances or sockets to return real-time power consumption values; and passive infrared (PIR) sensors and temperature and humidity sensors deployed in the space to return motion detection signals and environmental parameters, respectively. The core task of this unit is to perform unified standardization processing on these diverse and formatted raw data. The processing flow includes accurately attaching a collection timestamp to each data point, cleaning outliers, normalizing analog data, and finally encapsulating all data into a standardized key-value pair format to form environmental status information that can be directly accessed by the system.

[0086] One of the core technological effects of the data perception and feature extraction module lies in its final feature fusion step. This step is not a simple listing of the outputs of the above units, but rather a strict synchronization and alignment based on timestamps. The system aggregates the set of people and objects identified by the target detection unit at the same time or within a very small time window, the behavior patterns of the corresponding people parsed by the behavior analysis unit, and the environmental state information integrated by the environmental data processing unit into a single, comprehensive data record.

[0087] This data record is the structured perceptual event feature defined in this invention. It is a composite data structure that fully describes an atomic event that occurs within the home environment at a specific moment. It contains all the information about "who / what", "where", "what did", and "the state of the environment". This organic fusion of multimodal information provides a much richer and more reliable input for subsequent causal inference than a single information source, significantly reducing the possibility of misjudgment due to incomplete information, thus laying a solid data foundation for the prediction accuracy of the entire system.

[0088] In a specific embodiment, the knowledge graph construction and maintenance module is the core knowledge hub of the predictive home security system of the present invention. Its fundamental function is to receive discrete structured perceived event features generated by the data perception and feature extraction module, and transform them into a persistent knowledge model rich in causal relationships that can characterize the dynamic evolution of the home environment, namely, the home dynamic causal knowledge graph. The operation of this module enables the system to have learning, memory and adaptive capabilities, providing in-depth contextual basis for subsequent causal reasoning and risk assessment.

[0089] To achieve the above functions, the internal technical solution of the knowledge graph construction and maintenance module can be specifically set as including a knowledge graph storage unit, a knowledge graph update unit, a time-probability state prediction unit, a causal chain integrity monitoring unit, and a user intent hypothesis generation unit.

[0090] First, the knowledge graph storage unit is responsible for the physical storage, data structure definition, and persistent management of the family's dynamic causal knowledge graph. In a preferred implementation, professional graph database technology can be used as its underlying implementation. The data structure of this knowledge graph is rigorously defined to include the following basic elements: Entity nodes: used to represent specific physical objects or people in the home environment, and include specific attributes such as person's name, device type, and location; Event nodes: Used to record dynamically occurring events. Each event node has a precise timestamp and other relevant attributes. For example, a "door opening event" node may include attributes such as the time of occurrence and the duration. Contextual state nodes: These are used to characterize the macro-background state of a family at a specific time scale, such as "family routine status" or "security mode status". Their state values ​​are dynamically changing. Edges: used to represent rich semantic and causal relationships between nodes. For example, (entity node: person) and (entity node: device) can be connected by an edge with a "ownership" relationship; an (event node) can be pointed to a (entity node: location) by an edge with a "occurred" relationship; and two (event nodes) can be characterized by an edge with a "caused" relationship.

[0091] Based on this structure, the knowledge graph update unit acts as the execution mechanism for writing data into the knowledge graph. This unit receives structured awareness event features from upstream and performs parsing and transformation operations, mapping them into new elements in the knowledge graph. For example, when receiving a feature "{Time: T, Person ID: A, Behavior: Opening the door, Object ID: B, Location: Door}", the unit creates a new "Opening the door event" node in the knowledge graph and establishes an association edge between this event node and the existing "Person: A" entity node and "Equipment: Door B" entity node. In addition, this unit can also process planned events input by users through the human-computer interaction interface and pre-create them as event nodes, thereby incorporating future information into the management scope of the knowledge graph.

[0092] To enable the knowledge graph to capture the macro-rhythms of the environment, a time-probability state prediction unit is configured to dynamically infer and quantify the current value of situational state nodes. In this embodiment, this unit is constructed based on a Hidden Markov Model (HMM), which treats the macro-state of the household as a hidden state that cannot be directly observed, and uses the event sequence output by the data perception and feature extraction module. This is considered an observable sequence. This unit determines the key parameter of the HMM—the state transition probability matrix—through offline learning from historical data. That is, the probability of transitioning from one macroscopic state to another; emission probability matrix. That is, the probability of a specific event occurring under a certain macroscopic state; and the initial state probability vector. .

[0093] During system operation, this unit uses a forward algorithm to calculate the current moment in real time based on the observed event sequence. The system is in every hidden state Posterior probability: ; in, For the system at time state, For the first A specific possible state, For at any time The observed values, For probability, This is the observation sequence.

[0094] The calculated probability distribution results will be used to update the attribute values ​​of the "family routine status" context state node in the knowledge graph. This allows the knowledge graph to not only record micro-events, but also to overlay a probabilistic description of the macro-background status, providing key contextual information for subsequent reasoning.

[0095] Furthermore, to proactively detect potential anomalies, a causal chain integrity monitoring unit is introduced. This unit predefines and stores a series of conventional causal chain templates. These templates are standard event sequences summarized from daily life patterns. For example, a typical homecoming sequence template can be defined as: "garage door opens" event - "person enters garage" event - "door leading to the interior opens" event. The unit continuously compares the event sequences added in real time in the knowledge graph with these preset templates. When a real-time event sequence is detected that violates the conventional causal chain, such as the event "person moves indoors" occurring directly without the preceding event, the unit determines this as a causal break. At this time, it generates a "causal anomaly event" node and establishes a connection between it and the event node that triggered this anomaly, thereby highlighting this potential abnormal behavior at the knowledge graph level.

[0096] Furthermore, to enhance the system's predictive capabilities, the user intent hypothesis generation unit is configured to generate forward-looking hypothesis nodes based on the fusion analysis of multidimensional data. This unit does not process a single event but analyzes a set of concurrent or closely related event characteristics within a specific time window. For example, when the system simultaneously observes a set of characteristics such as "person A walks towards the window," "environmental data shows a decline in air quality," and "window sensor status remains unchanged," this unit can generate a "user intent hypothesis" event node, whose content can be labeled as "intent: open the window for ventilation," and inject it into the knowledge graph. This prediction of the user's potential intent provides the system with deeper semantic information for understanding the current situation, enabling subsequent risk assessments to be based on more comprehensive information.

[0097] In summary, the knowledge graph construction and maintenance module, through the collaborative work of its internal units, gradually transforms the raw data stream into a dynamically evolving, clearly structured family dynamic causal knowledge graph that includes macro-states and can proactively identify anomalies and predictive intentions, providing a solid knowledge foundation for achieving accurate predictive security.

[0098] In a specific embodiment, the causal reasoning and risk assessment module serves as the core of the decision analysis for the predictive home security system of the present invention. Its function is to receive structured perceived event features representing the current state generated by the data perception and feature extraction module, and combine them with the family dynamic causal knowledge graph-spectrum information representing the history and macro-context stored in the knowledge graph construction and maintenance module to make in-depth inferences about the root cause of the perceived event, and finally output a quantitative causal explanation hypothesis with probabilistic significance.

[0099] To achieve the above functions, the internal technical solution of the causal reasoning and risk assessment module can be specifically set to include a hypothesis generation unit and a Bayesian reasoning unit.

[0100] Specifically, the hypothesis generation unit functions to provide an initial, logically consistent set of explanations for each input perceptual event feature. When the unit receives a new perceptual event feature, it matches and generates a set of logically mutually exclusive causal explanation hypotheses from a pre-defined hypothesis library associated with the event type. For the aforementioned example event, the hypothesis set output by this unit could be: { 1: "Family members open windows for ventilation". 2: "Unauthorized intrusion by outsiders" 3: "The window opened on its own due to strong winds or structural failure". This set of hypotheses is then passed to the Bayesian inference unit for probability evaluation.

[0101] Building upon this, the function of the Bayesian inference unit is to quantitatively calculate the confidence level of each causal explanation hypothesis provided by the hypothesis generation unit, in order to determine the most probable explanation. The operational mechanism of this unit is based on Bayes' theorem, and its basic form can be expressed as: ; in, In order to provide evidence Assuming The posterior probability of its validity; In the assumption Evidence in cases where it is established The likelihood probability of occurrence; Assumption The prior probability.

[0102] In embodiments of the present invention, the evidence combined by the Bayesian inference unit It is a composite evidence set, the sources of which include real-time perceived event features provided by the data perception and feature extraction module, and information from the family dynamic causal knowledge graph provided by the knowledge graph construction and maintenance module.

[0103] To further illustrate, this inference process can be formalized more precisely. Let the set of hypotheses be... The characteristics of the perceived events observed in real time are The background information extracted from the knowledge graph is The goal of a Bayesian inference unit is to provide each hypothesis with... Calculate its posterior probability and find the hypothesis that maximizes that probability. The core calculation process can be expressed as follows: ; in, For the solution to be found, after combining all the evidence, the first... The posterior probability of a hypothesis. In the context of knowledge graphs Known and assumed Under the given conditions, real-time event characteristics were observed. The conditional probability; For background information based solely on knowledge graphs The resulting hypothesis The prior probability.

[0104] Background information extracted from the family dynamic causal knowledge graph Specifically, this includes, but is not limited to: user-planned event information stored in the knowledge graph, and the probability distribution of context state nodes generated by the time-probability state prediction unit. For example, when inferring the aforementioned "window open" event, if the "family routine status" node in the knowledge graph shows a probability of being in a "nighttime sleep" state of 0.99, and there is no user-planned "nighttime ventilation" event, this will significantly reduce the probability of the hypothesis being true. The prior probability of "family members opening windows for ventilation" At the same time, increase the hypothesis The prior probability of "illegal intrusion by external personnel".

[0105] After calculating the posterior probability of all hypotheses, the Bayesian inference unit determines the hypothesis with the highest probability value and uses it as the final inference result, i.e., the causal explanation hypothesis, which is then output to the human-machine collaboration and knowledge alignment module. This process transforms an isolated physical perception event into a logical judgment with a clear causal explanation and quantifiable confidence level, providing a decision-making basis for the system to achieve accurate predictive early warning.

[0106] In one specific embodiment, the human-machine collaboration and knowledge alignment module constitutes the bridge for interaction between the predictive home security system of the present invention and the user, and undertakes the key function of closed-loop learning. The function of this module is not only to transform the abstract decision results of the causal reasoning and risk assessment module into information that users can understand and act on, but also to systematically receive, analyze and utilize user feedback to continuously correct and supplement the dynamic causal knowledge graph of the family, thereby realizing the dynamic alignment between system knowledge and user cognition.

[0107] To achieve the above functions, the internal technical solution of the human-machine collaboration and knowledge alignment module can be specifically set to include an interpretable early warning generation unit and a feedback processing and knowledge update unit.

[0108] Specifically, the explanatory warning generation unit is responsible for externally presenting the inference results from within the system. This unit receives the most probable causal explanation hypothesis (denoted as ) output by the upstream causal inference and risk assessment module. ) and their corresponding posterior probabilities (denoted as ) A key technical feature of this unit is that when generating a warning notification, it does not only present the event itself, but also presents the inferred causal explanation hypothesis and the uncertainty value representing the hypothesis to the user.

[0109] The uncertainty here can be quantified as the probability that the assumption is false. In a preferred embodiment, the calculation formula is as follows: ; In the formula, That is, an assumption The uncertainty measure will be provided along with the assumptions. This is the hypothesis with the highest posterior probability. For posterior probability, For the evidence set.

[0110] Furthermore, to avoid information overload and improve the effectiveness of early warnings, the interpretable early warning generation unit is also configured to categorize and push early warning notifications according to the type and confidence level of the causal explanation hypothesis. This function can be implemented based on a dynamically calculated risk score, which comprehensively considers the inherent severity of the event and the system's confidence level in its judgment. Its calculation method can be formalized as follows: ; in, Assumption Risk score, For predefined, and hypothetical types The associated inherent severity level coefficient, The confidence level of this hypothesis, i.e., its posterior probability. .

[0111] Calculate the risk score Then, the unit classifies the warnings into different levels (e.g., high, medium, low) based on the preset threshold range, and matches them with different push channels and strategies. For example, high-risk events trigger an audible alarm and call emergency contacts, medium-risk events trigger an application push with a prominent prompt sound, and low-risk events may only generate a silent log record within the application.

[0112] After the system completes the push of the warning information, the feedback processing and knowledge update unit is responsible for processing the user's response to form a closed loop of knowledge alignment. This unit can receive and parse the user's feedback information, which can be structured input generated through preset interactive buttons on the application interface (such as "Confirm as my operation", "Mark as false alarm", "Add as trusted visitor") or natural language text entered by the user. For natural language feedback, this unit can embed a lightweight natural language understanding engine to extract the user's core intent and key entity information.

[0113] Please see Figure 6 , Figure 6 This is a flowchart illustrating another predictive method for home security provided in an embodiment of this application, as shown below. Figure 6As shown, the core function of this unit is to directly create or modify nodes and edges in the family dynamic causal knowledge graph based on the results obtained from parsing user feedback. Taking a specific scenario as an example: when the system issues an alarm for an unrecognized face, the user feedback information is "This is my friend A". The feedback processing and knowledge update unit first parses the feedback and extracts the intent as "confirm identity", the entity as {name: "person A", relationship: "friend"}.

[0114] Subsequently, the unit will initiate a transactional operation request to the knowledge graph storage unit. This request will contain a series of instructions to create a new entity node in the knowledge graph to represent the unknown person and establish a relationship between the new entity node and existing entity nodes in the knowledge graph. For example, in a graph database, the following operation might be performed: First, create a new node with the category "Person" and the attribute {Name: "Person A"}; then, find the "Person" node representing the current user and establish a directed edge with the category "Friend" between these two nodes.

[0115] Through this mechanism, every valid feedback from a user is directly and permanently solidified into part of the knowledge graph. In subsequent operations, when the system detects "person A" again, it can directly query its identity and relationships from the knowledge graph, thereby avoiding repeated false alarms. This human-machine collaborative knowledge alignment process enables the system to continuously learn from the user's experience, and its knowledge base becomes more accurate and complete over time, thereby continuously improving the accuracy of future predictions and judgments.

[0116] Corresponding to the aforementioned method for predicting home security, this invention also proposes a device for predicting home security. Since the device embodiments of this invention correspond to the method embodiments described above, details not disclosed in the device embodiments can be referred to in the method embodiments, and will not be repeated here.

[0117] Figure 7 This is a schematic diagram of the structure of a predictive device for home security provided in an embodiment of the present disclosure, as shown below. Figure 7 As shown, it includes: The acquisition unit 21 is used to acquire visual data and non-visual data, perform target detection and behavior analysis on the visual data, perform standardization processing on the non-visual data, and fuse the processed visual data and non-visual data to generate a first perceptual event feature. Construction unit 22 is used to construct a dynamic causal knowledge graph based on the features of the first perceived event; wherein, the knowledge graph includes nodes representing family entities, events and situational states; and edges representing the relationships between nodes; The matching unit 23 is used to receive the second perceived event features and match the second perceived event features with the preset causal chain template in the knowledge graph; The generation unit 24 is used to mark the second perceived event feature as a causal abnormal event and generate an associated node if the matching fails; The computing unit 25 is used to generate multiple sets of causal explanation hypotheses based on the dynamic causal knowledge graph, calculate the posterior probability of each hypothesis, and select the hypothesis with the highest posterior probability as the causal explanation result. The early warning unit 26 is used to calculate the uncertainty measure and risk score based on the posterior probability of the causal explanation result, generate interpretable early warning information containing causal logic, and execute early warning for the user.

[0118] Furthermore, in one possible implementation of this disclosure, the acquisition unit 21 is further configured to: The object detection algorithm is used to identify people and objects in the image and output category labels and location coordinates; the pose estimation algorithm is used to extract the coordinates of key points of the human skeleton, and the pose changes in continuous frames are analyzed by combining a time series model to identify human behavior patterns and generate behavior labels. An extreme value normalization device is used to map environmental parameters and equipment operation data in non-visual data to a preset value range, and a timestamp accurate to milliseconds is added to each processed data, which is then encapsulated as environmental status information in key-value pair format. By aggregating personnel information, behavioral tags, and environmental status information within the same time window, structured perception event features with multidimensional information are generated.

[0119] Furthermore, in one possible implementation of this disclosure, the computing unit 25 is further configured to: The dynamic time warping algorithm is used to calculate the similarity between the event sequence corresponding to the features of newly perceived events and the preset causal chain template; If the similarity is lower than a preset threshold, it is determined to be a causal anomaly, a causal anomaly event node is generated, and an association edge is established between the node and the event node that triggered the anomaly. Based on historical event records, current situational state probability distribution, and entity attributes in the knowledge graph, multiple mutually exclusive causal explanation hypotheses are generated. The prior probability of each hypothesis is calculated using the frequency of historical events in the knowledge graph as the hypothesis, and the posterior probability of each hypothesis is calculated using the characteristics of the current perceived event and the probability of the situational state as evidence. The hypothesis with the highest posterior probability is selected as the final causal explanation, and the relationship between the hypothesis and the evidence is recorded and stored in the knowledge graph.

[0120] Furthermore, in one possible implementation of this disclosure, the early warning unit 26 is further configured to: Warning levels are classified according to risk scores and corresponding push notification methods are assigned: different warning levels correspond to different push notification methods. Obtain user feedback and update the knowledge graph based on the user feedback.

[0121] Furthermore, in one possible implementation of this disclosure, the apparatus further includes: The acquisition unit 21 is also used to acquire non-visual data of the current time period as environmental baseline values ​​when no event is detected within a consecutive preset number of time windows, including temperature and humidity, gas concentration, and equipment standby power parameters. The calculation unit 25 is also used to calculate the drift coefficient between real-time environmental parameters and baseline values; The early warning unit 26 is also used to determine that the environmental parameters are abnormal if the drift coefficient exceeds a preset threshold, generate an environmental parameter drift event node, update the attributes of the corresponding environmental state node in the knowledge graph, and trigger a low-level early warning to prompt the user to check the environment or equipment.

[0122] It should be noted that the foregoing explanation of the method embodiments also applies to the apparatus of the embodiments of this disclosure, and the principle is the same. Therefore, the embodiments of this disclosure are not limited thereto.

[0123] According to embodiments of this disclosure, this disclosure also provides an electronic device, a readable storage medium, and a computer program product.

[0124] Figure 8 A schematic block diagram of an example electronic device 400 that can be used to implement embodiments of the present disclosure is shown. The electronic device is intended to represent various forms of digital computers, such as laptop computers, desktop computers, workstations, personal digital assistants, servers, blade servers, mainframe computers, and other suitable computers. The electronic device may also represent various forms of mobile devices, such as personal digital assistants, cellular phones, smartphones, wearable devices, and other similar computing devices. The components shown herein, their connections and relationships, and their functions are merely illustrative and are not intended to limit the implementation of the present disclosure described and / or claimed herein.

[0125] like Figure 8As shown, device 400 includes a computing unit 401, which can perform various appropriate actions and processes based on a computer program stored in ROM (Read-Only Memory) 402 or a computer program loaded from storage unit 408 into RAM (Random Access Memory) 403. RAM 403 may also store various programs and data required for the operation of device 400. The computing unit 401, ROM 402, and RAM 403 are interconnected via bus 404. I / O (Input / Output) interface 405 is also connected to bus 404.

[0126] Multiple components in device 400 are connected to I / O interface 405, including: input unit 406, such as keyboard, mouse, etc.; output unit 407, such as various types of monitors, speakers, etc.; storage unit 408, such as disk, optical disk, etc.; and communication unit 409, such as network card, modem, wireless transceiver, etc. Communication unit 409 allows device 400 to exchange information / data with other devices through computer networks such as the Internet and / or various telecommunications networks.

[0127] The computing unit 401 can be a variety of general-purpose and / or special-purpose processing components with processing and computing capabilities. Some examples of the computing unit 401 include, but are not limited to, CPUs (Central Processing Units), GPUs (Graphics Processing Units), various special-purpose AI (Artificial Intelligence) computing chips, various computing units running machine learning model algorithms, DSPs (Digital Signal Processors), and any suitable processor, controller, microcontroller, etc. The computing unit 401 performs the various methods and processes described above, such as predictive methods for home security. For example, in some embodiments, the predictive methods for home security may be implemented as computer software programs tangibly contained in a machine-readable medium, such as storage unit 408. In some embodiments, part or all of the computer program may be loaded and / or installed on device 400 via ROM 402 and / or communication unit 409. When the computer program is loaded into RAM 403 and executed by the computing unit 401, one or more steps of the methods described above may be performed. Alternatively, in other embodiments, the computing unit 401 may be configured to perform the aforementioned home security prediction method by any other suitable means (e.g., by means of firmware).

[0128] Various implementations of the systems and techniques described above herein can be implemented in digital electronic circuit systems, integrated circuit systems, FPGAs (Field Programmable Gate Arrays), ASICs (Application-Specific Integrated Circuits), ASSPs (Application-Specific Standard Products), SOCs (System-on-Chips), CPLDs (Complex Programmable Logic Devices), computer hardware, firmware, software, and / or combinations thereof. These various implementations may include implementations in one or more computer programs that can be executed and / or interpreted on a programmable system including at least one programmable processor, which may be a dedicated or general-purpose programmable processor, capable of receiving data and instructions from a storage system, at least one input device, and at least one output device, and transmitting data and instructions to the storage system, the at least one input device, and the at least one output device.

[0129] The program code used to implement the methods of this disclosure may be written in any combination of one or more programming languages. This program code may be provided to a processor or controller of a general-purpose computer, special-purpose computer, or other programmable data processing apparatus, such that when executed by the processor or controller, the program code causes the functions / operations specified in the flowcharts and / or block diagrams to be implemented. The program code may be executed entirely on a machine, partially on a machine, as a standalone software package partially on a machine and partially on a remote machine, or entirely on a remote machine or server.

[0130] In the context of this disclosure, a machine-readable medium can be a tangible medium that may contain or store a program for use by or in conjunction with an instruction execution system, apparatus, or device. A machine-readable medium can be a machine-readable signal medium or a machine-readable storage medium. A machine-readable medium can be, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, apparatus, or devices, or any suitable combination of the foregoing. More specific examples of machine-readable storage media include electrical connections based on one or more wires, portable computer disks, hard disks, RAM, ROM, EPROM (Electrically Programmable Read-Only Memory) or flash memory, optical fiber, CD-ROM (Compact Disc Read-Only Memory), optical storage devices, magnetic storage devices, or any suitable combination of the foregoing.

[0131] To provide interaction with a user, the systems and techniques described herein can be implemented on a computer having: a display device for displaying information to the user (e.g., a CRT (Cathode-Ray Tube) or LCD (Liquid Crystal Display) monitor); and a keyboard and pointing device (e.g., a mouse or trackball) through which the user provides input to the computer. Other types of devices can also be used to provide interaction with the user; for example, feedback provided to the user can be any form of sensory feedback (e.g., visual feedback, auditory feedback, or tactile feedback); and input from the user can be received in any form (including sound input, voice input, or tactile input).

[0132] The systems and technologies described herein can be implemented in computing systems that include backend components (e.g., as data servers), or computing systems that include middleware components (e.g., application servers), or computing systems that include frontend components (e.g., user computers with graphical user interfaces or web browsers through which users can interact with implementations of the systems and technologies described herein), or any combination of such backend, middleware, or frontend components. The components of the system can be interconnected via digital data communication of any form or medium (e.g., communication networks). Examples of communication networks include LANs (Local Area Networks), WANs (Wide Area Networks), the Internet, and blockchain networks.

[0133] Computer systems can include clients and servers. Clients and servers are generally geographically separated and typically interact via communication networks. The client-server relationship is created by computer programs running on the respective computers and having a client-server relationship with each other. A server can be a cloud server, also known as a cloud computing server or cloud host, a hosting product within the cloud computing service system that addresses the shortcomings of traditional physical hosts and VPS (Virtual Private Server) services, such as high management difficulty and weak business scalability. Servers can also be servers for distributed systems or servers incorporating blockchain technology.

[0134] It's important to note that artificial intelligence (AI) is the study of enabling computers to simulate certain human thought processes and intelligent behaviors (such as learning, reasoning, thinking, and planning). It encompasses both hardware and software technologies. AI hardware technologies generally include sensors, dedicated AI chips, cloud computing, distributed storage, and big data processing. AI software technologies primarily include computer vision, speech recognition, natural language processing, machine learning / deep learning, big data processing, and knowledge graph technologies.

[0135] It should be understood that the various forms of processes shown above can be used to rearrange, add, or delete steps. For example, the steps described in this disclosure can be executed in parallel, sequentially, or in different orders, as long as the desired result of the technical solution disclosed in this disclosure can be achieved, and this is not limited herein.

[0136] The specific embodiments described above do not constitute a limitation on the scope of protection of this disclosure. Those skilled in the art should understand that various modifications, combinations, sub-combinations, and substitutions can be made according to design requirements and other factors. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this disclosure should be included within the scope of protection of this disclosure.

Claims

1. A predictive method for home security, characterized in that, Includes the following steps: Visual data and non-visual data are collected, target detection and behavior analysis are performed on the visual data, the non-visual data are standardized, and the processed visual data and non-visual data are fused to generate a first perceptual event feature. A dynamic causal knowledge graph is constructed based on the features of the first perceived event; wherein, the knowledge graph includes nodes representing family entities, events and situational states; and edges representing the relationships between nodes; Receive the second perceived event features and match the second perceived event features with the preset causal chain template in the knowledge graph; If the match fails, the second perceived event feature is marked as a causal abnormal event and an associated node is generated; Based on the dynamic causal knowledge graph, multiple sets of causal explanation hypotheses are generated, the posterior probability of each hypothesis is calculated, and the hypothesis with the highest posterior probability is selected as the causal explanation result. Based on the posterior probability of the causal explanation results, an uncertainty measure and risk score are calculated, and interpretable early warning information containing causal logic is generated to issue warnings to users.

2. The method according to claim 1, characterized in that, The process of collecting visual and non-visual data, performing target detection and behavior analysis on the visual data, standardizing the non-visual data, and fusing the processed visual and non-visual data to generate the first perceptual event feature includes: The object detection algorithm is used to identify people and objects in the image and output category labels and location coordinates; the pose estimation algorithm is used to extract the coordinates of key points of the human skeleton, and the pose changes in continuous frames are analyzed by combining a time series model to identify human behavior patterns and generate behavior labels. The extreme value normalization method is used to map environmental parameters and device operation data in non-visual data to a preset value range, and a timestamp accurate to milliseconds is added to each processed data, which is then encapsulated as environmental status information in key-value pair format. By aggregating personnel information, behavioral tags, and environmental status information within the same time window, structured perception event features with multidimensional information are generated.

3. The method according to claim 1, characterized in that, The process of generating multiple causal explanation hypotheses based on the dynamic causal knowledge graph, calculating the posterior probability of each hypothesis, and selecting the hypothesis with the highest posterior probability as the causal explanation result includes: The dynamic time warping algorithm is used to calculate the similarity between the event sequence corresponding to the features of the newly perceived event and the preset causal chain template; If the similarity is lower than a preset threshold, it is determined to be a causal anomaly, a causal anomaly event node is generated, and an association edge is established between the node and the event node that triggered the anomaly. Based on historical event records, current situational state probability distribution, and entity attributes in the knowledge graph, multiple mutually exclusive causal explanation hypotheses are generated. Using the frequency of historical events in the knowledge graph as the prior probability of the hypothesis, and using the characteristics of the current perceived event and the probability of the situational state as evidence, the posterior probability of each hypothesis is calculated. The hypothesis with the highest posterior probability is selected as the final causal explanation, and the relationship between the hypothesis and the evidence is recorded and stored in the knowledge graph.

4. The method according to claim 1, characterized in that, The step of calculating uncertainty measures and risk scores based on the posterior probability of the causal explanation results, generating interpretable early warning information containing causal logic, and issuing early warnings to users includes: Warning levels are classified according to risk scores and corresponding push notification methods are assigned: different warning levels correspond to different push notification methods. Obtain user feedback and update the knowledge graph based on the user feedback.

5. The method according to claim 1, characterized in that, The method further includes: When no event is detected within a preset number of consecutive time windows, non-visual data for the current time period is collected as the environmental baseline value, including temperature, humidity, gas concentration, and equipment standby power parameters. Calculate the drift coefficients of real-time environmental parameters relative to baseline values; If the drift coefficient exceeds the preset threshold, the environmental parameters are determined to be abnormal, an environmental parameter drift event node is generated, the attributes of the corresponding environmental state node in the knowledge graph are updated, and a low-level warning is triggered to prompt the user to check the environment or equipment.

6. A predictive device for home security, characterized in that, Includes the following steps: The acquisition unit is used to acquire visual data and non-visual data, perform target detection and behavior analysis on the visual data, perform standardization processing on the non-visual data, and fuse the processed visual data and non-visual data to generate a first perceptual event feature. A construction unit is used to construct a dynamic causal knowledge graph based on the features of the first perceived event; wherein, the knowledge graph includes nodes representing family entities, events and situational states; and edges representing the relationships between nodes; The matching unit is used to receive the second perceived event features and match the second perceived event features with the preset causal chain template in the knowledge graph; A generation unit is used to mark the second perceived event feature as a causal abnormal event and generate an associated node if the matching fails. The computing unit is used to generate multiple sets of causal explanation hypotheses based on the dynamic causal knowledge graph, calculate the posterior probability of each hypothesis, and select the hypothesis with the highest posterior probability as the causal explanation result. The early warning unit is used to calculate uncertainty measure and risk score based on the posterior probability of the causal explanation result, generate interpretable early warning information containing causal logic, and execute early warning for the user.

7. The apparatus according to claim 6, characterized in that, The acquisition unit is also used for: The object detection algorithm is used to identify people and objects in the image and output category labels and location coordinates; the pose estimation algorithm is used to extract the coordinates of key points of the human skeleton, and the pose changes in continuous frames are analyzed by combining a time series model to identify human behavior patterns and generate behavior labels. An extreme value normalization device is used to map environmental parameters and equipment operation data in non-visual data to a preset value range, and a timestamp accurate to milliseconds is added to each processed data, which is then encapsulated as environmental status information in key-value pair format. By aggregating personnel information, behavioral tags, and environmental status information within the same time window, structured perception event features with multidimensional information are generated.

8. An electronic device, characterized in that, include: At least one processor; as well as A memory communicatively connected to the at least one processor; wherein, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the method of any one of claims 1-5.

9. A non-transitory computer-readable storage medium storing computer instructions, characterized in that, The computer instructions are used to cause the computer to perform the method according to any one of claims 1-5.

10. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method according to any one of claims 1-5.

Images