Event exception identification method and device and computer program product

Abstract

The invention relates to an event anomaly recognition method, an event anomaly recognition device and a computer program product. The event anomaly recognition method comprises the steps of recognizing a target event based on input data; determining a plurality of current modes of the target event based on identification results of the target event in a plurality of time periods satisfying a preset time association condition, each current mode corresponding to a time period; and determining whether the target event is abnormal or not based on the deviation values of the plurality of current modes and the historical mode, the step of determining whether the target event is abnormal or not comprises: responding to the deviation values of at least two current modes of a predetermined number in the plurality of current modes relative to the historical mode meeting a first standard, and determining whether the target event is abnormal or not. And determining that the target event has the first anomaly.

Description

Technical Field

[0001] This disclosure relates to the field of event anomaly identification, specifically to event anomaly identification methods, event anomaly identification devices, and computer program products used to identify anomalies in target events. Background Technology

[0002] With the development of artificial intelligence technology, security cameras have gradually evolved from devices with only video capture capabilities to video capture devices with a certain degree of intelligent analysis. Existing security cameras typically use AI models based on the video data they capture to achieve area intrusion and boundary crossing tracking and warnings for target objects (e.g., vehicles), or combine sensor data collected by sensors with preset behavioral models to identify simple behaviors of target objects (e.g., pets using the toilet). However, these methods usually rely on fixed behavioral patterns and mainly focus on judging single target events, failing to capture the trend changes of target events over time. This leads to the neglect of gradual anomalies accumulated in target events, affecting the user's analysis and warning of potential risks associated with target events. Summary of the Invention

[0003] In view of the above problems, this disclosure provides an event anomaly identification method, an event anomaly identification device, and a computer program product, which can analyze the changes of a target event in the time dimension to capture the anomalies of the target event.

[0004] According to one aspect of this disclosure, an event anomaly identification method is provided, comprising: identifying a target event based on input data; determining multiple current patterns of the target event based on the identification results of the target event in multiple time periods satisfying preset time association conditions, each current pattern corresponding to a time period; and determining whether the target event has an anomaly based on the deviation values ​​of the multiple current patterns from historical patterns, wherein determining whether the target event has an anomaly includes: determining that the target event has a first anomaly in response to at least two of the multiple current patterns having deviation values ​​relative to the historical patterns that satisfy a first criterion.

[0005] According to one embodiment of this disclosure, the preset time association condition is sequential.

[0006] According to one embodiment of this disclosure, the first criterion is that the absolute value of the deviation of each of the at least two current patterns relative to the historical pattern is greater than a threshold.

[0007] According to one embodiment of this disclosure, the first criterion is a monotonically changing deviation of the at least two current patterns relative to the historical patterns.

[0008] According to one embodiment of this disclosure, determining whether the target event is abnormal further includes: determining that the target event is abnormal in response to a deviation value of a first current mode among the plurality of current modes relative to the historical mode satisfying a second criterion.

[0009] According to one embodiment of this disclosure, the event anomaly identification method further includes: identifying the target event based on historical data; and determining the historical pattern of the target event based on the identification results of the target event in multiple historical time periods that meet preset time association conditions.

[0010] According to one embodiment of this disclosure, the description information of the target event includes features in multiple dimensions, which include at least: object features; and one or more of action features and state features.

[0011] According to one embodiment of this disclosure, the input data includes one or more of the following: image data acquired by an image acquisition device; and perception data acquired by a sensing device.

[0012] According to one embodiment of this disclosure, the current mode or historical mode of the target event includes one or more of the following: the occurrence time of the target event; the duration of the target event; the number of times the target event occurs; the number of times the target event is interrupted; the time interval between two adjacent target events; and the spatial distance between the event feature vectors of two target events.

[0013] According to one embodiment of this disclosure, determining the historical pattern of the target event based on the identification results of the target event in multiple historical time periods that satisfy preset time association conditions includes: determining multiple candidate historical patterns of the target event in the multiple historical time periods based on the identification results of the target event in the multiple historical time periods that satisfy preset time association conditions; and determining the historical pattern of the target event based on the average, maximum, minimum, median, mode, or predicted value of the multiple candidate historical patterns.

[0014] According to one embodiment of this disclosure, based on the first anomaly and the second anomaly, an anomaly identification result of the target event is output.

[0015] According to one embodiment of this disclosure, the event anomaly identification method further includes: in response to an anomaly occurring in the target event, outputting personalized information related to the anomaly of the target event.

[0016] According to another aspect of this disclosure, an event anomaly identification apparatus is provided, comprising: a processor; and a memory storing a computer program, wherein when the computer program is executed by the processor, the processor performs the event anomaly identification method as described above.

[0017] According to another aspect of this disclosure, a computer-readable medium is provided, including a computer program that, when executed by a processor, implements the event anomaly identification method as described above.

[0018] According to another aspect of this disclosure, a computer program product is provided, including a computer program that, when executed by a processor, performs the event anomaly identification method as described above.

[0019] According to embodiments of this disclosure, it is possible to analyze changes in a target event over time, thereby capturing anomalies in the target event and aiding in the early or more accurate identification of abnormal events. Attached Figure Description

[0020] Figure 1 This is a diagram illustrating an example scenario where an artificial intelligence model is used to identify and analyze the behavior of a target object.

[0021] Figure 2 This is a flowchart of an event anomaly identification method according to an embodiment of the present disclosure.

[0022] Figure 3 This is an example application scenario diagram of the event anomaly identification method according to embodiments of the present disclosure.

[0023] Figure 4 This is a diagram illustrating an example of a first standard in the event anomaly identification method of the embodiments of this disclosure.

[0024] Figure 5 This is a diagram illustrating another example of a first standard in the event anomaly identification method of the embodiments of this disclosure.

[0025] Figure 6 This is a diagram illustrating an example of a second standard in the event anomaly identification method of the embodiments of this disclosure.

[0026] Figure 7 This is a schematic block diagram illustrating an event anomaly identification device according to an embodiment of the present disclosure. Detailed Implementation

[0027] Hereinafter, embodiments of the present disclosure will be described in detail with reference to the accompanying drawings. While some embodiments of the present disclosure are shown in the drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as limited to the embodiments set forth herein. Rather, these embodiments are provided to provide a more thorough and complete understanding of the present disclosure. It should be understood that the accompanying drawings and embodiments of the present disclosure are for illustrative purposes only and are not intended to limit the scope of protection of the present disclosure.

[0028] It should be understood that the steps described in the method embodiments of this disclosure may be performed in different orders and / or in parallel. Furthermore, method embodiments may include other steps and / or omit certain steps.

[0029] The term "comprising" and its variations as used herein are open-ended inclusions, meaning "including but not limited to". The term "based on" means "at least partially based on". The term "one embodiment" means "at least one embodiment"; the term "another embodiment" means "at least one additional embodiment"; the term "some embodiments" means "at least some embodiments". Definitions of other terms will be given in the description below.

[0030] It should be understood that the concepts of "first" and "second" mentioned in this disclosure are used only to distinguish different devices, modules, units, models, data, etc., and are not used to limit the order of the functions performed by these devices, modules, units, models, data, or their interdependencies.

[0031] Existing security cameras typically rely on the video data they collect to identify the behavior of target objects through artificial intelligence models. However, these methods usually focus on judging single events and cannot capture the trend changes of target events over time. This leads to the neglect of gradual anomalies accumulated in target events, affecting users' analysis and early warning of potential risks associated with target events.

[0032] Figure 1 This diagram illustrates an example scenario where an artificial intelligence model is used to identify and analyze the behavior of a target object. In this scenario, a pet behavior analysis device based on a security camera uses an artificial intelligence model to count the number of times a pet cat grooms itself, based on surveillance video captured by the camera. Figure 1 The line graph shown is a statistical representation of the number of times a pet cat grooms itself per hour. The horizontal axis represents time, and the vertical axis represents the number of grooming sessions in the preceding hour. For example, the horizontal axis 11:00 corresponds to a vertical axis of 4, indicating that the pet cat groomed itself 4 times between 10:00 and 11:00. Figure 1As shown, the standard reference value for grooming behavior of the pet cat is set at 5 times / hour (as shown by the dotted line M). It is stipulated that as long as the deviation from the standard reference value is within ±2 times / hour, that is, as long as the number of grooming behaviors is within the range of 3 to 7 times / hour, the grooming behavior of the pet cat can be regarded as normal (as shown by the dotted lines N1 and N2). Figure 1 The "×" symbol indicates that the cat was detected grooming itself 8 times between 15:00 and 16:00, exceeding the preset range. Therefore, it is judged as an abnormal behavior of "excessive grooming" during this period. In this case, after determining that the cat's behavior is "excessive grooming," the pet behavior analysis device based on the security camera notifies the user of the abnormal behavior and preliminarily analyzes whether the abnormal behavior is related to the cat's skin problems or environmental anxiety, thereby monitoring and issuing early warnings for abnormal pet behavior.

[0033] However, as Figure 1 As shown, during the period from 3:00 PM to 4:00 PM, the cat's grooming behavior gradually increased from 3 times / hour to 4 times / hour, then to 6 times / hour, and finally to 7 times / hour, with an overall upward trend in the number of grooming sessions per hour. In this case, because the number of grooming sessions never exceeded the preset range, it was not judged as an abnormal behavior of "excessive grooming." However, even if it does not exceed the preset range, a gradual increase in the frequency of grooming may indicate early skin problems in the cat. In such cases, failure to provide timely warnings may delay the optimal treatment period.

[0034] As mentioned above, even if the characteristics of the target event corresponding to the target object (e.g., frequency of occurrence, duration, etc.) always fall within the normal range, it does not necessarily mean that the target object is not abnormal. In other words, even if the characteristics of a single target event are within the normal range, it is sometimes necessary to consider the trend of the target event's characteristics over time to determine whether an anomaly has occurred.

[0035] In view of the above problems, embodiments of this disclosure propose an event anomaly identification method, including: identifying a target event based on input data, and determining multiple current patterns of the target event based on the identification results of the target event in multiple time periods that satisfy preset time association conditions; determining whether the target event is abnormal based on the deviation values ​​of the multiple current patterns and historical patterns, wherein determining whether the target event is abnormal includes: determining that the target event is abnormal in response to at least two of the multiple current patterns having deviation values ​​relative to the historical patterns that satisfy a specific standard.

[0036] The event anomaly identification method according to embodiments of the present disclosure will be described in detail below. Figure 2This is a flowchart of an event anomaly identification method 200 according to an embodiment of this disclosure. (As follows) Figure 2 As shown, the event anomaly identification method 200 may include steps S210 to S230.

[0037] In step S210, a target event is identified based on the input data. The input data is data related to the behavior of the target object, and may include image data acquired by an image acquisition device (e.g., video of a pet eating captured by a surveillance camera) and / or sensory data acquired by a sensing device (e.g., sensory data on the weight change of a pet food bowl acquired by a weight sensor, ambient temperature data acquired by a temperature sensor, etc.). A target event refers to the state of the target object of interest and / or an event with clear behavioral semantics performed by the target object, such as "a pet cat eating," "an elderly person falling," or "a child studying." The descriptive information used to describe the target event may include features in multiple dimensions. For example, the multiple dimensions of features may include object features, action features, and one or more of state features. In some embodiments, the object feature may be a pet cat, and the action feature may be behaviors such as running or eating. Furthermore, the object feature may be a pet cat's water bowl, and the state feature may be the water level in the water bowl, etc. In some embodiments, an artificial intelligence model may be used to analyze the input data and identify the target event. For example, an artificial intelligence model with video content analysis capabilities (e.g., Inflated 3DConvNet (I3D)) can be used to analyze video data captured by security cameras and identify target events in the video. Of course, the methods for analyzing input data are not limited to this. For instance, a pattern library containing behavioral patterns corresponding to target events can be pre-set, and the input data can be matched and analyzed based on the pattern library to identify the target events.

[0038] In step S220, based on the identification results of the target event in multiple time periods that satisfy preset time association conditions, multiple current patterns of the target event are determined, each current pattern corresponding to a time period. In some embodiments, the preset time association conditions can be sequential, that is, multiple current patterns of the target event are determined based on the identification results of the target event in multiple sequential time periods. In this embodiment, the sequential nature of the preset time association conditions is used as an example for explanation. The identification results can indicate whether the target event was detected in the corresponding time period, and can also include how many times the target event was detected in the corresponding time period, the start time of the target event, the end time of the target event, etc. Each current pattern corresponds to a time period, and different lengths of time periods can be set according to the characteristics of different target events. For example, if the target event is "pet cat eating", the time period can be set to one day; if the target event is "pet cat drinking water", the time period can be set to one hour. The current pattern obtained based on the recognition results may include one or more of the following: the occurrence time of the target event, the duration of the target event, the number of times the target event occurs, the number of times the target event is interrupted, the time interval between two adjacent target events, and the spatial distance between the event feature vectors of two target events. For example, the current pattern may be the number of times a pet cat eats in a day, the time, duration, and amount of food consumed at each feeding, and / or the number of times and amount of water consumed by the pet cat within an hour. In some embodiments, the target event can be converted into an event feature vector using an AI model, and the spatial distance between the event feature vectors of two target events can be calculated to reflect the similarity or difference between the two target events.

[0039] In step S230, based on the deviation values ​​between the plurality of current modes and historical modes, it is determined whether the target event has become abnormal. Determining whether the target event has become abnormal includes: in response to at least two of the plurality of current modes having a predetermined number of deviation values ​​relative to the historical mode that meet a first criterion, it is determined that the target event has become abnormal.

[0040] According to the first anomaly identification method in step S230, it is not based on the judgment of a single pattern in a single time period, but on the comprehensive judgment of multiple patterns in multiple sequential time periods, which can improve the accuracy of identification and may help in the early identification of abnormal events.

[0041] The following reference Figure 3 An example implementation of step S230 will be described below. Figure 3 This is an example application scenario diagram of the event anomaly identification method according to embodiments of the present disclosure.

[0042] Historical patterns are used to describe the state characteristics of a target event over a historical time dimension and serve as a standard for calculating the deviation value of the current pattern. These historical patterns can be obtained by summarizing or statistically analyzing patterns of the target event across multiple historical time periods. Similar to the aforementioned current pattern, historical patterns may include one or more of the following: the occurrence time of the target event, the duration of the target event, the number of times the target event occurred, the number of times the target event was interrupted, the time interval between two consecutive target events, and the spatial distance between the event feature vectors of two consecutive target events.

[0043] In some embodiments, the target event can be identified based on historical data, and a historical pattern of the target event can be determined based on the identification results of the target event in multiple historical time periods that satisfy preset time correlation conditions. The historical data consists of multiple input data recorded in the past. In some embodiments, multiple candidate historical patterns of the target event in the multiple historical time periods that satisfy preset time correlation conditions can be determined based on the identification results of the target event in the multiple historical time periods, and the historical pattern of the target event can be determined based on the average, maximum, minimum, median, mode, or fitted number of the multiple candidate historical patterns. For example, such as... Figure 3 As shown, multiple historical data points can be sequentially selected from multiple historical time periods in chronological order to obtain multiple candidate historical patterns for a target event. Then, based on these candidate historical patterns, the historical pattern of the target event can be obtained. For example, "pet cat eating" can be taken as the target event, the time period can be set to one day, and video data of the pet from January 1st to January 20th can be selected as historical data. Based on this video data, the "pet cat eating" event can be identified each day, thereby calculating the daily eating frequency of the pet cat over the 20 days from January 1st to January 20th as candidate historical patterns, and calculating the average daily eating frequency over these 20 days as the historical pattern of the pet cat eating event.

[0044] The deviation value between the current pattern and the historical pattern is a quantitative indicator of the degree of deviation between the current pattern and the historical pattern of the target event. For example, the deviation value can be the statistical difference between the current pattern and the historical pattern of the target event. In some embodiments, the deviation value can be the difference between the current pattern and the historical pattern of the target event, and the degree of deviation can be the absolute value of the deviation value. The larger the absolute value of the deviation value, the greater the change in the current state relative to the historical state, and the greater the possibility that the current state is abnormal; the smaller the absolute value of the deviation value, the closer the current state is to the historical state, and the smaller the possibility that the current state is abnormal. For example, if the historical pattern of a pet cat's eating event is 4 times / day, and if on a certain day the pet cat is found to have eaten only once, then the deviation value between the current pattern and the historical pattern for that day is -3. The absolute value of the deviation value is relatively large, which can be used to determine that the pet cat's eating behavior is abnormal on that day.

[0045] It should be understood that the historical pattern is not limited to fixed values. In some embodiments, the historical pattern can also be a trend prediction calculated based on the multiple candidate historical patterns. For example, when the target event is the weight of a pet cat's water bowl, if the selected candidate historical data are 500g, 450g, 400g, and 350g, a trend function for the weight change of the water bowl can be fitted based on the above weight data. In this case, when analyzing the current pattern within a certain time period, the predicted value corresponding to the trend within that time period can be used as the historical pattern based on the above trend function, and the deviation between the current pattern and the predicted value can be calculated to determine whether the target event has occurred abnormally within that time period.

[0046] It should be understood that when acquiring the historical patterns, the candidate historical patterns selected from the historical data are not limited to multiple candidate historical patterns corresponding to multiple consecutive time periods, as long as the selected candidate historical patterns are data in chronological order within the historical data. In some embodiments, multiple candidate historical patterns can be selected intermittently and discontinuously in chronological order, thereby avoiding the influence of anomalous data from individual sporadic behaviors on the final statistical results, making the historical patterns obtained based on historical data more reflective of the true characteristics of the target event. For example, video data from January 1st, January 4th, January 8th, January 9th, and January 15th to January 20th (a total of 10 days) can be selected as historical data. Based on this video data, the "pet cat eating" event can be identified each day, thereby counting the number of times the pet cat eats each day during these 10 days as candidate historical patterns, and calculating the average number of times the pet cat eats each day during these 10 days as the historical pattern of the pet cat eating event. Furthermore, selecting candidate historical patterns in chronological order is to ensure that the above fitting process follows the temporal relationship, making the fitting results conform to the true changing patterns of the target event.

[0047] In step S230, determining whether the target event is abnormal includes: determining that the target event is abnormal in response to the deviation values ​​of at least two current modes out of a predetermined number of current modes relative to the historical mode satisfying a first criterion. In some embodiments, the first criterion is: the absolute value of the deviation value of each of the at least two current modes relative to the historical mode is greater than a threshold. In other embodiments, the first criterion is: the deviation values ​​of the at least two current modes relative to the historical mode change monotonically, for example, gradually increasing. In some embodiments, the at least two current modes can be at least two sequential current modes. In this embodiment, the description will be based on the example of at least two sequential current modes. The following will refer to... Figure 3 , Figure 4 and Figure 5 The two embodiments described above will be explained in detail below.

[0048] First, refer to Figure 3 and Figure 4 The following is a specific description of an embodiment where the absolute value of the deviation of each of the plurality of sequential current patterns relative to the historical patterns is greater than a threshold. The aforementioned threshold is a standard used to determine whether an anomaly has occurred in the current pattern. It can be preset manually or set after analyzing historical data using an artificial intelligence model. If the absolute value of the deviation between the current pattern and the historical pattern, i.e., the absolute value of the difference between the two, is greater than the threshold, then the target event is determined to be abnormal.

[0049] Figure 4 This diagram illustrates an example of a first standard in the event anomaly identification method according to an embodiment of the present disclosure. In this embodiment, the current mode is set to the number of times a pet cat eats, and the time period is set to 1 day. Figure 4 In the diagram, the horizontal axis represents the date, and the vertical axis represents the total number of times the cat ate on the corresponding date. For example, the horizontal axis for February 2nd corresponds to a vertical axis of 1, indicating that the cat ate once on February 2nd. The dashed line K, for example, represents an average eating frequency of 3 times / day calculated from historical data of the cat's eating events, and uses this as the historical pattern for that particular feeding event. For instance, a threshold of 1 for the absolute value of the deviation can be predefined. In other words, as long as the difference between the current pattern and the historical pattern is within ±1 times / day, i.e., as long as the number of feeding behaviors is within the range of 2 to 4 times / day, the cat's eating behavior can be considered normal (as shown by dashed lines L1 and L2). If the absolute value of the deviation between the current pattern and the historical pattern, i.e., the absolute value of the difference, is greater than the threshold, then the current pattern of the cat's eating event is considered abnormal. Figure 4As shown, the current daily pattern of the pet cat's eating events from February 1st to February 6th is 5 times, 1 time, 1 time, 5 times, 5 times, and 1 time, respectively. The absolute value of the deviation from the historical pattern is 2 for each of the multiple sequential current patterns, which is greater than the preset threshold of 1. That is, the absolute value of the deviation of each of the multiple sequential current patterns from the historical pattern is greater than the threshold. In this case, it is determined that the pet cat's eating behavior is persistently abnormal, and the abnormality is reported to the user.

[0050] In some embodiments, personalized information related to the anomaly of the target event can be output in response to an anomaly occurring in the target event. For example, such as Figure 3 As shown, if the deviation between a pet cat's daily feeding frequency and its historical average daily feeding frequency consistently exceeds a threshold, personalized information related to this abnormality can be output, such as "The feeding frequency is consistently too high, which may be related to anxiety, sleep disturbances, and nutrient absorption efficiency." This prompts the user to pay attention to the pet's health and intervene promptly when necessary. It should be understood that while artificial intelligence models can analyze abnormal data and output the aforementioned personalized information, they are not limited to this. For example, connections can be established with professional medical personnel or relevant service platforms, allowing professionals to assess the target's condition based on abnormal data and provide corresponding guidance.

[0051] The above method helps to capture continuous deviations from the current pattern, avoiding inaccurate identification caused by a small amount of abnormal data. Using this method, if the absolute value of the deviation of each of the current patterns relative to the historical patterns is greater than a preset threshold, a persistent anomaly can be identified, and a corresponding anomaly warning can be output.

[0052] Next, refer to Figure 3 and Figure 5 The first criterion is specifically described in an embodiment where the deviation of the at least two current modes from the historical modes changes monotonically. Figure 5 This diagram illustrates another example of the first standard in the event anomaly identification method according to embodiments of the present disclosure. In this embodiment, the current mode is set to the number of times a pet cat grooms, and the time period is set to 1 hour. Figure 5 In the graph, the horizontal axis represents time, and the vertical axis represents the number of grooming behaviors performed by the cat within the corresponding time period. For example, the horizontal axis 11:00 corresponds to a vertical axis of 4, indicating that the cat groomed itself 4 times between 10:00 and 11:00. The dashed line P, for example, represents an average grooming frequency of 5 times per hour calculated from historical data of the cat's grooming events, and uses this as the historical pattern for that cat's grooming activities. If we follow... Figure 4The anomaly identification method in this embodiment can, for example, pre-define a threshold of 2 for the absolute value of the deviation. In other words, as long as the difference between the current pattern and the historical pattern is within ±2 times / day, that is, as long as the number of grooming behaviors is within the range of 3 to 7 times / hour, the grooming behavior of the pet cat can be considered as not abnormal (as shown by dashed lines Q1 and Q2). Figure 5 As shown, the pet cat groomed itself 3 times, 4 times, 5 times, 6 times, and 7 times respectively during the five time periods of 9:00-10:00, 10:00-11:00, 11:00-12:00, 12:00-13:00, and 13:00-14:00, respectively. The deviations from the historical pattern were -2, -1, 0, 1, and 2 respectively. In other words, if we follow... Figure 4 In the anomaly detection method of this embodiment, the absolute values ​​of the deviations of the aforementioned multiple current modes do not exceed a preset threshold of 2, thus no event anomaly is detected. However, the aforementioned deviation values ​​gradually increase over time, showing an upward trend over several consecutive time periods. At this time, although the absolute values ​​of the deviations in the pet cat's grooming behavior are not very large, this gradually increasing trend may not be an isolated fluctuation, indicating that the pet cat's skin condition may already have a potential abnormal risk. In this case, it is determined that the pet cat's grooming behavior has a trend of abnormality, and the anomaly is reported to the user. For example, as... Figure 3 As shown, in response to the above anomaly, personalized information related to the anomaly, such as "The number of grooming sessions continues to increase, and the pet's skin may be experiencing problems," can be output to the user's terminal device.

[0053] It should be understood that, in some embodiments, when performing the analysis of the first anomaly described above, the selected data is not limited to data from multiple consecutive time periods, but may also be data from multiple discontinuous sequential time periods.

[0054] The above implementation method avoids ignoring situations where the current pattern of the target event is still within the specified range but is gradually changing. By analyzing the changing trend of the target event over time, it helps to identify the anomaly before it develops significantly, and to take intervention measures when necessary to prevent the anomaly from escalating.

[0055] In some embodiments, determining whether the target event is abnormal may further include: determining that the target event has a second anomaly in response to a deviation value of a first current pattern relative to the historical pattern satisfying a second criterion. In some embodiments, the second criterion may be: the absolute value of the deviation value of the first current pattern relative to the historical pattern is greater than a threshold. As mentioned above, this threshold is a standard used to determine whether the current pattern is abnormal, which can be preset manually or set after analyzing historical data through an artificial intelligence model. If the absolute value of the deviation value between a current pattern and a historical pattern, i.e., the absolute value of the difference between the two, is greater than the threshold, then the current pattern is determined to have a second anomaly. In some embodiments, the threshold in the second criterion corresponding to the second anomaly is greater than the threshold in the first criterion corresponding to the first anomaly. This is because single data is more susceptible to the influence of sporadic behavior, environmental changes, or acquisition errors. If the threshold used in the second criterion is too low, it is easy to misjudge short-term fluctuations as anomalies in the target event, affecting the accuracy of anomaly identification. In contrast, the judgment of the first anomaly depends on the changes in data over multiple time periods, and its results have higher stability and reliability. Therefore, the first criterion can use a lower threshold to improve the sensitivity to continuous and trend changes in data. Furthermore, if an anomaly needs to be determined based on a single data point, the deviation of that single data point usually needs to be relatively large. This is because a small deviation in a single behavior may not reflect a substantial change in the target event and may not require special attention. In contrast, if the small deviation persists or gradually changes (e.g., gradually increases) over multiple time periods, it may reflect a substantial change in the target event and may therefore require special attention. (See below for further details.) Figure 3 and Figure 6 The following is a detailed description of an embodiment in which a second anomaly is determined to occur in response to the second criterion described above.

[0056] Figure 6 This diagram illustrates an example of a second standard in the event anomaly identification method according to an embodiment of this disclosure. In this embodiment, the current mode is set to the number of times a pet cat eats, and the time period is set to one day. Figure 6 In the diagram, the threshold for the absolute value of the deviation is set to 2. The meanings of the horizontal and vertical axes, and the dashed lines P, Q1, and Q2 are as follows: Figure 4 The same applies, so the explanation is omitted here. As shown by the "×" symbol in the image, the pet cat ate 8 times on March 3rd, with an absolute deviation of 5, which exceeds the preset threshold of 2. Therefore, the pet cat's eating behavior on March 3rd is determined to be abnormal. In this case, it is determined that the pet cat's eating behavior has experienced a single instance of abnormality, and this abnormality is reported to the user. Figure 3As shown, for example, personalized information related to the abnormality can be output to the user's terminal device, such as "The number of times the pet eats in a single day is too high. It is recommended to pay attention to the pet's emotional state and whether it is experiencing stress."

[0057] As described above, for persistent anomalies in a target event, a lower threshold can be set, and the occurrence of a first anomaly can be determined based on the deviation values ​​over multiple sequential time periods, or based on the trend of the deviation values ​​over multiple sequential time periods. For single anomalies in a target event, a higher threshold can be set, and a second anomaly can be determined if the deviation value at the target time exceeds the higher threshold. Therefore, target events can be monitored from different time dimensions. This allows for the capture of sudden, large-amplitude changes based on single data anomalies, and also for the identification of persistent or trend-based changes based on data information over multiple time periods, avoiding reliance on single data points while ignoring anomalies in long-term data trends. This improves the completeness and reliability of anomaly detection for target events and helps identify anomalies before they develop significantly, enabling intervention measures to be taken when necessary to prevent the escalation of the anomaly.

[0058] In some embodiments, the anomaly identification result of the target event can be output based on the first anomaly and the second anomaly. For example, the identification results of the first and second anomalies can be combined to comprehensively confirm the anomaly result of the target event, reducing the possibility of misjudgment and improving the reliability of anomaly identification. For example, if a pet cat is monitored to have groomed 12 times in a certain period of time, and the number of grooming times in several periods before and after that period is 3, 4, 3, 3, and 4 times respectively, remaining within the standard range (i.e., no continuous anomaly in the first anomaly) and without any deviation trend (i.e., no trend anomaly in the first anomaly), the possibility of early skin problems in the pet cat can be basically ruled out, and no anomaly warning will be triggered. In some embodiments, the standard for comprehensively confirming the anomaly result of the target event based on the first and second anomalies can be predetermined or determined by an artificial intelligence model based on data from multiple historical time periods.

[0059] According to another aspect of this disclosure, an event anomaly identification device 700 is also provided. Figure 7 A schematic block diagram of an event anomaly identification device 700 according to an embodiment of the present disclosure is shown.

[0060] like Figure 7As shown, the event anomaly identification device 700 of this embodiment includes a processor 710 and a memory 720. The memory 720 stores a computer program. The processor 710 is any processing device such as a microprocessor, which operates according to the program installed in the memory 720. The memory 720 is any volatile or non-volatile storage element, such as a hard disk, solid-state drive, ROM, or RAM. The program executed by the processor 710, etc., is stored in the memory 720. Figure 7 The event anomaly identification device 700 shown can be used to implement the event anomaly identification method according to embodiments of the present disclosure.

[0061] According to another aspect of this disclosure, a computer-readable recording medium is also provided, including a computer program that, when executed by a processor, enables the processor to implement an event anomaly identification method 200 according to embodiments of this disclosure.

[0062] According to another aspect of this disclosure, a computer program product is also provided, the computer program product including a computer program that, when executed by a processor, enables the processor to implement an event anomaly identification method 200 according to an embodiment of this disclosure.

[0063] The block diagrams of apparatuses, devices, methods, systems, etc., involved in this disclosure are merely exemplary and are not intended to require or imply that connections, arrangements, or configurations must be made in the manner shown in the block diagrams. As those skilled in the art will recognize, these circuits, devices, apparatuses, and systems can be connected, arranged, and configured in any manner that achieves the desired purpose.

[0064] In the foregoing description, the present invention has been described based on embodiments. These embodiments are merely illustrative, and those skilled in the art should understand that the combination of constituent elements and processes of these embodiments can be modified in various ways, and such modifications are also within the scope of the present invention.

Claims

1. A method for identifying event anomalies, comprising: Identify target events based on input data; Based on the identification results of the target event in multiple time periods that meet preset time association conditions, multiple current patterns of the target event are determined, and each current pattern corresponds to a time period; as well as Based on the deviation values ​​between the multiple current patterns and historical patterns, it is determined whether the target event has occurred abnormally. The determination of whether the target event is abnormal includes: in response to at least two of the predetermined number of current modes among the plurality of current modes satisfying a first criterion, determining that the target event is abnormal.

2. The event anomaly identification method as described in claim 1, wherein, The preset time association conditions are sequential.

3. The event anomaly identification method as described in claim 1, wherein, The first criterion is that the absolute value of the deviation of each of the at least two current patterns relative to the historical pattern is greater than a threshold.

4. The event anomaly identification method as described in claim 1, wherein, The first criterion is that the deviation of the at least two current patterns from the historical patterns changes monotonically.

5. The event anomaly identification method as described in claim 1, wherein, Determining whether the target event is abnormal also includes: In response to a second criterion being met by a deviation of the first current pattern among the plurality of current patterns relative to the historical patterns, the target event is determined to have a second anomaly.

6. The event anomaly identification method as described in claim 1, further comprising: Identify the target event based on historical data; as well as Based on the identification results of the target event in multiple historical time periods that meet preset time association conditions, the historical pattern of the target event is determined.

7. The event anomaly identification method as described in claim 1, wherein, The description information of the target event includes features in multiple dimensions, and the features in multiple dimensions include at least: Object characteristics; and One or more of the action characteristics and state characteristics.

8. The event anomaly identification method as described in claim 1, wherein, The input data includes one or more of the following: Image data acquired through image acquisition devices; and Sensing data acquired through sensing devices.

9. The event anomaly identification method as described in claim 1, wherein, The current or historical pattern of the target event includes one or more of the following: The time of occurrence of the target event; The duration of the target event; The number of times the target event occurs; The number of interruptions for the target event; The time interval between two consecutive target events; and The spatial distance between the event feature vectors of the two target events.

10. The event anomaly identification method as described in claim 6, wherein, Based on the identification results of the target event in multiple historical time periods that meet preset time association conditions, the historical pattern of the target event is determined as follows: Based on the identification results of the target event in multiple historical time periods that meet preset time association conditions, multiple candidate historical patterns of the target event in the multiple historical time periods are determined; and The historical pattern of the target event is determined based on the average, maximum, minimum, median, mode, or predicted value of the multiple candidate historical patterns.

11. The event anomaly identification method as described in claim 5, wherein, Based on the first anomaly and the second anomaly, the anomaly identification result of the target event is output.

12. The event anomaly identification method as described in claim 1, further comprising: In response to an anomaly occurring in the target event, personalized information related to the anomaly of the target event is output.

13. An event anomaly identification device, comprising: processor; as well as Memory, which stores computer programs. When the computer program is executed by the processor, the processor performs the event anomaly identification method as described in any one of claims 1-12.

14. A computer program product comprising a computer program that, when executed by a processor, performs the event anomaly identification method as described in any one of claims 1-12.

Images