Illegal behavior identification analysis processing method and system for training

Abstract

The invention provides an illegal behavior identification analysis processing method and system for training, and belongs to the technical field of image data analysis, and the method specifically comprises the steps: taking the number as a constraint condition, employing the association condition of the personnel with illegal behaviors in the video of each person as the basis, determining a secondary identification object in the personnel, and carrying out the recognition of the secondary identification object; according to the method, the video of the secondary recognition object is recognized to obtain the secondary recognition result of the violation behavior of each person, and the person for further recognition processing of the violation behavior is determined by using the secondary recognition result and the recognition result on the basis of the recognition data of the secondary recognition object, so that the recognition processing efficiency of the violation behavior is improved.

Description

Technical Field

[0001] This invention belongs to the field of image data analysis technology, and in particular relates to a method and system for identifying and analyzing violations in training. Background Technology

[0002] To address the technical problem of indiscriminate training in existing solutions, personalized training course recommendations can be made by analyzing the identification results of violations recorded by law enforcement officers' recorders, which can greatly improve the relevance of training.

[0003] However, due to the large number of videos recorded by law enforcement recorders, and the fact that the recorded law enforcement actions often involve multiple law enforcement officers, how to identify and process violations recorded by recorders during daily operations, improve the efficiency of violation identification and processing, and ensure the matching degree of training courses has become an urgent technical problem to be solved.

[0004] To address the aforementioned technical problems, this application provides a method and system for identifying, analyzing, and processing violations during training. Summary of the Invention

[0005] To achieve the objectives of this invention, the following technical solution is adopted: Specifically, this application provides a method for identifying, analyzing, and processing violations during training, which includes: S1 uses the personnel data corresponding to the recordings of each event in the recorder to determine the analysis objects among the personnel. Based on the identification results of the violations of each person and the corresponding personnel data of the recordings, when it is determined that there is no need to perform full recording analysis, the number of secondary identification objects is determined based on the identification results of the violations of the personnel. S2 uses the quantity as a constraint, and uses the correlation between the people with violations in the video recordings of each person to determine the secondary identification objects among the people. The video recordings of the secondary identification objects are processed to obtain the secondary identification results of the violations of each person. Based on the identification data of the secondary identification objects, the secondary identification results and the identification results are used to determine the people to be further identified and processed for violations.

[0006] The beneficial effects of this invention are as follows: By performing step-by-step identification and processing of video recordings for different personnel, it ensures that all law enforcement videos of some personnel can be comprehensively analyzed and processed to reliably identify violations. At the same time, by considering the correlation with other personnel, i.e., the appearance data of other personnel in the law enforcement videos, it also ensures the reliability of identifying and processing the violations of other personnel. This avoids the technical problem that the real-time performance and efficiency of the identification and processing cannot meet the requirements when there are a large number of videos of personnel that need to be analyzed and processed.

[0007] Based on the identification data of the secondary identification objects, the secondary identification results and the identification results are used to determine the personnel for further identification and processing of violations. This ensures that personnel with a high frequency of violations in the secondary identification results and the identification results are identified. By further identifying and processing violations, the reliability of violation identification and processing is improved. This avoids the technical problem of excessive pressure on the course recommendation server due to insufficient analysis and repeated course recommendations. At the same time, the number of secondary identification objects also enables the evaluation of the comprehensiveness of the identification and processing in this department. This allows for the determination of the number of personnel for further identification and processing of violations based on the differences in the comprehensiveness of the identification and processing, thus ensuring the comprehensiveness of the identification and processing.

[0008] Furthermore, the video recordings of each event are recordings corresponding to different law enforcement events.

[0009] Furthermore, the personnel data corresponding to the video recordings of each event is determined based on the law enforcement personnel data corresponding to each law enforcement event.

[0010] Furthermore, the method for determining the analysis subjects among the personnel is as follows: Based on the personnel data of the video subjects of each event, determine the law enforcement personnel data in the video recording of the law enforcement event corresponding to the personnel; Using the law enforcement personnel data, identify the law enforcement personnel appearing in the corresponding video footage; Based on the video data of law enforcement officers appearing in the video recordings corresponding to the individuals, it is determined whether the individuals are the subjects of analysis.

[0011] Furthermore, the method for determining the personnel responsible for further identifying and processing violations is as follows: Based on the identification data of the secondary identification object, the number of secondary identification objects is determined; Using the secondary identification results and the identification results, the number of violations committed by the person under different types of violations is determined; Based on the number of secondary identification objects, the number of violations committed by the person under different types of violations, and the parsing and processing data of the video recordings of the person, it is determined whether the person belongs to the group for further identification and processing of violations.

[0012] In a second aspect, the present invention provides a computer system comprising: a memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-described method for identifying and analyzing violations for training when running the computer program.

[0013] Other features and advantages will be set forth in the following description, and the objects and other advantages of the invention are realized and obtained through the structures particularly pointed out in the description and the drawings.

[0014] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0015] The above and other features and advantages of the present invention will become more apparent from a detailed description of exemplary embodiments thereof with reference to the accompanying drawings.

[0016] Figure 1 This is a flowchart of a method for identifying, analyzing, and processing violations used in training. Figure 2 This is a flowchart illustrating the method for identifying the analysis objects among personnel; Figure 3 This is a flowchart illustrating the method for determining the number of objects to be identified in a secondary manner; Figure 4 This is a flowchart illustrating the method for identifying secondary identification targets among personnel; Figure 5 This is a flowchart illustrating the method for identifying and processing personnel for further violations; Figure 6 It is a framework diagram of a computer system. Detailed Implementation

[0017] To enable those skilled in the art to better understand the technical solutions in this specification, the technical solutions in the embodiments of this specification will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this specification, and not all embodiments. Based on the embodiments of this specification, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of this specification.

[0018] This application identifies law enforcement officers for violation identification based on the presence of other law enforcement officers in the body camera footage. Violation identification is then performed on the video and audio data within the body camera. This approach ensures both comprehensive analysis of some law enforcement officers and efficient identification of their violations, laying the foundation for further course recommendation based on violation data.

[0019] Example 1 like Figure 1 As shown, this application provides a method for identifying, analyzing, and processing violations during training, specifically including: S1 uses the personnel data corresponding to the recordings of each event in the recorder to determine the analysis objects among the personnel. Based on the identification results of the violations of each person and the corresponding personnel data of the recordings, when it is determined that there is no need to perform full recording analysis, the number of secondary identification objects is determined based on the identification results of the violations of the personnel. Furthermore, the video recordings of each event are recordings corresponding to different law enforcement events.

[0020] Furthermore, the personnel data corresponding to the video recordings of each event is determined based on the law enforcement personnel data corresponding to each law enforcement event.

[0021] It should be noted that the analysis results of the video recording include the analysis results obtained by identifying violations in the audio and image data of the video recording. For example, the image data includes the use of police equipment and law enforcement actions, and the audio data includes uncivilized language and non-compliant language. The image data can be identified using a CNN image recognition model, and the audio data can be identified using a deep neural network.

[0022] The core decision-making objective of this method is to identify, by analyzing the co-occurrence relationships among law enforcement officers in video recordings, those officers who appear most frequently in the recordings and are most closely associated with other high-frequency responders, as the core analysis targets. This enables efficient collaborative analysis and supervision of the behavior of multiple potentially high-risk law enforcement officers. The core logic is to construct a law enforcement network based on personnel co-occurrence, identifying "high-frequency responders" involved in multiple cases as key nodes, and then finding the personnel directly associated with the largest number of key nodes as the analysis targets. This method concentrates limited analytical resources on the individuals with the greatest network influence, allowing the simultaneous assessment of the law enforcement compliance of multiple relevant personnel by analyzing the video recordings of one individual.

[0023] Specifically, such as Figure 2 As shown, the method for determining the analysis subjects among the personnel is as follows: Based on the personnel data of the video subjects of each event, determine the law enforcement personnel data in the video recording of the law enforcement event corresponding to the personnel; Keyword Explanation: Personnel Data of Video Recording Subjects: Refers to the collection of identity information of all law enforcement personnel present in each law enforcement video segment.

[0024] Video recordings of law enforcement incidents involving the personnel in question: This refers to the collection of all video recordings in the database that show the appearance of the target law enforcement officer.

[0025] Law enforcement personnel data: A list of other law enforcement personnel extracted from all video recordings featuring the target law enforcement officer (i.e., all colleagues who appeared in the same video recording as the target officer).

[0026] This step forms the data foundation of the entire methodology. It constructs a raw dataset of the target individual's "collaboration network" by aggregating all law enforcement incidents they have participated in. The significance of this aggregation method lies in its departure from viewing individual law enforcement incidents in isolation; instead, it builds a network view from the perspective of personnel collaboration relationships. In practice, law enforcement officers often partner with or collaborate with relatively fixed colleagues. This collaboration pattern may reflect deeper information such as shift assignments, complementary professional skills, or the transfer of work experience. Collecting this data makes it possible to subsequently analyze "who they frequently collaborate with."

[0027] example: Suppose the target law enforcement officer appears in four different law enforcement videos. In the first video, he appears with colleagues A and B; in the second video, he appears with colleagues A, C, and D; in the third video, he appears with colleagues B and E; and in the fourth video, he appears with colleagues C and F. Through this step, the system will integrate the "law enforcement officer data" set corresponding to the target officer: {colleague A, colleague B, colleague C, colleague D, colleague E, colleague F}. This set represents his direct collaboration network.

[0028] Using the law enforcement personnel data, identify the law enforcement personnel appearing in the corresponding video footage; Law enforcement officers appearing in the video recordings corresponding to the aforementioned personnel: This step is a direct presentation of the results of step S1, namely, the collection of other law enforcement officers extracted above.

[0029] This step, though seemingly simple, transforms the data from "event logs" to "personnel networks." Its significance lies in clarifying the object of analysis: we focus not on the events themselves, but on the connections between people established within those events. It provides a clear list of objects for subsequent quantitative analysis, enabling calculations to move from "who I've worked with" to "who I've worked with frequently." This is a prerequisite for conducting network centrality or influence analysis.

[0030] Example: Continuing from the previous example, the result of this step is to clearly list: in all the law enforcement videos in which the target person participated, they appeared together with colleagues A, B, C, D, E, and F. This list is a static, non-quantifiable list of associations.

[0031] Based on the video data of law enforcement officers appearing in the video recordings corresponding to the individuals, it is determined whether the individuals are the subjects of analysis.

[0032] It is understood that the analysis object is the person whose video recordings of law enforcement officers appearing in the corresponding video recordings meet the requirement the most.

[0033] In one possible embodiment, law enforcement officers who appear in two or more law enforcement cases in the video recordings of the personnel are considered as law enforcement officers whose number of law enforcement cases meets the requirement. The person with the largest number of video recordings in the video recordings of the personnel, i.e., the number of law enforcement cases that meets the requirement, is selected as the analysis object.

[0034] Keyword Explanation: Video Data of Law Enforcement Officers: This specifically refers to the video recording participation of the "other law enforcement officers" listed in step S2.

[0035] Video recording count requirement: For a specific "other law enforcement officer," count the number of times he / she appears in the target person's video recordings. When this number is greater than or equal to 2, the "other law enforcement officer" is considered to meet the requirement. This indicates that he / she has had multiple law enforcement experiences with the target person.

[0036] The target personnel with the largest number of eligible individuals is identified as follows: Calculate the total number of "other law enforcement personnel" who meet the above requirements for each target personnel. Compare this number with all candidate target personnel and select the target personnel with the largest number of eligible individuals as the analysis object.

[0037] This is the core decision-making step of this method. Setting "≥2 times" as the threshold for "meeting the requirements" is to filter out occasional, one-off collaborations and focus on stable, repetitive collaborative relationships. Such repetitive collaborations may mean partnerships, belonging to the same fixed team, or frequently performing similar tasks, thus potentially leading to mutual influence or convergence in law enforcement behavior patterns. Selecting the person with the "highest number of met-upon requirements" as the analysis object is significant in maximizing the leverage effect of analytical resources. This person acts as a network "hub," connecting the largest number of "active nodes" (high-frequency responders). Analyzing their video footage is equivalent to simultaneously observing the law enforcement performance of multiple active personnel, enabling the most efficient discovery of potential systemic problems or contagion chains of violation patterns.

[0038] example: Suppose there are three candidate individuals: A, B, and C. For A: In his video recording, colleague A appears 3 times, colleague B appears 2 times, and colleague C appears 1 time. Therefore, the individuals who meet the requirement (appearing ≥ 2 times) are A and B, totaling 2.

[0039] For person B: In his video recording, colleague A appeared twice, colleague D appeared twice, and colleague E appeared twice. The personnel who meet the requirements are A, D, and E, totaling 3.

[0040] For C: In his video recording, colleague B appeared once, colleague C appeared three times, and colleague F appeared once. The person who meets the requirements is C, and the number is 1.

[0041] According to the rules, B (number 3) is identified as the subject of analysis. Because analyzing B's video recordings allows for the simultaneous verification of the behavior of A, D, and E, three individuals with whom he has collaborated multiple times.

[0042] It should be noted that each law enforcement case corresponds to a single video recording, and the number of video recordings must meet the requirement to indicate that the number of law enforcement cases involving the personnel in question meets the requirement.

[0043] Furthermore, it was determined that full video recording analysis was unnecessary, specifically including: The core decision-making objective of this method is to intelligently determine whether a comprehensive analysis of all law enforcement personnel's video recordings is necessary by analyzing the distribution characteristics of identified violations within law enforcement personnel and their corresponding video recordings. The core logic is: if violations are highly concentrated in a few video recordings (i.e., multiple violators appear in the same recording), and the total number of violators involved is small, it indicates that the violation problem has "scene clustering," and only these key video recordings need to be analyzed, without initiating a costly full-scale analysis; conversely, if violations are scattered or involve a large number of personnel, a full-scale analysis is required to ensure comprehensive coverage.

[0044] Based on the identification results of the aforementioned violations, the law enforcement personnel who committed the violations are identified; Keyword explanation: The results of the identification of violations refer to the records of law enforcement violations that have been confirmed through preliminary screening, verification of reports, or partial random checks. Each record includes the identity of the law enforcement officer who violated the regulations, the type of violation, the time of occurrence, and the corresponding law enforcement video number.

[0045] Law enforcement officers with violations: The set of law enforcement officers with at least one violation record extracted based on the above identification results.

[0046] This step is the starting point for defining the scope of the problem. Using confirmed violation cases as input, the analysis shifts from an "event" dimension to a "personnel" dimension, clearly identifying the known groups of people at risk. Its significance lies in providing a clear set of objects requiring focused attention for subsequent analysis, enabling the analysis to start from known issues and explore unknown risks, rather than blindly searching for problems in massive amounts of data.

[0047] example: Suppose that a preliminary investigation confirms 5 instances of law enforcement violations involving the following law enforcement personnel (represented by their numbers): Incident 1 (Video V001): Law enforcement officer S01 violated procedures.

[0048] Incident 2 (Video V005): Law enforcement officer S02 made inappropriate remarks.

[0049] Incident 3 (Video V007): Law enforcement officers S01 (again) and S03 both mishandled evidence.

[0050] Incident 4 (Video V010): Law enforcement officer S04's recording was not in accordance with regulations.

[0051] Incident 5 (Video V012): Law enforcement officer S02 (again) committed a procedural violation.

[0052] Therefore, the set of law enforcement personnel identified as having committed violations is: {S01, S02, S03, S04}.

[0053] Based on the personnel data in the video footage corresponding to law enforcement officers who committed violations, video footage of multiple law enforcement officers who committed violations was identified; It should be noted that when the number of law enforcement officers involved in violations is large, exceeding the threshold, then it is necessary to perform full video analysis.

[0054] Keyword explanation: Video footage of law enforcement officers who committed violations: This refers to all law enforcement videos in which the aforementioned officers participated, including not only the video in which they committed the violation, but also other videos in which they participated but were not marked as violating the rules.

[0055] Videos featuring multiple law enforcement officers with violations: Select videos from all the above videos that simultaneously contain at least two law enforcement officers with violations.

[0056] This step aims to identify spatial (scene) clustering of violations. If multiple problematic individuals frequently appear in the same scene (same video), it may mean that: (1) the specific law enforcement scene (such as night inspection, handling of complex disputes) is inherently high-risk; (2) there are systemic management or cultural problems within the law enforcement team (team) in that scene; (3) there is mutual influence or learning of misconduct among the problematic individuals. Identifying these "high-risk scene videos" is key to determining whether targeted analysis or full analysis can be conducted.

[0057] Example: Continuing from the previous example, list the video recordings (hypothetical data) of each law enforcement officer involved in the issue: Videos involving S01: V001, V003, V007, V009; Videos involving S02: V005, V007, V012, V015; Videos involving S03: V007, V011, V014; Videos involving S04: V010, V013; Filter the videos that contain at least two problematic individuals: V007: contains S01, S02, and S03 → meets the condition. Other videos, such as V001, V005, and V010, contain only one problematic individual.

[0058] Therefore, the video collection containing multiple videos of law enforcement officers who have committed violations is: {V007}.

[0059] By using video recordings of multiple law enforcement officers with violations, as well as the corresponding video recordings of those officers, it can be determined whether full video analysis is necessary.

[0060] First-level judgment: Based on the number of violators, keyword explanation: Number threshold: An absolute number or relative proportion threshold set based on team size and problem tolerance (e.g., more than 10% of the total number of people who violate the rules, or more than 5 people).

[0061] Judgment rule: When the number of law enforcement officers with violations exceeds the threshold, it is determined that full video recording analysis is required.

[0062] If the number of violators is already large, it indicates that the problem is likely not isolated but rather has become widespread. In this case, analyzing only the clustered video footage is likely to miss violations occurring in other scenarios or when collaborating with other non-violators. Initiating a full-scale analysis is necessary to comprehensively assess the severity of the problem, identify all potential risk points, and prevent any violations from slipping through the cracks.

[0063] example: Assume the total number of law enforcement personnel is 50, and the threshold for the number of violations is set at 5 or 10%. Continuing the previous example, the number of violators is 4 (S01, S02, S03, S04). 4 < 5, and 4 / 50 = 8% < 10%. Therefore, the first-level judgment does not trigger a full analysis, and the second-level judgment is initiated.

[0064] In one possible specific embodiment, the video recording of multiple law enforcement officers who have committed violations is a video recording of at least two law enforcement officers who have committed violations.

[0065] Furthermore, if the number of law enforcement officers with violations is not large, and the proportion of videos of multiple law enforcement officers with violations in the videos of different law enforcement officers with violations is high (i.e., exceeding the threshold), then the overlap of videos of different law enforcement officers with violations is high, and therefore there is no need to perform full video analysis.

[0066] Keyword explanation: Videos featuring multiple law enforcement officers who have committed violations: i.e., the set of high-risk scene videos selected in step S2 (such as {V007}).

[0067] Video recordings of law enforcement officers involved in violations: This refers to the union of all video recordings of each individual involved in the violation.

[0068] Proportion: For each law enforcement officer involved in a problem, calculate the proportion of high-risk scene recordings to the total number of recordings they participated in.

[0069] The proportions are all too high: This means that the above proportions exceed a set threshold (e.g., 50%) for most (e.g., exceeding a certain proportion, such as 80%) or all law enforcement officers with problems.

[0070] Judgment rule: If the number of law enforcement officers with violations is not large (first layer not triggered), but when the proportion of videos of multiple law enforcement officers with violations in the videos of different law enforcement officers with violations is high (i.e., greater than the threshold), then the videos of different law enforcement officers with violations have a high degree of overlap, so there is no need to perform full video analysis.

[0071] The core of this judgment is assessing the "concentration" or "coupling" of the problem. If most of the law enforcement activities (videos) of each problematic individual are concentrated in a few high-risk scenarios (i.e., they frequently work together in these scenarios), and the total number of these problematic individuals is not large, then the root cause of the problem is likely limited to these specific scenarios or teams. In this case, a thorough analysis of these highly overlapping "core risk videos" is sufficient to effectively locate and resolve the problem, without the need for a large-scale full-scale analysis. This greatly improves resource utilization efficiency.

[0072] Example (data continues and expands): Assuming the extended data is as follows, calculate the percentage of "high-risk scene recordings" for each problematic individual: High-risk scene recording set (from S2): {V007}, recording participation and percentage for each problematic individual: ; S01: Participated in recording {V001, V003, V007, V009}, a total of 4 segments. Among them, V007 is a high-risk scenario. Percentage = 1 / 4 = 25%.

[0073] S02: Participated in recording {V005, V007, V012, V015}, a total of 4 segments. Among them, V007 is a high-risk scenario. Percentage = 1 / 4 = 25%.

[0074] S03: Participated in recording {V007, V011, V014}, a total of 3 segments. Among them, V007 is a high-risk scenario. Percentage = 1 / 3 ≈ 33%.

[0075] S04: Participated in recording {V010, V013}, a total of 2 segments. V007 is not part of his recording. Percentage = 0 / 2 = 0%.

[0076] Suppose the judgment threshold is set to "percentage > 50%" and "most (e.g., > 75%) of the problem personnel meet the requirements".

[0077] In this example: the percentage of uninhabited individuals is greater than 50% (maximum 33%), and the percentage of problematic individuals meeting the criteria is 0% (0 / 4). Therefore, the condition of "all percentages being too high" does not hold.

[0078] Decision: Since the first layer (few individuals) has passed, but the second layer (high overlap) has not, the system determines that a full analysis cannot be waived. Because the violations of the problematic individuals (S01-S04) are not highly concentrated in the same batch of recordings (only V007 is common, and its proportion is very low), their violations may have occurred in different independent scenarios, posing a risk of dispersion. Therefore, a full analysis is still necessary for a comprehensive investigation.

[0079] Suppose another scenario (satisfying the second-level condition): If the data changes to: the video recordings of most problematic individuals (such as S01, S02, S03) highly overlap, appearing together in multiple videos such as {V007, V008, V009}, and these high-risk videos account for more than 60% of each individual's total video count, while the total number of problematic individuals remains 4 (not exceeding the threshold), then the second condition is met.

[0080] Decision: No need to perform full video analysis and processing.

[0081] Specifically, such as Figure 3 As shown, the method for determining the number of secondary recognition objects is as follows: Based on the initial identification of violations, this method scientifically determines the number of law enforcement personnel required for a second round of in-depth identification (secondary identification) by analyzing the reliability (based on personnel-video matching) and scale of the identified violations. The core logic is: when the number of initially identified violations is small, these personnel are directly subjected to secondary identification; when the number of initially identified violations is large or the identification reliability is low, the scope of secondary identification needs to be expanded to include more suspicious individuals to ensure comprehensiveness and accuracy. This method aims to balance identification accuracy with review costs, avoiding the risks of over-review or under-review.

[0082] Based on the identification results of the personnel's violations, determine the number of personnel who have committed violations; The results of violation identification refer to records that are suspected or confirmed to be violations, which are initially identified through the first round of automatic analysis or manual spot checks.

[0083] Number of personnel involved in violations: The total number of different law enforcement personnel identified from the above identification results.

[0084] This step quantifies the initial scale of the identified problems. Aggregating violations to the personnel dimension allows for an assessment of the problem's prevalence. If only a few people are involved, it may be an isolated incident; if many are involved, it may indicate a systemic problem. This is the primary basis for determining the scale of subsequent resource allocation.

[0085] example: Assuming the first round identifies 15 violations, and these involve 8 different law enforcement officers, then the total number of individuals involved in the violations is 8.

[0086] By analyzing the video recordings of individuals, the identification and matching of their violations can be determined. Personnel-specific video recordings: This refers to the collection of all law enforcement videos that the law enforcement officer has participated in.

[0087] Video recordings that have undergone analysis: Video recordings of the person who have been thoroughly reviewed (analyzed) by the system or by a human during the first round of identification in order to detect violations.

[0088] Identifying matching results: For a specific individual, calculate the ratio of "the number of videos they have analyzed" to "the total number of videos they participated in". That is, matching degree = number of analyzed videos / total number of videos participated in. This ratio reflects the adequacy of the first round of identification in covering the individual's law enforcement activities.

[0089] This step assesses the depth and reliability of the first round of identification. If most of a person's video footage has been analyzed, the first round's conclusions regarding their violations are relatively reliable (comprehensive coverage); conversely, if only a small portion of their video footage has been analyzed, many violations may remain undetected, and the first round's conclusions may only represent the "tip of the iceberg." Match accuracy is a crucial indicator for determining whether a second, deeper identification of the individual is necessary.

[0090] example: For law enforcement officer A: He participated in a total of 20 law enforcement video recordings. In the first round of identification, the system analyzed 5 of these recordings in detail and discovered violations. Therefore, A's identification matching score (match degree) = 5 / 20 = 0.25.

[0091] The number of individuals with violations and the matching results are used to determine the number of secondary identification targets.

[0092] It is understandable that when the number of people with violations does not meet the requirements, for example, when the number of people with violations is more than 3, in order to improve the reliability of the identification and processing of violations, a preset number is used to determine the number of secondary identification objects.

[0093] The problem is quite large, so we'll expand the scope directly: Judgment criteria: When the number of people with violations does not meet the requirements, for example, when the number of people with violations is more than 3.

[0094] Decision: The number of objects to be identified in the secondary identification is determined by using a preset number.

[0095] Setting a low threshold (e.g., 3 people) allows for a rapid response to the scale of the problem. If more than 3 individuals are identified in the initial screening, it indicates that the violation may be widespread, and the initial screening results should not be relied upon solely. To prevent omissions due to sampling bias or imperfect rules in the first round of screening, it is necessary to proactively expand the scope of the secondary screening, i.e., increase the number of individuals subject to secondary screening (a preset number), to ensure a more comprehensive discovery of problems. This reflects the prudent principle of "better to be strict than lenient when problems first emerge."

[0096] Additionally, it should be noted that if the number of individuals with violations meets the requirements, then it is necessary to further determine the identification and matching of the individuals' violations. If the identification and matching of violations does not meet the requirements, then a preset number is used to determine the number of secondary identification objects. In other cases, the number of individuals with violations is used as the number of secondary identification objects. The identification and matching of individuals' violations is determined based on the proportion of the number of video recordings of the individuals being analyzed in the analysis objects to the number of video recordings of the individuals corresponding to those individuals.

[0097] The problem is small in scale but insufficient in identification; the scope needs to be expanded. Judgment criteria: If the number of people with violations meets the requirements (e.g., ≤3 people), but the identification and matching of violations does not meet the requirements.

[0098] Regarding the judgment of identification matching: When the average value of identification matching of violations by different personnel is less than 0.3, it is determined that the identification matching of violations does not meet the requirements.

[0099] Decision: The number of objects to be identified in the secondary identification is determined by using a preset number.

[0100] Even if only one or two individuals are found to have violated regulations, if the first round of identification has a low coverage of these individuals (average match rate <0.3), it means that the conclusions drawn about them are highly unreliable. They may have other undiscovered violations, or the current findings may be merely coincidental. In this case, vigilance should not be relaxed simply because the number of individuals is small. A second round of identification needs to be expanded to include more individuals related to them or with similar characteristics to ensure the reliability of the conclusions. A match rate threshold of 0.3 means that if the first round of review covers less than 30% of a person's law enforcement record, the review is considered insufficient.

[0101] In one possible embodiment, if the average number of matching results for the identification of violations by different individuals is less than 0.3, then it is determined that the identification of violations does not meet the requirements.

[0102] Small question size and sufficient identification, accurate review: Judgment criteria: The number of people with violations meets the requirements (e.g., ≤3 people) and the identification and matching meet the requirements (average matching degree ≥0.3).

[0103] Decision: The number of individuals with violations will be used as the number of secondary identification targets.

[0104] In this scenario, the problem is concentrated on a very small number of individuals, and the first round of identification has already provided a relatively thorough review of them (covering more than 30% of their law enforcement records). At this point, the secondary identification can precisely target these few individuals identified as high-risk, allowing for a more in-depth and detailed review without expanding the scope. This achieves optimal resource allocation, ensuring that resources are used effectively.

[0105] It should be noted that the preset quantity is determined by the sum of the number of personnel with violations and the preset ratio excluding the number of personnel with violations and the analysis subjects. The preset ratio is determined based on the average value of the identification and matching of violations of different personnel. The smaller the average value of the identification and matching of violations of different personnel, the larger the preset ratio is. In one possible embodiment, its value ranges from 0.1 to 0.2.

[0106] The method for calculating the preset quantity: formula: Preset quantity = Number of personnel with violations + Round (Preset ratio × Number of remaining personnel available for review) Variable Explanation: Number of personnel with violations: The number of personnel identified in the first round of violations. Remaining number of personnel available for review: Total number of law enforcement personnel - Number of personnel with violations - Analytical subjects (subtract if the analytical subjects are a specific group different from the violators; do not subtract if the analytical subjects are the violators themselves).

[0107] Preset ratio: A dynamic value determined based on the average of the identification and matching results of violations by different individuals. The smaller the average value, the larger the preset ratio, with a range of 0.1 to 0.2.

[0108] Calculation logic: Basic component (number of personnel with violations): Ensure that all identified problematic personnel are re-identified.

[0109] The expanded portion (Round(preset ratio × remaining number of personnel available for review)): A portion of the remaining personnel is proportionally selected to expand the scope of review. The ratio is dynamically adjusted based on the reliability of the first round of identification: the lower the reliability (the smaller the average matching degree), the higher the proportion of expanded review.

[0110] example: Assume the enforcement team has 50 members. In the first round, 3 individuals (A, B, C) were identified as violators. Their identification match scores were 0.2, 0.25, and 0.15, respectively, with an average score of (0.2 + 0.25 + 0.15) / 3 = 0.2.

[0111] "Analysis objects" specifically refers to the number of analysis objects determined earlier.

[0112] Calculate the remaining number of personnel who can be reviewed = 50 - 3 (violation personnel) - 5 (analysis subjects) = 42 people. Determine the preset ratio: Since the average matching degree of 0.2 is low (<0.3) and the reliability is poor, a higher ratio is adopted, such as 0.18.

[0113] The preset number is calculated as follows: 3 + Round(0.18 × 42) = 3 + Round(7.56) = 3 + 8 = 11 people.

[0114] This means that, in addition to the initial three violators, the second round of identification will involve randomly selecting eight more individuals from the remaining 42 ordinary personnel, for a total of 11 people undergoing in-depth review. The second preset number is 0.7 times the preset number.

[0115] S2 uses the quantity as a constraint, and uses the correlation between the people with violations in the video recordings of each person to determine the secondary identification objects among the people. The video recordings of the secondary identification objects are processed to obtain the secondary identification results of the violations of each person. Based on the identification data of the secondary identification objects, the secondary identification results and the identification results are used to determine the people to be further identified and processed for violations.

[0116] The core decision-making objective of this method is to, given that some individuals have engaged in misconduct and some video recordings have already been analyzed, precisely select a group of individuals from the remaining pool as secondary identification targets through a complex combinatorial optimization algorithm. This maximizes the analysis coverage of new video recordings (those not covered by the initial analysis). The core logic is as follows: First, exclude those individuals who cannot be adequately covered by analyzing existing video recordings (records of other individuals) (they must be secondary identified); then, while satisfying the total number of personnel constraints, freely combine the remaining candidates; finally, evaluate the coverage capability of each combination for "unanalyzed video recordings," and select the combination with the strongest coverage capability as the final secondary identification target. This logic aims to achieve "covering the most unreviewed video recordings with the fewest personnel," thereby efficiently improving the comprehensiveness of misconduct identification.

[0117] Suppose we have the following basic data: The entire group is represented by: P = {P1, P2, P3, P4, P5, P6, P7, P8}; Initial recognition results: Personnel who violated regulations: V = {P1, P2, P3}; Analysis object ( Figure 2 (Identified core personnel): A = {P4}; Video participation matrix: Each person participated in a series of video recordings.

[0118] Parsed Videos: In the initial analysis, we have parsed all the video recordings of the analysis object P4, as well as the parts of the video recordings of the violators P1, P2, and P3 that appeared in the video recordings of the analysis object P4.

[0119] Objective: To identify the group of individuals who require further in-depth identification.

[0120] Specifically, such as Figure 4 As shown, the method for determining the secondary identification objects among the personnel is as follows: The number of videos showing violations among other personnel and the personnel whose violations do not meet the requirements are used as secondary violation identification targets. Personnel with violations other than the secondary violation identification targets are used as violation analysis personnel. Keyword explanation: Other personnel: refers to all other personnel besides the person who violated the rules.

[0121] The sum of the number of videos appearing among the other individuals: For a violator Pi, calculate the union of the sets of videos of all other individuals. Then count how many of Pi's videos are included in this "union of videos of the other individuals".

[0122] If the proportion does not meet the requirements: Define a threshold, for example, proportion < 60%. That is, if less than 60% of the videos of violator Pi can be found in other people's video libraries, it is considered that Pi cannot be fully reviewed by analyzing other people's videos alone.

[0123] Judgment and Classification: For each violator Pi ∈ V: Collect the set of video IDs of all persons except Pi, denoted as Other_Videos(Pi). Obtain the set of video IDs of Pi himself, denoted as Self_Videos(Pi).

[0124] Calculate coverage: Coverage_Ratio(Pi) = |Self_Videos(Pi) ∩ Other_Videos(Pi)| / |Self_Videos(Pi)|.

[0125] If Coverage_Ratio(Pi) < 0.6, then Pi is marked as a secondary review violator (Must-ReviewViolator).

[0126] If Coverage_Ratio(Pi) >= 0.6, then Pi is marked as an AnalyzableViolator.

[0127] This step achieves separation of responsibility. For violators who act alone or whose video tracks are unique (with little overlap with others), it is difficult to obtain a comprehensive law enforcement record by analyzing the videos of others; therefore, they must be included in secondary identification without any other options. However, for violators whose videos are highly integrated into a group (with much overlap with others), theoretically, they can be indirectly investigated through in-depth analysis of the videos of others; therefore, they do not necessarily need to be directly subject to secondary identification, thus freeing up slots for subsequent optimization selection.

[0128] Calculate the percentage of video recordings of the three individuals in V that are covered by others: P1: 10 self-owned video clips. Only 2 clips {V1, V2} appear in other people's (P2-P8) video clips. Proportion = 2 / 10 = 20% < 60%. Therefore, P1 ∈ Secondary identification of the violation target (M).

[0129] P2: 8 self-made video clips. Of these, 5 clips {V3, V4, V5, V6, V7} appear in other people's video clips. Percentage = 5 / 8 = 62.5% ≥ 60%.

[0130] P3: 5 self-made video clips. Among them, 3 clips {V8, V9, V10} appear in other people's video clips. Percentage = 3 / 5 = 60% ≥ 60%.

[0131] Therefore, P2, P3 ∈ Violation Analyst (V_a).

[0132] Results: M = {P1}; V_a = {P2, P3}.

[0133] Results: The set of objects identified in the secondary identification (M): {P1}, and the set of personnel analyzed for violations (V_analyzable): {P2, P3}.

[0134] The goal is to analyze and process all videos except those of the identified object, by different violation analysts. The constraint is the number of secondary identification objects minus the number of secondary violation identification objects. The personnel other than the identified object and the secondary violation identification objects are freely combined, and multiple schemes are obtained by combining the secondary violation identification objects. Keyword explanation: Objective: To ensure that "all videos of different violator analysts, excluding those appearing in the identified target's videos, contain parsed and processed footage." In other words, the goal is to achieve the most comprehensive coverage of unparsed videos of violator analysts (V_a).

[0135] Constraint Quantity (K): The number of people available for free combination. K = N - |M|. N is the total number of secondary identification objects, and |M| is the number of secondary identification violation objects that must be included.

[0136] Candidate pool: This refers to "personnel excluding the secondary identification of violators and those not identified." Mathematically, it is represented as: P ∪ A (M ∪ A).

[0137] Scheme: A complete list of secondary identification targets. It consists of M, which must be included, and a K-person combination (Combo) freely selected from the candidate pool.

[0138] Focusing on the core contradiction: The target is "unanalyzed video recordings covering unauthorized analysts," which clarifies that the primary task of secondary identification is to clarify known doubts (V_a), rather than blindly expanding the scope of the review.

[0139] Structured search space: By fixing M and transforming the remaining choice problem into selecting K people from the candidate pool, we generate all possible candidate solutions (combinations). This lays the foundation for subsequent quantitative calculations to find the "optimal solution".

[0140] Example (following the previous step): The total number of slots is N = 4, and the number of people to be reviewed is |M| = 1 (P1).

[0141] The number of freely chosen spots is K = 4 - 1 = 3.

[0142] Candidate pool = P \ (M ∪ A) = {P1, P2, P3, P4, P5, P6, P7, P8} \ ({P1} ∪{P4}) = {P2, P3, P5, P6, P7, P8}.

[0143] We need to select 3 people from the candidate pool {P2, P3, P5, P6, P7, P8} to form a scheme with P1.

[0144] Four exemplary candidate solutions are generated (actual C(6,3)=20): S1 = M ∪ {P2, P3, P5} = {P1, P2, P3, P5}; S2 = M ∪ {P2, P6, P7} = {P1, P2, P6, P7}; S3 = M ∪ {P3, P5, P8} = {P1, P3, P5, P8}; S4 = M ∪ {P5, P6, P7} = {P1, P5, P6, P7}.

[0145] Based on the number of videos of different personnel in the scheme that are analyzed and processed, excluding the videos of the identified target and the secondary identified violators, the preferred scheme is determined, and the personnel corresponding to the preferred scheme are used as the secondary identification targets.

[0146] It is understandable that when the sum of the number of videos of the person who violated the rules appearing among other people does not meet the requirements, in a possible specific embodiment, that is, when the proportion of the sum of the number of videos of the person who violated the rules appearing among other people in the video of the person who violated the rules is less than 60%, the video of the other people cannot fully realize the identification and processing of the violation. Therefore, the person who violated the rules is determined to be a secondary identification object.

[0147] Additionally, it should be noted that the proposed scheme is based on the personnel other than the secondary identification of the violation object and the identification object, to obtain multiple personnel combinations. These personnel combinations, together with the secondary identification of the violation object, construct personnel in different schemes. In the video recordings of the personnel in the scheme, there are videos of the violation analysis personnel excluding those appearing outside the identification object. The number of people in the personnel combination is the constraint number.

[0148] Furthermore, based on the number of video recordings processed by different personnel excluding those appearing in the video recordings of the identified object, a preferred scheme is determined, specifically including: The video recordings excluding those appearing in the identified object are considered unparsed videos. Based on the number of unparsed videos in the scheme that are parsed by different violation analysts and by personnel other than the violation analysts, the analyzed object, and the secondary identified violation object, the preferred scheme is determined.

[0149] In one possible embodiment, the preferred scheme is determined based on the number of unparsed video recordings of different violation analysts and persons other than the violation analysts and the analysis targets in the scheme, specifically including: The preferred scheme is the one that maximizes the sum of the weights of the number of unparsed video recordings of the violation analyst in the scheme that are parsed and processed, and the sum of the weights of the number of unparsed video recordings of the violation analyst, the analysis object, and the secondary identification violation object in the scheme that are parsed and processed.

[0150] In one possible embodiment, the weight of the number of unparsed video recordings of the violation analyst in the scheme that are parsed is determined by the sum of the products of the number of unparsed video recordings of different violation analysts in the scheme that are parsed and a preset weight value. In one possible embodiment, the preset weight value of the violation analyst is 0.3, and the preset weight value of personnel other than the violation analyst, the analysis object, and the secondary identification of violation objects is 0.1.

[0151] Keyword explanation: Unresolved video: For any person Pi, it refers to "videos excluding those appearing in the video of the identified object", that is, the set of videos of Pi and the set of videos of A.

[0152] Number of parsed processes: For the violating analyst (V_a): In a given scenario S, the number of times the analyst's "unparsed video" is overwritten by the video recordings of all other personnel in S except for the analyst himself.

[0153] For other personnel (i.e., those from the candidate pool in the scheme who do not belong to V_a): refers to the number of "unparsed recordings" for that person (because these recordings are parsed when they are directly reviewed).

[0154] Weighted sum: The default weight for the violation analyst (V_a) is 0.3.

[0155] The preset weight for "personnel other than the aforementioned violation analysts, analysis objects, and secondary identification violation objects" (i.e., non-V_a members in the candidate pool) is 0.1.

[0156] Solution value: Value(S) = 0.3 * Σ (number of unresolved recordings for each V_a member covered by the solution) + 0.1 * Σ (number of unresolved recordings for non-V_a members in the solution).

[0157] Preferred solution: The solution that maximizes Value(S).

[0158] Quantitative evaluation and fair comparison: Calculate a comprehensive score for each option so that all options can be compared under the same standard.

[0159] The weighting reflects priority: assigning a higher weight (0.3 > 0.1) to the coverage of the violation analyst (V_a) clearly indicates that the optimization direction of the system is to prioritize ensuring the verification coverage of known suspicious points, and secondarily to acquire more new video information. This ensures strategic focus in decision-making.

[0160] Example calculation (using analog video data): 1. Define the video recording sets and unparsed video recordings for each person; The video set of object P4 is analyzed: {V_a, V_b, V_c} (used as a baseline to calculate the "unresolved" portion). Each person's video recording and unanalyzed video recording (self-recording / P4 recording): P1 (M): {V_a, V2, V3} -> Unresolved: {V_a,V2, V3} (V_a is overwritten by P4); P2 (V_a): {V_a, V4, V5} -> Unresolved: {V4, V5} (V_a is overwritten by P4); P3 (V_a): {V_b, V6} -> Unresolved: {V6} (V_b is overwritten by P4); P5: {V3, V4, V7} -> Unresolved: {V3, V4, V7} (assuming no intersection with P4); P6: {V5, V6, V8} -> Unresolved: {V5, V6, V8}; P7: {V7, V9} -> Unresolved: {V7, V9}; P8: {V8, V10} -> Unresolved: {V8, V10}; 2. Calculate the value of the proposed solution; Solution S1 = {P1, P2, P3, P5}; Covering V_a: Is the unresolved {V4, V5} in P2 covered by {P1, P3, P5} in S1? P5 has V4, but P1 / P3 do not have V5. Yes, it covers V4. Quantity = 1.

[0161] Is the unresolved {V6} in P3 covered by {P1, P2, P5} in S1? No one has V6. Quantity = 0.

[0162] The weighted sum of V_a is 0.3 * (1 + 0) = 0.3; Coverage of other personnel: In S1, the non-V_a member is {P5}.

[0163] The number of unparsed recordings in P5 = |{V3, V4, V7}| = 3; The weighted average of other personnel is 0.1 * 3 = 0.3; Total value of S1 = 0.3 + 0.3 = 0.6; Solution S2 = {P1, P2, P6, P7}; Covering V_a: The unresolved {V4, V5} of P2 is covered by {P1, P6, P7}, and P6 has V5. V5 is covered. Quantity = 1.

[0164] The unresolved {V6} of P3 is covered by {P1, P2, P6, P7}, and P6 has V6. Quantity = 1.

[0165] The sum of the weights of V_a is 0.3 * (1 + 1) = 0.6; Coverage of other personnel: In S2, non-V_a members are {P6, P7}.

[0166] Unresolved numbers in P6 = 3, unresolved numbers in P7 = 2. Total = 5.

[0167] The weighted average of other personnel is 0.1 * 5 = 0.5; Total value of S2 = 0.6 + 0.5 = 1.1; 3. Determine the optimal solution; Comparing the values ​​of each option: S2 (1.1) > S1 (0.6), the preferred option is S2, which has the highest value.

[0168] The final output, based on the above calculations, is S2 = {P1, P2, P6, P7}.

[0169] Therefore, the secondary identification targets identified this time are: Officer 01 (P1), Officer 02 (P2), Officer 06 (P6), and Officer 07 (P7).

[0170] This result means that, given the necessity of reviewing P1, simultaneously reviewing P2, P6, and P7 is the most effective way (at a quantitative score of 1.1) to cover the unanalyzed law enforcement footage of violators P2 and P3, while also reviewing a significant number of new video clips.

[0171] Example 2 Secondly, such as Figure 6As shown, the present invention provides a computer system comprising: a memory and a processor connected in communication, and a computer program stored in the memory and capable of running on the processor, wherein the processor executes the above-described method for identifying and analyzing violations for training when running the computer program.

[0172] Specifically, such as Figure 5 As shown, the method for determining the personnel responsible for further identifying and processing violations is as follows: Scenario: Following the "secondary identification" stage, a deeper audit of the entire department's law enforcement behavior is needed to determine which personnel need to be included in the final in-depth analysis list covering all their law enforcement videos.

[0173] Input data: Secondary identification object: from the previous stage ( Figure 4 The process identifies the set of personnel requiring in-depth verification. Let's assume it's {P1, P2, P6, P7}.

[0174] The number of objects identified in the secondary identification: that is, the number of people in the above set, denoted as N_s = 4.

[0175] Violation Record Database: Contains all types of violations and their frequency identified by all personnel in all preliminary analysis stages (initial analysis, secondary identification).

[0176] Video recording status: Records the percentage of each person's video recording that has been processed.

[0177] Department of Personnel: This embodiment focuses on a specific department, whose entire personnel set is D = {P1, P2, P3, P4, P5, P6, P7, P8}, totaling 8 people. All secondary identification targets are from this department.

[0178] Based on the identification data of the secondary identification object, the number of secondary identification objects is determined; Using the secondary identification results and the identification results, the number of violations committed by the person under different types of violations is determined; Based on the number of secondary identification objects, the number of violations committed by the person under different types of violations, and the parsing and processing data of the video recordings of the person, it is determined whether the person belongs to the group for further identification and processing of violations.

[0179] Optionally, when the number of violations by the personnel does not meet the requirement, in a specific embodiment, that is, when there are more than 3 violations, in order to avoid the technical problem of excessive difficulty in recommending training courses for the personnel due to incomplete identification, the personnel are then regarded as personnel who need further identification and processing of violations, thereby realizing comprehensive analysis and processing of the video recordings corresponding to the personnel.

[0180] Identify "Further Identification Personnel" (based on frequency of serious violations): Keyword Explanation: Violation frequency does not meet requirements: For a specific type of violation, the number of times the person has committed the violation exceeds the preset severity threshold (e.g., 3 times).

[0181] Further Identification Personnel (F): These are individuals who are not subject to secondary identification but have accumulated more than three violations of a single offense. These personnel pose an extremely high risk due to repeatedly committing the same serious error and must undergo a comprehensive review and thorough assessment.

[0182] Risk Focus: This step targets high-frequency, recurring violations. More than three violations of a single item indicate a potential systemic habitual problem or a lack of awareness. This cannot be eradicated by piecemeal reviews alone; a comprehensive review of past violations is necessary to prevent repeated training sessions triggered by the same issue in the future, which would increase management costs (i.e., "recommended handling is too difficult").

[0183] Example (data simulation): Suppose that from the violation database, we find the following individuals have committed more than 3 violations of a single type: P3: There were a total of 4 violations in the "processing error" category.

[0184] P5: There were a total of 5 violations in the category of "improper use of equipment".

[0185] Therefore, the set of personnel to be further identified is determined as F = {P3, P5}.

[0186] (Note: P1, P2, P6, and P7 are secondary identification objects. Their videos have been or will be analyzed in depth during the secondary identification process. They may also meet the criteria in this step, but they have been prioritized for overwriting.) Furthermore, individuals whose number of violations does not meet the requirement are considered as further identification personnel. When the proportion of these further identification personnel and secondary identification objects in the total number of personnel meets the requirement, which is greater than 70% in a possible specific embodiment, the comprehensiveness of the identification of the department corresponding to the personnel is already sufficient. Therefore, no further identification processing is required for personnel whose number of violations does not meet the requirement, thereby ensuring the efficiency of violation identification processing when there are a large number of departments.

[0187] It should be noted that the personnel referred to are those corresponding to the department.

[0188] The assessment department's overall review coverage "meets the requirements." Key terms explained: Composition Ratio (R): This refers to the proportion of the total number of people in the department who have been marked as requiring in-depth analysis (including secondary identification subjects S and further identification personnel F). R = |S ∪ F| / |D|.

[0189] The requirement is met: the proportion of the composition R is greater than 70%.

[0190] Balancing efficiency and comprehensiveness: If over 70% of the people in a department have been placed on the in-depth review list, then the remaining minority (<30%) constitutes the low-risk majority. From a management efficiency perspective, it can be assumed that the overall risk of the department has been fully revealed through the review of this majority of people, eliminating the need to invest resources in a comprehensive review of the remaining low-risk personnel. This ensures that overall efficiency is not too low when multiple departments require auditing.

[0191] Example (continued from the previous step): Secondary identification targets S = {P1, P2, P6, P7} (4 people), further identification personnel F = {P3, P5} (2 people), total number of personnel requiring in-depth review S ∪ F = {P1, P2, P3, P5, P6, P7}, a total of 6 people, total number of people in the department |D| = 8.

[0192] The composition ratio R = 6 / 8 = 75%. Therefore, R = 75% > 70%, and the overall departmental review coverage "meets the requirements." Proceed to step 3 for more refined judgment.

[0193] Additionally, it should be noted that if the proportion of the further identified personnel and the secondary identified objects in the total number of personnel meets the requirements, the identification frequency threshold is determined based on the proportion of the further identified personnel and the secondary identified objects in the total number of personnel. If the number of violations by the personnel exceeds the identification frequency threshold, but the number of violations does not meet the requirements, then the personnel is determined to be personnel for further identification and processing of violations. In other cases, if the personnel have violations that exceed the identification frequency threshold, then the proportion of the video recordings corresponding to the personnel that are processed for analysis is determined.

[0194] Once the department coverage meets the requirements, further refined screening of the remaining personnel is conducted: Overall logic: When R>70%, for the remaining personnel not in S ∪ F (D \ (S ∪ F) = {P8} in this example), they are not directly exempted from review. Instead, a higher, dynamic threshold for the number of violations is set for secondary screening.

[0195] Keyword explanation: Identification frequency threshold (T): A dynamic threshold for the number of violations, used to determine whether the remaining personnel are "high enough to warrant exceeding efficiency principles for review". T is positively correlated with the proportion of the composition R: the higher the proportion (the wider the scope of review), the higher the threshold T can be set (more lenient towards the remaining personnel).

[0196] The number of violations exceeding the identification threshold does not meet the requirements: This statement in the original text needs to be understood in conjunction with the examples in the following text. It usually refers to whether the number of violation types that the person has that exceed the threshold T reaches a "dangerous number" (e.g., no less than 2).

[0197] 3.1 Example of determining the dynamic "recognition count threshold (T)" (applied rule): Given that the composition percentage R = 75% (between 30% and 70%). According to the rule: "When it is between 30% and 70%, the recognition count threshold is 3." Therefore, the threshold for the number of recognitions is determined to be T = 3.

[0198] Judgment rules (summarized from examples below): "Not meeting the requirements" should be understood as less than 2. That is, if a remaining person has more than 2 violations exceeding the threshold T=3, then they are considered extremely high-risk and need to be included in in-depth review. Otherwise (0 or 1), they proceed to the next round (step 3.3) of screening.

[0199] Examples (analyzing pages 4 and 8): Checking the violation records for P8: {"Incorrect processing procedure": 4 times, "Inappropriate language": 3 times, "Improper use of equipment": 1 time}.

[0200] The behavior with more than 3 violations is "processing error" (4 times). Therefore, the only violation that exceeds the threshold T=3 is "processing error".

[0201] The quantity is 1, which is less than 2. Therefore, P8 was not immediately identified as a deep reviewer for violating this rule. P8 proceeds to the next step.

[0202] In one possible specific embodiment, if the proportion of the video recordings corresponding to the person being analyzed is less than 0.6, then the person is determined to be a person for further identification and processing of violations. If the person does not have any violations with a number of violations exceeding the identification threshold, or if the proportion of the video recordings corresponding to the person being analyzed is not less than 0.6, then based on the identification data of the video recordings of the person for further identification and processing of violations, it is determined whether the person needs to undergo further identification and processing of violations.

[0203] Percentage of video clips analyzed: This refers to the proportion of video clips in the individual's law enforcement video that have been analyzed (regardless of the stage).

[0204] This is the last line of defense. For the remaining personnel, even if their individual violations are not extremely frequent, if we have too little information about them (low video resolution), then the existing violation records may only be the tip of the iceberg. To avoid underestimating the risk due to incomplete information, supplementary review of their unresolved video recordings is necessary.

[0205] Example (Judging P4 and P8): Assume that: P8 has a total of 15 video segments, of which only 8 have been parsed. Parsing percentage = 8 / 15 ≈ 0.53 (<0.6).

[0206] Judgment: P8: Analysis percentage 0.53 < 0.6. Therefore, P8 is identified as a person requiring further identification and processing of violations.

[0207] In one possible embodiment, the larger the proportion of the further identified personnel and the secondary identified objects in the personnel, the larger the threshold for the number of identifications, thereby avoiding the technical problem of high identification processing efficiency caused by excessive video recording for identification when there are too many departments.

[0208] Additionally, it is understandable that after all the video recordings of personnel who have undergone further identification and processing of violations have been analyzed, in a possible specific embodiment, if a person who has not undergone comprehensive video analysis has committed more than three violations, in order to avoid the technical problem of excessive difficulty in recommending training courses for such a person due to incomplete identification, this person will be considered as someone who needs further identification and processing of violations. This will enable comprehensive analysis and processing of the video recordings of such a person. In other cases, no identification and processing is required.

[0209] It is understood that the types of violations include improper use of equipment, use of inappropriate language, and errors in the processing procedure.

[0210] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to mutually. Each embodiment focuses on describing the differences from other embodiments. In particular, the embodiments of apparatus, devices, and non-volatile computer storage media are basically similar to the method embodiments, so the descriptions are relatively simple; relevant parts can be referred to the descriptions of the method embodiments.

[0211] The foregoing has described specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recited in the claims may be performed in a different order than that shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily require the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are possible or may be advantageous.

[0212] The above description is merely one or more embodiments of this specification and is not intended to limit this specification. Various modifications and variations can be made to the one or more embodiments of this specification by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principle of one or more embodiments of this specification should be included within the scope of the claims of this specification.

Claims

1. A method for identifying, analyzing, and processing violations during training, characterized in that, Specifically, it includes: Using the personnel data corresponding to the recordings of each event in the recorder, the analysis targets among the personnel are determined. Based on the identification results of the violations of each person and the corresponding personnel data in the recordings, when it is determined that there is no need to perform full recording analysis, the number of secondary identification targets is determined based on the identification results of the violations of the personnel. Using the quantity as a constraint, and leveraging the correlation between individuals exhibiting violations in the video recordings of each individual, secondary identification targets among the individuals are determined. The video recordings of the secondary identification objects are processed to obtain secondary identification results of each person's violations. Based on the identification data of the secondary identification objects, the secondary identification results and the identification results are used to determine the personnel who will be further identified and processed for violations.

2. The method for identifying, analyzing, and processing violations in training as described in claim 1, characterized in that, The video recordings of each event are for different law enforcement incidents.

3. The method for identifying, analyzing, and processing violations in training as described in claim 1, characterized in that, The personnel data corresponding to the video recordings of each event are determined based on the law enforcement personnel data corresponding to each law enforcement event.

4. The method for identifying, analyzing, and processing violations in training as described in claim 1, characterized in that, The method for determining the number of secondary recognition objects is as follows: Based on the identification results of the personnel's violations, determine the number of personnel who have committed violations; By analyzing the video recordings of individuals, the identification and matching of their violations can be determined. The number of individuals with violations and the matching results are used to determine the number of secondary identification targets.

5. The method for identifying, analyzing, and processing violations in training as described in claim 4, characterized in that, If the number of individuals with violations does not meet the requirements, a preset number will be used to determine the number of secondary identification objects.

6. The method for identifying, analyzing, and processing violations in training as described in claim 1, characterized in that, The method for determining the secondary identification targets among the personnel is as follows: The number of videos showing violations among other personnel and the personnel whose violations do not meet the requirements are used as secondary violation identification targets. Personnel with violations other than the secondary violation identification targets are used as violation analysis personnel. The goal is to analyze and process all videos except those of the identified object, by different violation analysts. The constraint is the number of secondary identification objects minus the number of secondary violation identification objects. The personnel other than the identified object and the secondary violation identification objects are freely combined, and multiple schemes are obtained by combining the secondary violation identification objects. Based on the number of videos of different personnel in the scheme that are analyzed and processed, excluding the videos of the identified target and the secondary identified violators, the preferred scheme is determined, and the personnel corresponding to the preferred scheme are used as the secondary identification targets.

7. The method for identifying, analyzing, and processing violations in training as described in claim 6, characterized in that, The proposed scheme involves taking the individuals other than the identified violators (excluding those identified in the secondary identification process) as a basis to obtain multiple personnel combinations, and then using these personnel combinations together with the identified violators in the secondary identification process to construct personnel in different schemes.

8. The method for identifying, analyzing, and processing violations in training as described in claim 7, characterized in that, The number of people in the personnel group is a constraint number.

9. The method for identifying, analyzing, and processing violations in training as described in claim 1, characterized in that, The method for determining the personnel responsible for further identifying and processing violations is as follows: Based on the identification data of the secondary identification object, the number of secondary identification objects is determined; Using the secondary identification results and the identification results, the number of violations committed by the person under different types of violations is determined; Based on the number of secondary identification objects, the number of violations committed by the person under different types of violations, and the parsing and processing data of the video recordings of the person, it is determined whether the person belongs to the group for further identification and processing of violations.

10. A computer system, comprising: A memory and processor connected by communication, and a computer program stored in the memory and capable of running on the processor, characterized in that, when the processor runs the computer program, it executes a method for identifying and analyzing violations for training as described in any one of claims 1-9.

Images