Fighting behavior recognition method and device, electronic equipment and storage medium

Abstract

The application provides a fighting behavior recognition method and device, electronic equipment and a storage medium, and relates to the technical field of video processing. The method comprises the following steps: in the case that fighting behavior is detected in a target video, determining that suspected fighting behavior occurs; according to a fighting behavior recognition rule, identifying whether a target behavior determined as fighting behavior in the target video is fighting behavior, and in the case that the target behavior is identified as fighting behavior, determining that fighting behavior occurs; the fighting behavior recognition rule comprises the following steps: determining a behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction evaluation information in the behavior influence area; according to the crowd reaction evaluation information, calculating a fighting behavior occurrence probability; in the case that the fighting behavior occurrence probability is greater than a set probability threshold, determining that the target behavior is fighting behavior. The application can improve the recognition accuracy of fighting behavior.

Description

Technical Field

[0001] This invention relates to the field of video processing technology, and in particular to a method and apparatus for identifying fighting behavior, an electronic device, and a storage medium. Background Technology

[0002] Currently, when analyzing human behavior based on artificial intelligence (AI), the method involves capturing human posture and then analyzing that posture to determine the behavior.

[0003] For fighting behavior, the false detection rate is high because the human body posture changes frequently and has a high degree of freedom during the behavior, and there are many variables such as occlusion and image blurring. Summary of the Invention

[0004] This invention provides a method, device, electronic equipment, and storage medium for identifying fighting behavior, in order to address the shortcomings of the high false detection rate of fighting behavior in the prior art and to improve the accuracy of fighting behavior identification.

[0005] This invention provides a method for recognizing fighting behavior, comprising the following steps.

[0006] If fighting is detected in a target video, it is determined that a fight is suspected to have occurred. Based on fighting behavior identification rules, the system identifies whether the target behavior deemed to be fighting in the target video is actually a fight. If the target behavior is identified as a fight, it is determined that a fight has occurred. The fighting behavior identification rules include: determining the area of ​​influence corresponding to the target behavior in the target video; obtaining crowd reaction assessment information within the area of ​​influence; calculating the probability of a fight occurring based on the crowd reaction assessment information; and determining that the target behavior is a fight if the probability of a fight occurring is greater than a set probability threshold.

[0007] According to a method for identifying fighting behavior provided by the present invention, the crowd reaction assessment information includes: multiple crowd reaction information corresponding to multiple confirmation frames; obtaining crowd reaction assessment information in the behavior influence area, including: obtaining multiple confirmation frames corresponding to the target behavior; obtaining crowd reaction information in the behavior influence area of ​​each confirmation frame, thus obtaining multiple crowd reaction information corresponding to multiple confirmation frames; obtaining multiple evaluation feature values ​​corresponding to multiple evaluation dimensions of each confirmation frame based on multiple evaluation dimensions and the multiple crowd reaction information corresponding to multiple confirmation frames; and calculating the probability of fighting behavior occurring based on the crowd reaction assessment information, including: calculating the probability of fighting behavior occurring based on the multiple evaluation feature values ​​corresponding to multiple evaluation dimensions of each confirmation frame.

[0008] According to a method for identifying fighting behavior provided by the present invention, multiple evaluation feature values ​​corresponding to multiple evaluation dimensions of each confirmation frame are obtained based on multiple crowd reaction information corresponding to multiple evaluation dimensions and multiple confirmation frames. The method includes: obtaining initial feature values ​​for each evaluation dimension of a first confirmation frame based on crowd reaction information related to the first confirmation frame; wherein the crowd reaction information related to the first confirmation frame includes: crowd reaction information corresponding to the first confirmation frame and crowd reaction information corresponding to the previous frame of the first confirmation frame; the first confirmation frame is any one of the multiple confirmation frames; normalizing the initial feature values ​​of each evaluation dimension of each confirmation frame to obtain normalized feature values ​​for each evaluation dimension of each confirmation frame; assigning first weights corresponding to the first evaluation dimensions to the normalized feature values ​​of the first evaluation dimensions of each confirmation frame according to multiple first weights corresponding to the multiple evaluation dimensions, thereby obtaining multiple evaluation feature values ​​corresponding to the multiple evaluation dimensions of each confirmation frame; wherein the first evaluation dimension is any one of the multiple evaluation dimensions.

[0009] According to a method for identifying fighting behavior provided by the present invention, the probability of a fighting behavior occurring is calculated based on multiple evaluation feature values ​​corresponding to multiple evaluation dimensions of each confirmation frame. The method includes: obtaining a crowd reaction dimension score for the second evaluation dimension of each confirmation frame based on the evaluation feature values ​​corresponding to the second evaluation dimension of each confirmation frame; wherein the second evaluation dimension is any one of multiple evaluation dimensions; assigning a second weight corresponding to the second evaluation dimension to the crowd reaction dimension score of the second evaluation dimension of each confirmation frame based on multiple second weights corresponding to the multiple evaluation dimensions, thereby obtaining a weighted score for the crowd reaction dimension corresponding to the multiple evaluation dimensions of each confirmation frame; obtaining a total crowd reaction score for the target confirmation frame based on the weighted score for the multiple crowd reaction dimensions corresponding to the multiple evaluation dimensions of the target confirmation frame; wherein the target confirmation frame is any one of multiple confirmation frames; obtaining the probability of a fighting behavior occurring in the target confirmation frame based on the total crowd reaction score of the target confirmation frame, thereby obtaining multiple occurrence probabilities corresponding to the multiple confirmation frames; and obtaining the probability of a fighting behavior occurring based on the multiple occurrence probabilities corresponding to the multiple confirmation frames.

[0010] According to a method for identifying fighting behavior provided by the present invention, before determining that a suspected fight has occurred when a fighting behavior is detected in a target video, the method further includes: acquiring training samples; wherein the training samples include multiple video samples and video sample labels labeled for each of the multiple video samples, the video sample labels including fighting labels and non-fighting labels; training an initial training model based on fighting behavior recognition rules according to the training samples, and determining multiple first weights and multiple second weights according to the training results.

[0011] According to the method for identifying fighting behavior provided by the present invention, the crowd reaction information includes at least one of the following: audio, total number of people, and face orientation; the multiple evaluation dimensions include at least one of the following: crowd clustering degree, rate of change of the number of people in the crowd, crowd dispersion, rate of change of the number of people in the dispersed crowd, rate of change of the number of onlookers, number of onlookers, and crowd voice characteristics.

[0012] According to a method for identifying fighting behavior provided by the present invention, the multiple confirmation frames include at least one of the following: multiple subsequent frames within a first time period after the target time, and multiple past frames collected within a second time period before the target time; wherein, the target time is the time when fighting behavior is detected in the target video.

[0013] The present invention also provides a fighting behavior recognition device, comprising the following modules: a determination module and a processing module.

[0014] The determination module is used to determine whether a fight is suspected to have occurred when a fight is detected in a target video.

[0015] The processing module is used to identify whether a target behavior in a target video that is judged to be a fighting behavior is actually a fighting behavior, based on fighting behavior recognition rules, and to determine whether a fighting behavior has occurred if the target behavior is identified as a fighting behavior. The fighting behavior recognition rules include: determining the behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction assessment information in the behavior influence area; calculating the probability of fighting behavior occurring based on the crowd reaction assessment information; and determining that the target behavior is a fighting behavior if the probability of fighting behavior occurring is greater than a set probability threshold.

[0016] The present invention also provides an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement any of the fighting behavior recognition methods described above.

[0017] The present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements any of the fighting behavior recognition methods described above.

[0018] The present invention also provides a computer program product, including a computer program that, when executed by a processor, implements any of the fighting behavior recognition methods described above.

[0019] This invention provides a method, apparatus, electronic device, and storage medium for identifying fighting behavior. When fighting behavior is detected in a target video, it determines that a fight is suspected to have occurred. Based on fighting behavior identification rules, it identifies whether the target behavior in the target video, which is judged to be fighting, is actually a fight. If the target behavior is identified as fighting, it determines that a fight has occurred. The fighting behavior identification rules include: determining the behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction assessment information within the behavior influence area; calculating the probability of fighting occurrence based on the crowd reaction assessment information; and determining that the target behavior is a fight if the probability of fighting occurrence is greater than a set probability threshold. Therefore, this invention, when a fighting behavior is detected in a target video, does not immediately determine that the target behavior, which is judged to be fighting, is actually fighting. Instead, it confirms whether the target behavior is actually fighting based on the crowd reaction assessment information surrounding the target behavior. Only when the crowd reaction assessment information also confirms that the target behavior is fighting is a fight considered to have occurred, making the identification result of fighting behavior more accurate. Therefore, the fighting behavior detection method provided by the present invention can effectively solve the problem of high false detection rate of fighting behavior in the prior art, and achieve the goal of improving the accuracy of fighting behavior identification. Attached Figure Description

[0020] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0021] Figure 1 This is one of the flowcharts of the fighting behavior recognition method provided in the embodiments of the present invention.

[0022] Figure 2 This is the second flowchart of the fighting behavior recognition method provided in the embodiment of the present invention.

[0023] Figure 3 This is a schematic diagram of the behavior influence area in the fighting behavior recognition provided in the embodiment of the present invention.

[0024] Figure 4 This is the third flowchart of the fighting behavior recognition method provided in the embodiments of the present invention.

[0025] Figure 5 This is the fourth flowchart of the fighting behavior recognition method provided in the embodiments of the present invention.

[0026] Figure 6This is the fifth flowchart of the fighting behavior recognition method provided in the embodiments of the present invention.

[0027] Figure 7 This is a schematic diagram of the structure of the fighting behavior recognition transpose provided in an embodiment of the present invention.

[0028] Figure 8 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0029] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0030] The following is combined with Figures 1-6 The present invention describes a method for recognizing fighting behavior.

[0031] Figure 1 This is one of the flowcharts illustrating the fighting behavior recognition method provided by the present invention, such as... Figure 1 As shown, the method includes the following steps S110~S120.

[0032] S110: If fighting is detected in the target video, it is determined that the fighting is suspected to have occurred.

[0033] Video capture devices (such as cameras) can be pre-set to capture video of a target area (such as a target street, shopping mall, or other public area with high pedestrian traffic). This video is the target video. Then, the target video is checked for any fighting behavior. For example, AI algorithms can be used to analyze the captured target video data, identifying various features related to fighting behavior. Based on these identified features, it can be determined whether fighting has occurred. If fighting behavior is detected in the target video, it is determined that a fight is suspected to have occurred.

[0034] In practice, the AI ​​algorithm and the specific method for determining whether a fight has occurred based on the identified features can be set by those skilled in the art according to the actual situation, and the embodiments of the present invention do not limit this.

[0035] S120: Based on the fighting behavior recognition rules, identify whether the target behavior in the target video that is judged to be fighting behavior is actually fighting behavior, and if the target behavior is identified as fighting behavior, determine that fighting behavior has occurred.

[0036] The target behavior identified as fighting in the target video may or may not be fighting. To reduce the probability of false positives, this embodiment of the invention sets up fighting behavior recognition rules to perform a secondary judgment on the target behavior identified as fighting, thereby reducing the probability of false positives.

[0037] like Figure 2 As shown, the rules for identifying fighting behavior may include S210~S240.

[0038] S210: Identify the area of ​​influence of the target behavior in the target video.

[0039] In real-world scenarios, after a fight occurs, people typically gather and disperse within a certain area surrounding the location of the fight. For example, after a fight breaks out, people nearby may move away from the fight to avoid being affected; conversely, some people may gather closer to the fight to watch or stop it. Simultaneously, during a fight, onlookers usually gather around the location of the fight, their faces typically facing the direction of the incident. To accurately identify the gathering, dispersal, and onlooker behavior caused by a fight, when a fight is detected in the target video, the area of ​​influence of the target behavior on surrounding people is determined. This allows for the identification of the gathering, dispersal, and onlooker behavior caused by the target behavior based on changes in the number of people in the area, their facial orientation, and their posture. The embodiments of the present invention utilize the reactions and changes in personnel within the area affected by the target behavior after it occurs to make a secondary determination as to whether the target behavior is a fight, thereby improving the accuracy of the fight behavior identification results.

[0040] In order to accurately obtain the reactions and changes of people around the location of the target behavior, when determining the area of ​​influence of the target behavior, we can first obtain the target position of the target behavior in the video frame of the target video, and then determine the area of ​​influence of the target behavior based on the target position.

[0041] For example, such as Figure 3As shown, in video frame A of the target video, the area where the target behavior occurs is region B. A center point can be determined in region B (e.g., the intersection of diagonals, a point of central symmetry, etc., which can be set by those skilled in the art according to the actual situation). This center point is determined as the location O where the target behavior occurs. In video frame A, a circle ⊙O is drawn with location O as the center and a preset distance (which can be set by those skilled in the art according to the actual situation). The area of ​​⊙O is determined as the behavior influence area of ​​the target behavior (see...). Figure 3 (Middle shaded area).

[0042] Of course, it is understood that the shape of the area affected by the target behavior can be, but is not limited to, a circle. For example, it can also be a rectangle, an irregular polygon, etc., and this embodiment of the invention does not limit this.

[0043] The size of the area affected by the behavior can be fixed or can be changed according to the location where the target behavior occurs in the target video.

[0044] For example, when a fight is detected in a target video, the background image of the current frame in the target video is used to analyze the location where the fight is determined to occur. The location could include places like shopping malls, streets, scenic spots, restaurants, schools, etc. Depending on the location, different sizes of the area affected by the behavior are determined.

[0045] For example, in a certain area, shopping malls usually have a large flow of people, while restaurants have a small flow of people. If the behavior occurs in a shopping mall, the determined area of ​​influence is D1, and if the behavior occurs in a restaurant, the determined area of ​​influence is D2. By pre-setting the corresponding parameters, the determined area of ​​influence of behavior D1 can be made larger than the area of ​​influence of behavior D2.

[0046] Of course, other reference factors can also be set to comprehensively determine the size of the behavior's influence area. For example, if the location is a shopping mall, where foot traffic is usually lower in the morning and higher in the afternoon and evening, if the target behavior occurs in the morning (e.g., 9:00-12:00), the behavior influence area corresponding to the target behavior is the first area; if the target behavior occurs in the afternoon (e.g., 12:00-18:00), the behavior influence area corresponding to the target behavior is the second area. By setting appropriate parameters, the first area can be made smaller than the second area.

[0047] S220: Obtain assessment information on population responses in the area affected by the behavior.

[0048] Crowd response assessment information may include multiple crowd response information corresponding to multiple confirmation frames. In this case, when obtaining crowd response assessment information in the area affected by the behavior, multiple confirmation frames corresponding to the target behavior can be obtained first, then the crowd response information in the area affected by the behavior in each of the multiple confirmation frames can be obtained, and finally, multiple crowd response information corresponding to multiple confirmation frames can be obtained.

[0049] In some embodiments, the plurality of confirmation frames may include at least one of the following: a plurality of subsequent frames within a first time period after the target time, and a plurality of past frames acquired within a second time period before the target time; wherein the target time is the time when fighting behavior is detected in the target video.

[0050] Typically, fights are followed by onlookers and people running away, so multiple subsequent frames can be captured and used as multiple confirmation frames. These multiple subsequent frames can be consecutive video frames.

[0051] The first time period can be set by those skilled in the art according to the actual situation, such as 15s, 20s, 30s, etc.

[0052] The reason for acquiring multiple past frames is that fighting often has a delayed effect; for example, before a fight breaks out, there is usually some minor physical contact. Therefore, to further improve the accuracy of fighting identification, multiple past frames can be acquired. By analyzing the crowd reactions in the affected areas of multiple past frames, it is possible to more accurately determine whether a fight has occurred. For example, if the crowd reactions in the affected areas of multiple past frames do not match the typical reactions to a fight—for example, if there is no obvious gathering, dispersal, or onlooker behavior in the affected areas—the probability of the target behavior being identified as a fight is lower; conversely, the more obvious the gathering, dispersal, and onlooker behavior in the affected areas, the higher the probability of the target behavior being identified as a fight. This idea can be applied to subsequent model training. When weighting multiple evaluation dimensions for gathering, dispersal, and onlooker behavior, accurate weighted parameter values ​​can be obtained through model training.

[0053] The second time period can be set by those skilled in the art according to the actual situation, such as 15s, 20s, 30s, etc.

[0054] The first time period and the second time period can be the same or different.

[0055] It should be noted that since the reactions of the surrounding crowd after a fight occurs (i.e., the stress response of the surrounding people to the fight, such as watching, running away, etc.) are often more obvious than the reactions of the surrounding crowd before the fight occurs, when acquiring multiple confirmation frames corresponding to the target behavior, it is preferable to acquire multiple subsequent frames in the first time period after the target time when the fight is detected in the target video, and identify the multiple subsequent frames as multiple confirmation frames.

[0056] The following describes the process of obtaining crowd reaction information within the behavioral impact area of ​​each of the multiple confirmation frames, thus obtaining multiple crowd reaction information corresponding to each of the multiple confirmation frames.

[0057] Crowd reaction information can include information related to how pedestrians would typically react to witnessing a fight. In specific implementations, crowd reaction information may include, for example, audio, the total number of people, and the orientation of faces.

[0058] Audio refers to the sound data synchronized with each frame of the video. When a fight occurs, it is usually accompanied by screams, cries for help, shouts, and other sounds.

[0059] The total number of people refers to the total number of people in the area affected by the behavior. It can reflect the gathering behavior, evacuation behavior, etc. of people around the target behavior under the influence of the target behavior.

[0060] The orientation of a face refers to the orientation of faces within the area affected by the behavior, which can reflect the onlookers' behavior around the target behavior under its influence.

[0061] Of course, it is understood that the crowd reaction information listed above is merely exemplary. The crowd reaction information in the embodiments of the present invention includes, but is not limited to, the information listed above. For example, it may also include the walking speed of people in the crowd (for example, when people see a fight, their attention will suddenly focus and their walking speed will suddenly slow down or speed up).

[0062] Of course, it is understandable that the crowd reaction assessment information can also be the crowd reaction information in the behavior influence area of ​​the current video frame (i.e., the video frame when the fighting behavior is detected). The process of obtaining the crowd reaction information in the current video frame is the same as the process of obtaining the crowd reaction information in each confirmation frame mentioned above. For details, please refer to the relevant description in the section on obtaining the crowd reaction information in each confirmation frame in the above embodiments, which will not be repeated here.

[0063] S230: Calculate the probability of fighting based on crowd response assessment information.

[0064] Based on the population response information corresponding to multiple evaluation dimensions and multiple confirmation frames one by one, multiple evaluation feature values corresponding to multiple evaluation dimensions of each confirmation frame can be obtained, and then based on the multiple evaluation feature values corresponding to multiple evaluation dimensions of each confirmation frame, the probability of a fighting behavior occurring can be calculated.

[0065] In some embodiments, such as Figure 4 shown, the execution process of obtaining multiple evaluation feature values corresponding to multiple evaluation dimensions of each confirmation frame based on the population response information corresponding to multiple evaluation dimensions and multiple confirmation frames one by one may include the following S410 - S430.

[0066] S410: Obtain the initial feature value of each evaluation dimension of the first confirmation frame according to the population response information related to the first confirmation frame; wherein, the population response information related to the first confirmation frame includes: the population response information corresponding to the first confirmation frame and the population response information corresponding to the previous frame of the first confirmation frame; the first confirmation frame is any one of the multiple confirmation frames.

[0067] The multiple evaluation dimensions may include at least two of the following: population aggregation degree, population aggregation number change rate, population dispersion degree, population dispersion number change rate, population onlooker number change rate, onlooker number, and population sound characteristics.

[0068] For each of the multiple evaluation dimensions, a rule for obtaining the initial feature value corresponding to each evaluation dimension based on the population response information is preset. Based on this, after obtaining the population response information corresponding to each confirmation frame, the initial feature value of this evaluation dimension of this confirmation frame can be obtained according to the population response information corresponding to each confirmation frame and the rule for obtaining the initial feature value corresponding to each evaluation dimension.

[0069] Taking the seven dimensions of population aggregation degree, population aggregation number change rate, population dispersion degree, population dispersion number change rate, population onlooker number change rate, onlooker number, and population sound characteristics as an example for the multiple evaluation dimensions below, the obtaining of the initial feature value of each evaluation dimension of each confirmation frame will be described.

[0070] The multiple confirmation frames, for example, include n (n is a positive integer and n is greater than or equal to 2) confirmation frames. Taking the above multiple evaluation dimensions as the to - be - calculated evaluation dimensions in sequence, calculating the initial feature value of the to - be - calculated evaluation dimension of the m - th (0 < m ≤ n) confirmation frame among the n confirmation frames can obtain the initial feature values of multiple evaluation dimensions of the m - th confirmation frame (which can be regarded as the above first confirmation frame).

[0071] For example, when the to - be - calculated evaluation dimension is the population aggregation degree, the initial feature value of the population aggregation degree of the m - th confirmation frame can be calculated by the following method: ; wherein, Let be the initial feature value of the crowd aggregation degree in the m-th confirmed frame. Let m be the total number of people in the crowd response information of the m-th confirmation frame. The area of ​​the region affected by the behavior.

[0072] For example, if the evaluation dimension to be calculated is the rate of change in the number of people gathered in a crowd, the initial feature value of the rate of change in the number of people gathered in a crowd in the m-th confirmation frame can be calculated as follows: ; in, Let be the initial feature value of the rate of change in the number of people in the crowd at the m-th confirmed frame. The total number of people in the crowd response information of the (m-1)th confirmation frame is the frame preceding the m-th confirmation frame in terms of time sequence.

[0073] For example, if the evaluation dimension to be calculated is the crowd dispersion, the initial feature value of the crowd dispersion of the m-th confirmation frame can be calculated as follows: ; in, It is the initial feature value of the crowd dispersion of the m-th confirmation frame. The target location of the target behavior within the video frame of the target video. Let the action of the m-th acknowledgment frame affect the location of the i-th person within the area. Let be the distance between the location of the i-th person and the target location.

[0074] For example, if the evaluation dimension to be calculated is the rate of change of the discrete number of people in the crowd, the initial feature value of the rate of change of the discrete number of people in the crowd in the m-th confirmation frame can be calculated as follows: .

[0075] in, Let be the initial characteristic value of the rate of change of the discrete number of people in the crowd for the m-th confirmed frame.

[0076] For example, if the evaluation dimension to be calculated is the number of onlookers, the initial feature value of the number of onlookers in the m-th confirmation frame can be obtained as follows: In the m-th confirmation frame, based on the face orientation and the target position in the target video frame, determine whether the person corresponding to the face is an onlooker. If so, increment the number of onlookers by 1; otherwise, leave the number of onlookers unchanged, thus obtaining the number of onlookers in the m-th confirmation frame. .

[0077] The method for determining whether a person corresponding to a face is a bystander based on the face's orientation and the target's location in the video frame can be set by those skilled in the art according to the actual situation, and this embodiment of the invention does not limit this.

[0078] For example, if the evaluation dimension to be calculated is the rate of change in the number of onlookers, the initial feature value of the rate of change in the number of onlookers in the m-th confirmed frame can be calculated as follows: ; in, Let be the initial feature value for the change in the number of onlookers in the m-th confirmed frame. Let m be the number of people watching the (m-1)th confirmed frame.

[0079] For example, if the evaluation dimension to be calculated is the crowd sound feature, the initial feature value of the crowd sound feature in the m-th confirmation frame can be calculated as follows: extract the volume peak of the audio in the m-th confirmation frame, obtain the feature vector Vo of the volume peak, and determine the feature vector as the initial feature value of the crowd sound feature in the m-th confirmation frame. In specific implementation, after extracting the volume peak of the audio in the m-th confirmation frame, the feature vector of the extracted volume peak can be obtained based on the Mel Frequency Cepstral Coefficient (MFCC).

[0080] For each of the n confirmation frames, the initial feature values ​​of the above multiple evaluation dimensions are calculated in the same way as the m-th confirmation frame. This yields the initial feature values ​​of the 1st to nth confirmation frames in multiple evaluation dimensions, which is equivalent to obtaining the initial feature values ​​of the multiple evaluation dimensions for each confirmation frame in the multiple confirmation frames.

[0081] S420: Normalize the initial feature values ​​of each evaluation dimension of each confirmation frame to obtain the normalized feature values ​​of each evaluation dimension of each confirmation frame.

[0082] Based on the initial feature values ​​of each confirmation frame in multiple evaluation dimensions, the normalized feature values ​​of each evaluation dimension of the confirmation frame can be obtained, thus obtaining multiple normalized feature values ​​corresponding one-to-one to the multiple evaluation dimensions of the confirmation frame.

[0083] Continuing with the example above, if we assume the initial feature values ​​of the evaluation dimension to be calculated for the 1st to nth confirmation frames are F1, F2, F3, ..., Fn, respectively, then the normalized feature value of the evaluation dimension to be calculated for the mth confirmation frame can be obtained based on the following formula. .

[0084] .

[0085] in, The normalized feature value of the m-th confirmed frame in the evaluation dimension to be calculated; The initial feature value of the m-th confirmed frame in the evaluation dimension to be calculated; It is the minimum value among the initial feature values ​​of the evaluation dimension to be calculated from the first confirmation frame to the nth confirmation frame, that is, the minimum value among F1, F2, F3, ..., Fn; It is the maximum value among the initial feature values ​​of the evaluation dimension to be calculated in the first to the nth confirmation frames, that is, the maximum value among F1, F2, F3, ..., Fn.

[0086] To illustrate further, based on the multiple initial feature values ​​corresponding to the multiple evaluation dimensions of the m-th confirmation frame obtained above: Using the above formula for calculating normalized feature values, we can obtain the following normalized feature values ​​corresponding to the multiple evaluation dimensions of the m-th confirmation frame: ;in, It is the normalized feature value of the crowd aggregation degree evaluation dimension of the m-th confirmed frame. It is the normalized feature value of the evaluation dimension of the rate of change in the number of people gathered in the m-th confirmed frame. It is the normalized feature value of the crowd dispersion evaluation dimension of the m-th confirmed frame. It is the normalized feature value of the evaluation dimension of the rate of change of the discrete number of people in the m-th confirmed frame. It is the normalized feature value of the number of onlookers in the m-th confirmed frame as an evaluation dimension. It is the normalized feature value of the evaluation dimension of the rate of change in the number of onlookers in the m-th confirmed frame. It is the normalized feature value of the crowd sound features evaluation dimension of the m-th confirmed frame.

[0087] S430: Based on the multiple first weights corresponding to the multiple evaluation dimensions, assign the first weight corresponding to the first evaluation dimension to the normalized feature value of the first evaluation dimension of each confirmation frame, thereby obtaining multiple evaluation feature values ​​corresponding to the multiple evaluation dimensions of each confirmation frame; wherein, the first evaluation dimension is any one of the multiple evaluation dimensions.

[0088] Based on the importance of each evaluation dimension among multiple evaluation dimensions, a weight is set for each evaluation dimension, resulting in multiple first weights corresponding to each of the multiple evaluation dimensions.

[0089] Continuing with the example above, for the m-th confirmation frame, the multiple evaluation feature values ​​corresponding to the multiple evaluation dimensions of the m-th confirmation frame can be obtained using the following formula: ; in, ; The first weight corresponding to the degree of crowd aggregation The evaluation feature value corresponding to the degree of crowd agglomeration. The first weight corresponding to the rate of change in the number of people gathered is... The evaluation feature value corresponding to the rate of change in the number of people gathered. The first weight corresponding to the dispersion of the population. These are the evaluation feature values ​​corresponding to the population dispersion. The first weight corresponding to the rate of change of the discrete number of people in the population. The evaluation feature value corresponding to the rate of change of the discrete number of people in the population. The first weight corresponding to the number of onlookers. The evaluation feature value corresponding to the number of onlookers. The first weight corresponding to the rate of change in the number of onlookers. The evaluation feature value corresponding to the rate of change in the number of onlookers. The first weight corresponding to the voice features of the crowd. V represents the evaluation feature value corresponding to the voice characteristics of the crowd. V is a set of multiple evaluation feature values.

[0090] After that, such as Figure 5 As shown, the execution process of calculating the probability of a fighting behavior occurring, which is a fighting behavior, based on the multiple evaluation feature values ​​corresponding to the multiple evaluation dimensions of each confirmation frame, may include the following steps S510~S550.

[0091] S510: Based on the evaluation feature value corresponding to the second evaluation dimension of each confirmation frame, obtain the population response dimension score of the second evaluation dimension of each confirmation frame; wherein, the second evaluation dimension is any one of multiple evaluation dimensions.

[0092] In some embodiments, the evaluation feature value of the second evaluation dimension of the target confirmation frame can be directly determined as the evaluation dimension score of the second evaluation dimension of the target confirmation frame.

[0093] In some embodiments, multiple evaluation dimension thresholds corresponding to multiple evaluation dimensions can be preset in advance. The evaluation dimension score of the second evaluation dimension of the second confirmation frame can be obtained based on the evaluation feature value of the second evaluation dimension and the evaluation dimension threshold of the second evaluation dimension of the second confirmation frame (any one of the multiple confirmation frames).

[0094] For example, the evaluation feature value of the second evaluation dimension of the second confirmation frame is K1, the evaluation dimension threshold of the second evaluation dimension is G1, and the evaluation dimension score of the second evaluation dimension of the second confirmation frame is Score_U=K1 / G1.

[0095] Taking the above example, when multiple evaluation dimensions include crowd density, rate of change of the number of people in the crowd, crowd dispersion, rate of change of the number of people in the dispersed crowd, rate of change of the number of onlookers, number of onlookers, and crowd sound characteristics, if the second confirmation frame is the m-th confirmation frame, based on the above formula Score_U=K1 / G1, the evaluation dimension score of crowd density, Score_De, can be obtained. / The evaluation dimension score for the rate of change in the number of people gathered is: Score_ = / The evaluation dimension score of population dispersion. = / The evaluation dimension score of the rate of change in the discrete number of people in the population. = / The evaluation dimension score for the rate of change in the number of onlookers is: = / The score for the number of onlookers. = / Score for the evaluation dimensions of crowd voice characteristics = / ;in, This represents the crowd density threshold for crowd aggregation. This represents the threshold for the rate of change in the number of people gathered in a crowd. This represents the population dispersion threshold. This is the threshold for the rate of change of the discrete population size. The threshold for the number of onlookers. This is the threshold for the rate of change in the number of onlookers. The threshold for crowd voice characteristics.

[0096] S520: Based on the multiple second weights corresponding to the multiple evaluation dimensions, assign the second weight corresponding to the second evaluation dimension to the population response dimension score of the second evaluation dimension of each confirmation frame, and obtain the weighted score of the population response dimension corresponding to the multiple evaluation dimensions of each confirmation frame.

[0097] Based on the importance of the population response dimension score under each evaluation dimension in the population response dimension scores of multiple evaluation dimensions, a weight corresponding to each evaluation dimension is set (which can also be regarded as setting the weight corresponding to the population response dimension score under each evaluation dimension), resulting in multiple second weights.

[0098] Taking the above example, the evaluation dimension for crowd density is: Score_De V = Score_De; where... This is the second weight corresponding to the degree of crowd aggregation.

[0099] Evaluation dimensions for the rate of change in the number of people gathered: Score_ V = Score_ ; The second weight corresponds to the rate of change in the number of people gathered.

[0100] Evaluation dimensions for population dispersion: Score_ V = Score_ ; This is the second weight corresponding to the dispersion of the population.

[0101] Evaluation dimensions for the rate of change in the discrete number of people in a population: Score_ V = Score_ ; The second weight corresponding to the rate of change of the discrete number of people in the population Evaluation criteria for the number of onlookers: Score_ V = Score_ ; 5 represents the weight of the first evaluation dimension corresponding to the number of onlookers. Evaluation dimensions for crowd density: Score_ V = Score_ ;, 6 represents the second weight corresponding to the rate of change in the number of onlookers. Evaluation dimensions for crowd voice characteristics: Score_ V = Score_ ; 7 represents the second weight corresponding to the voice characteristics of the crowd.

[0102] S530: Obtain the total population response score for the target confirmation frame by weighting the scores of multiple population response dimensions corresponding to the multiple evaluation dimensions of the target confirmation frame; wherein, the target confirmation frame is any frame among multiple confirmation frames.

[0103] The weighted scores of multiple population response dimensions corresponding to multiple evaluation dimensions of the target confirmation frame can be summed one-to-one, and the summation result can be determined as the total population response score of the target confirmation frame. Taking the above example, if the target confirmation frame is the m-th confirmation frame, the total crowd response score for the m-th confirmation frame can be obtained using the following formula: Score=Score_De V +Score_ V +Score_ V +Score_ V +Score_ V +Score_ V +Score_ V .

[0104] S540: Based on the total score of the crowd response in the target confirmation frame, obtain the probability of fighting occurring in the target confirmation frame, and obtain multiple occurrence probabilities corresponding to multiple confirmation frames.

[0105] For example, if the target confirmation frame is the m-th confirmation frame, the probability P of the fighting behavior corresponding to the m-th confirmation frame can be obtained using the following formula. .

[0106] S550: Based on the multiple occurrence probabilities corresponding to multiple confirmation frames, obtain the target probability and determine the target probability as the probability of the fighting behavior occurring.

[0107] There are several ways to obtain the target probability based on the one-to-one correspondence between multiple confirmation frames and multiple occurrence probabilities. For example, the average of multiple occurrence probabilities can be used to determine the target probability.

[0108] For example, multiple occurrence probabilities can be obtained, their expected values ​​can be calculated, and the result of the expected value can be determined as the target probability.

[0109] S240: If the probability of a fight is greater than a set probability threshold, the target behavior is determined to be a fight.

[0110] The probability threshold can be set by those skilled in the art according to the actual situation, and the embodiments of the present invention do not limit this.

[0111] Of course, it is understandable that if the probability of a fight is less than or equal to a set probability threshold, then the target behavior is determined not to be a fight.

[0112] In some implementations, before executing S110, such as Figure 6 As shown, embodiments of the present invention may also execute the following steps S610-S620.

[0113] S610: Obtain training samples; wherein, the training samples include multiple video samples and video sample labels labeled for each of the multiple video samples, the video sample labels include fighting labels and non-fighting labels.

[0114] It can acquire a large number of fighting videos and a large number of non-fighting videos, resulting in multiple video samples. The fighting videos are labeled with the "fighting" tag, and the non-fighting videos are labeled with the "non-fighting" tag, thus obtaining the labeled video sample tags for each of the multiple video samples.

[0115] S620: Based on the training samples, train the initial training model according to the fighting behavior recognition rules, and determine multiple first weights and multiple second weights based on the training results.

[0116] Based on the aforementioned fighting behavior recognition rules (S210~S240), the initial training model is trained. When the model converges, multiple first weights and multiple second weights obtained from the training of the model are acquired. Thus, multiple first weights and multiple second weights in S210~S240 can be obtained.

[0117] In some embodiments, the following evaluation criteria can be used when training the initial training model, as detailed in Table 1.

[0118] Table 1

[0119] Specifically, referring to Table 1, for the three items "Number of People Evacuated," "Number of People Gathered," and "Number of Bystanders," a "+" indicates an increase in the number of people. For the item "Abnormal Sounds," a "+" indicates that the abnormal sounds have worsened, such as increased volume or a higher frequency of screams and cries for help. Referring to Table 1, if there is a significant increase in the number of bystanders and abnormal sounds, while the number of people evacuated and gathered remains relatively stable, it is likely that the current stage is in the early part of the initial phase of the fight. If there is a significant increase in the number of people evacuated, bystanders, and abnormal sounds, while the number of people gathered remains relatively stable, it is likely that the current stage is in the later part of the initial phase of the fight. If there is a significant increase in the number of people gathered and bystanders, and abnormal sounds, while the number of people evacuated remains relatively stable, it is likely that the current stage is in the early part of the intense phase of the fight. If there is a significant increase in the number of people evacuated, bystanders, and abnormal sounds, while the number of people gathered remains relatively stable, it is likely that the current stage is in the later part of the intense phase of the fight.

[0120] It should be noted that the judgment criteria listed in Table 1 are merely exemplary. In specific implementations, the judgment criteria for the fighting stage in this embodiment of the invention include, but are not limited to, the judgment criteria listed in Table 1.

[0121] This invention provides a method for identifying fighting behavior. When fighting behavior is detected in a target video, a suspected fight is identified. Based on fighting behavior identification rules, the method determines whether the target behavior identified as fighting in the target video is actually a fight. If the target behavior is identified as fighting, the method determines that a fight has occurred. The fighting behavior identification rules include: determining the area of ​​influence corresponding to the target behavior in the target video; obtaining crowd reaction assessment information within the area of ​​influence; calculating the probability of a fight occurring based on the crowd reaction assessment information; and determining that the target behavior is a fight if the probability of a fight occurs is greater than a set probability threshold. Therefore, this invention, when a fight is detected in a target video, does not immediately identify the target behavior identified as fighting. Instead, it confirms whether the target behavior is actually fighting based on the crowd reaction assessment information. Only when the crowd reaction assessment information also confirms that the target behavior is fighting is a fight considered to have occurred, making the identification of fighting behavior more accurate. Therefore, the fighting behavior detection method provided by the present invention can effectively solve the problem of high false detection rate of fighting behavior in the prior art, and achieve the goal of improving the accuracy of fighting behavior identification.

[0122] The following describes the fighting behavior recognition device provided by the present invention. The fighting behavior recognition device described below can be referred to in correspondence with the fighting behavior recognition method described above.

[0123] Figure 7 This is a schematic diagram of the fighting behavior recognition device provided by the present invention. Figure 7 As shown, the fighting behavior recognition device 700 includes: a determination module 701 and a processing module 702.

[0124] The determination module 701 is used to determine whether a fight is suspected to have occurred when a fight is detected in a target video.

[0125] The processing module 702 is used to identify whether a target behavior in a target video that is judged to be a fighting behavior is actually a fighting behavior according to the fighting behavior recognition rules, and to determine that a fighting behavior has occurred if the target behavior is identified as a fighting behavior; wherein, the fighting behavior recognition rules include: determining the behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction assessment information in the behavior influence area; calculating the probability of fighting behavior occurring based on the crowd reaction assessment information; and determining that the target behavior is a fighting behavior if the probability of fighting behavior occurring is greater than a set probability threshold.

[0126] This invention provides a fighting behavior recognition device. Upon detecting fighting behavior in a target video, it determines that a fight is suspected to have occurred. Based on fighting behavior recognition rules, it identifies whether the target behavior deemed as fighting in the target video is actually a fight, and if the target behavior is identified as fighting, it determines that a fight has occurred. The fighting behavior recognition rules include: determining the behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction assessment information within the behavior influence area; calculating the probability of a fight occurring based on the crowd reaction assessment information; and determining that the target behavior is a fight if the probability of a fight occurs is greater than a set probability threshold. Therefore, this invention, upon detecting fighting behavior in a target video, does not immediately confirm the target behavior deemed as fighting. Instead, it confirms whether the target behavior is actually fighting based on the crowd reaction assessment information. Only when the crowd reaction assessment information also confirms that the target behavior is fighting is a fight considered to have occurred, making the recognition result of fighting behavior more accurate. Therefore, the fighting behavior detection method provided by the present invention can effectively solve the problem of high false detection rate of fighting behavior in the prior art, and achieve the goal of improving the accuracy of fighting behavior identification.

[0127] Figure 8 An example is a schematic diagram of the physical structure of an electronic device, such as... Figure 8 As shown, the electronic device may include: a processor 810, a communications interface 820, a memory 830, and a communication bus 840, wherein the processor 810, the communications interface 820, and the memory 830 communicate with each other through the communication bus 840. The processor 810 can call logical instructions in the memory 830 to execute a fighting behavior recognition method. This method includes: determining that a fighting behavior is suspected to have occurred when a fighting behavior is detected in a target video; identifying whether a target behavior in the target video that is judged to be a fighting behavior is actually a fighting behavior according to fighting behavior recognition rules; and determining that a fighting behavior has occurred if the target behavior is identified as a fighting behavior. The fighting behavior recognition rules include: determining the behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction assessment information in the behavior influence area; calculating the probability of the fighting behavior occurring based on the crowd reaction assessment information; and determining that the target behavior is a fighting behavior if the probability of the fighting behavior occurring is greater than a set probability threshold.

[0128] Furthermore, the logical instructions in the aforementioned memory 830 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0129] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the fighting behavior recognition method provided by the above methods. The method includes: determining that a fighting behavior is suspected to have occurred when a fighting behavior is detected in a target video; identifying whether a target behavior in the target video that is judged to be a fighting behavior is actually a fighting behavior according to a fighting behavior recognition rule, and determining that a fighting behavior has occurred if the target behavior is identified as a fighting behavior; wherein the fighting behavior recognition rule includes: determining the behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction assessment information in the behavior influence area; calculating the probability of the fighting behavior occurring based on the crowd reaction assessment information; and determining that the target behavior is a fighting behavior if the probability of the fighting behavior occurring is greater than a set probability threshold.

[0130] In another aspect, the present invention also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the fighting behavior recognition method provided by the above methods. The method includes: determining, upon detecting fighting behavior in a target video, that a fighting behavior is suspected to have occurred; identifying, according to fighting behavior recognition rules, whether a target behavior in the target video that is determined to be fighting behavior is indeed fighting behavior, and determining that fighting behavior has occurred if the target behavior is identified as fighting behavior; wherein the fighting behavior recognition rules include: determining the behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction assessment information in the behavior influence area; calculating the probability of fighting behavior occurring based on the crowd reaction assessment information; and determining that the target behavior is fighting behavior if the probability of fighting behavior occurring is greater than a set probability threshold.

[0131] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0132] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0133] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for recognizing fighting behavior, characterized in that, include: If fighting is detected in the target video, it is determined that the fighting is suspected to have occurred. According to the fighting behavior recognition rules, the system identifies whether the target behavior in the target video that is judged to be fighting behavior is actually fighting behavior, and if the target behavior is identified as fighting behavior, it determines that fighting behavior has occurred; wherein, the fighting behavior recognition rules include: Determine the area of ​​influence of the target behavior in the target video; Obtain assessment information on the crowd response in the area affected by the behavior; Calculate the probability of a fight occurring based on the crowd response assessment information. If the probability of the fighting behavior occurring is greater than a set probability threshold, the target behavior is determined to be a fighting behavior.

2. The method for identifying fighting behavior according to claim 1, characterized in that, The crowd response assessment information includes: multiple crowd response information corresponding to multiple confirmation frames; The acquisition of population response assessment information in the area affected by the behavior includes: Obtain multiple confirmation frames corresponding to the target behavior; Obtain crowd reaction information within the behavioral influence area of ​​each of the multiple confirmation frames, and obtain multiple crowd reaction information corresponding to each of the multiple confirmation frames; The step of calculating the probability of a fight occurring based on the crowd response assessment information includes: Based on multiple evaluation dimensions and the multiple population response information corresponding to the multiple confirmation frames, multiple evaluation feature values ​​corresponding to the multiple evaluation dimensions of each confirmation frame are obtained. The probability of the fighting behavior occurring is calculated based on the multiple evaluation feature values ​​corresponding to the multiple evaluation dimensions of each confirmation frame.

3. The method for identifying fighting behavior according to claim 2, characterized in that, The step of obtaining multiple evaluation feature values ​​corresponding to the multiple evaluation dimensions of each confirmation frame based on multiple evaluation dimensions and multiple population response information corresponding to the multiple confirmation frames includes: Based on the crowd response information related to the first confirmation frame, the initial feature value of each evaluation dimension of the first confirmation frame is obtained; wherein, the crowd response information related to the first confirmation frame includes: the crowd response information corresponding to the first confirmation frame, and the crowd response information corresponding to the previous frame of the first confirmation frame; the first confirmation frame is any one of the plurality of confirmation frames; The initial feature values ​​of each evaluation dimension of each confirmation frame are normalized to obtain the normalized feature values ​​of each evaluation dimension of each confirmation frame. Based on the multiple first weights corresponding to the multiple evaluation dimensions, the normalized feature value of the first evaluation dimension of each confirmation frame is assigned the first weight corresponding to the first evaluation dimension, thereby obtaining multiple evaluation feature values ​​corresponding to the multiple evaluation dimensions of each confirmation frame; wherein, the first evaluation dimension is any one of the multiple evaluation dimensions.

4. The method for recognizing fighting behavior according to claim 3, characterized in that, The step of calculating the probability of the fighting behavior occurring based on the multiple evaluation feature values ​​corresponding one-to-one with the multiple evaluation dimensions of each confirmed frame includes: Based on the evaluation feature value corresponding to the second evaluation dimension of each confirmation frame, obtain the population response dimension score of the second evaluation dimension of each confirmation frame; wherein, the second evaluation dimension is any one of the plurality of evaluation dimensions; Based on the multiple second weights corresponding to the multiple evaluation dimensions, the second weight corresponding to the second evaluation dimension of each confirmation frame is assigned to the crowd response dimension score of the second evaluation dimension, so as to obtain the crowd response dimension weighted score corresponding to the multiple evaluation dimensions of each confirmation frame. The total population response score of the target confirmation frame is obtained by weighting the scores of the multiple population response dimensions corresponding to the multiple evaluation dimensions of the target confirmation frame; wherein, the target confirmation frame is any one of the multiple confirmation frames. Based on the total crowd response score of the target confirmation frame, the probability of a fight occurring in the target confirmation frame is obtained, and multiple occurrence probabilities corresponding to the multiple confirmation frames are obtained one by one. Based on the occurrence probabilities corresponding to the multiple confirmation frames, a target probability is obtained, and the target probability is determined as the probability of the fighting behavior occurring.

5. The method for identifying fighting behavior according to claim 4, characterized in that, Before determining that a fight is suspected to have occurred when a fight is detected in the target video, the process also includes: Obtain training samples; wherein, the training samples include multiple video samples and video sample labels labeled for each of the multiple video samples, the video sample labels including fighting labels and non-fighting labels; Based on the training samples, the initial training model is trained according to the fighting behavior recognition rules, and the plurality of first weights and the plurality of second weights are determined based on the training results.

6. The method for identifying fighting behavior according to any one of claims 2-5, characterized in that, Crowd response information includes at least one of the following: audio, total number of people, and the orientation of faces; The multiple evaluation dimensions include at least one of the following: crowd density, rate of change of the number of people in the crowd, crowd dispersion, rate of change of the number of people in the dispersed crowd, rate of change of the number of onlookers, number of onlookers, and crowd sound characteristics.

7. The method for identifying fighting behavior according to any one of claims 2-5, characterized in that, The plurality of confirmation frames include at least one of the following: a plurality of subsequent frames within a first time period after the target time, and a plurality of past frames collected within a second time period before the target time; wherein, the target time is the time when fighting behavior is detected in the target video.

8. A fighting behavior recognition device, characterized in that, include: The determination module is used to determine whether a fight is suspected to have occurred when a fight is detected in a target video. The processing module is configured to identify whether a target behavior in the target video that is determined to be a fight is actually a fight, based on a fight behavior recognition rule, and to determine that a fight has occurred if the target behavior is identified as a fight. The fight behavior recognition rule includes: determining the behavior influence area corresponding to the target behavior in the target video; obtaining crowd reaction assessment information within the behavior influence area; calculating the probability of a fight occurring based on the crowd reaction assessment information; and determining that the target behavior is a fight if the probability of a fight occurring is greater than a set probability threshold.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the computer program, it implements the fighting behavior recognition method as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the fighting behavior recognition method as described in any one of claims 1 to 7.

Images