Abnormal event detection method, device and equipment

Abstract

The present application provides a method, device, and equipment for detecting abnormal events. The method includes: obtaining at least one storage queue corresponding to a target camera, each storage queue including abnormal event images captured by the target camera; determining user preference information corresponding to the storage queue for each storage queue, and determining image filtering parameters corresponding to the storage queue based on the user preference information; for a target image captured by the target camera and having an abnormal event, selecting a target storage queue corresponding to the target image from all storage queues corresponding to the target camera, and determining whether to filter the target image based on the image filtering parameters corresponding to the target storage queue; if so, filtering the target image; if not, sending the target image to a management platform, which performs abnormal event detection based on the target image. The technical solution of the present application avoids the existence of a large number of redundant target images on the management platform, saving human resources and audit time.

Description

Technical Field

[0001] The present application relates to the field of artificial intelligence, and in particular to a method, apparatus, and device for detecting abnormal events. Background Art

[0002] Machine learning is a path to artificial intelligence and a multidisciplinary field, encompassing probability theory, statistics, approximation theory, convex analysis, and algorithmic complexity theory. It studies how computers can simulate or implement human learning behaviors to acquire new knowledge or skills and reorganize existing knowledge structures to continuously improve their performance. Machine learning focuses on algorithm design, enabling computers to automatically learn patterns from data and use these patterns to make predictions about unknown data. Machine learning has found a wide range of applications, including deep learning, data mining, computer vision, natural language processing, biometrics, search engines, medical diagnostics, speech recognition, and handwriting recognition.

[0003] Based on machine learning technology, an abnormal event detection model can be trained to detect abnormal events in images. For example, an image captured by a camera can be fed into the abnormal event detection model, which then performs abnormal event detection on the image and obtains a result. If an abnormal event is detected in the image, the image can be sent to a management platform, where management personnel can view and review the image and then visit the site (i.e., the camera location) to verify whether an abnormal event has occurred.

[0004] However, a large number of images with abnormal events may be sent to the management platform, and managers usually only review a small number of images, resulting in a large number of redundant images. Moreover, it is difficult for managers to find truly valuable images for review from a large number of images, which causes managers to frequently visit the site to check, requiring a lot of human resources, consuming a lot of time, wasting workload, and having a poor user experience. Summary of the Invention

[0005] The present application provides a method for detecting abnormal events, the method comprising:

[0006] Acquire at least one storage queue corresponding to a target camera, each storage queue including abnormal event images captured by the target camera; wherein the abnormal event images in the same storage queue have similar features, and the abnormal event images in different storage queues have dissimilar features;

[0007] For each storage queue, determining user preference information corresponding to the storage queue, and determining image filtering parameters corresponding to the storage queue based on the user preference information;

[0008] For a target image with an abnormal event captured by a target camera, a target storage queue corresponding to the target image is selected from all storage queues corresponding to the target camera, and whether to filter the target image is determined based on an image filtering parameter corresponding to the target storage queue;

[0009] If so, the target image is filtered.

[0010] The present application provides an abnormal event detection device, the device comprising:

[0011] An acquisition module is configured to acquire at least one storage queue corresponding to a target camera, each storage queue including abnormal event images captured by the target camera; wherein abnormal event images in the same storage queue have similar features, and abnormal event images in different storage queues have dissimilar features;

[0012] a determination module, configured to determine, for each storage queue, user preference information corresponding to the storage queue, and determine image filtering parameters corresponding to the storage queue based on the user preference information;

[0013] A processing module is used to select a target storage queue corresponding to the target image with an abnormal event captured by the target camera from all storage queues corresponding to the target camera, and determine whether to filter the target image based on the image filtering parameters corresponding to the target storage queue; if so, filter the target image.

[0014] The present application provides an abnormal event detection device, comprising: a processor and a machine-readable storage medium, wherein the machine-readable storage medium stores machine-executable instructions that can be executed by the processor; the processor is used to execute the machine-executable instructions to implement the abnormal event detection method disclosed in the above embodiment.

[0015] As can be seen from the above technical solutions, in the embodiment of the present application, for the target image with abnormal events captured by the target camera, the target storage queue corresponding to the target image can be selected, and the target image can be filtered based on the image filtering parameters corresponding to the target storage queue. If yes, the target image is filtered, that is, the target image is not sent to the management platform. If not, the target image is sent to the management platform, so that the truly valuable target images can be sent to the management platform, rather than a large number of target images with abnormal events being sent to the management platform, thereby avoiding the existence of a large number of redundant target images on the management platform. The management personnel can review a small number of target images, saving human resources and review time. In the above method, the image filtering parameters corresponding to the storage queue can be determined based on user preference information, and the target image can be filtered based on the image filtering parameters, thereby meeting the management personnel's review habits, sending the target images that meet the management personnel's review habits to the management platform, avoiding sending redundant target images to the management platform, facilitating the management personnel to quickly obtain the target images with abnormal events, effectively solving the problems of redundant review and a large number of invalid events, improving the accuracy of abnormal event push, reducing the amount of invalid review by the management personnel, avoiding the management personnel from frequently visiting the site for inspection, and providing the management personnel with a better user experience. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present application or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments of the present application or the description of the prior art. Obviously, the drawings described below are only some embodiments recorded in this application. For ordinary technicians in this field, other drawings can also be obtained based on these drawings of the embodiments of the present application.

[0017] Figure 1 This is a flow chart of an abnormal event detection method in one embodiment of the present application;

[0018] Figure 2 This is a schematic diagram of an application scenario in one embodiment of the present application;

[0019] Figure 3 is a schematic diagram of a filtering and recommendation system for abnormal events in one embodiment of the present application;

[0020] Figure 4 is a schematic diagram of a filtering and recommendation system for abnormal events in one embodiment of the present application;

[0021] Figure 5A This is a schematic diagram of detecting whether an abnormal event exists in one embodiment of the present application;

[0022] Figure 5B is a schematic diagram of extracting image features in one embodiment of the present application;

[0023] Figure 5C is a schematic diagram of the storage queues corresponding to each camera in one embodiment of the present application;

[0024] Figure 5D This is a schematic diagram of collecting user preference information in one embodiment of the present application;

[0025] Figure 6 is a schematic diagram of a sorting process of target images in one embodiment of the present application;

[0026] Figure 7 is a structural diagram of an abnormal event detection device in one embodiment of the present application;

[0027] Figure 8 This is a hardware structure diagram of an abnormal event detection device in one embodiment of the present application. DETAILED DESCRIPTION

[0028] The terms used in the embodiments of this application are only for the purpose of describing specific embodiments and are not intended to limit this application. The singular forms "a," "the," and "the" used in this application and claims are also intended to include plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to any or all possible combinations of one or more associated listed items.

[0029] It should be understood that although the terms first, second, third, etc. may be used to describe various information in the embodiments of the present application, such information should not be limited to these terms. These terms are only used to distinguish information of the same type from each other. For example, without departing from the scope of the present application, the first information may also be referred to as the second information, and similarly, the second information may also be referred to as the first information. Depending on the context, the word "if" used may also be interpreted as "at the time of" or "when" or "in response to determining".

[0030] In the embodiment of the present application, a method for detecting abnormal events is proposed. The method can be applied to abnormal event detection equipment. Figure 1 FIG. 5 is a flow chart of the method, which may include:

[0031] Step 101: Acquire at least one storage queue corresponding to a target camera, each storage queue including abnormal event images captured by the target camera; wherein abnormal event images in the same storage queue have similar features, and abnormal event images in different storage queues have dissimilar features.

[0032] Exemplarily, obtaining at least one storage queue corresponding to the target camera may include but is not limited to: after obtaining the abnormal event image captured by the target camera, if the similarity between the image features of the abnormal event image and the stored image features of the storage queue corresponding to the target camera is greater than a preset similarity threshold, then storing the abnormal event image in the storage queue; if the similarity between the image features of the abnormal event image and the stored image features of all storage queues corresponding to the target camera is not greater than the preset similarity threshold, then creating a new storage queue for the target camera, and storing the abnormal event image in the new storage queue.

[0033] Step 102: For each storage queue, determine user preference information corresponding to the storage queue, and determine image filtering parameters corresponding to the storage queue based on the user preference information.

[0034] Exemplarily, determining the image filtering parameter corresponding to the storage queue based on the user preference information may include, but is not limited to: if the user preference information is a focus preference, determining that the image filtering parameter corresponding to the storage queue is a first filtering probability value. If the user preference information is a non-focus preference, determining that the image filtering parameter corresponding to the storage queue is a second filtering probability value. The second filtering probability value may be greater than the first filtering probability value, and there is no restriction on the second filtering probability value and the first filtering probability value.

[0035] Exemplarily, determining the user preference information corresponding to the storage queue may include, but is not limited to: obtaining the configured preferences corresponding to the storage queue, and determining the user preference information corresponding to the storage queue based on the configured preferences. Wherein, if the configured preferences indicate paying attention to abnormal event images in the storage queue, then the user preference information corresponding to the storage queue is determined to be a pay attention preference; if the configured preferences indicate not paying attention to abnormal event images in the storage queue, then the user preference information corresponding to the storage queue is determined to be a don't pay attention preference. Alternatively, obtaining the review label corresponding to the storage queue, and determining the user preference information corresponding to the storage queue based on the review label. Wherein, if the review label indicates that the target image corresponding to the storage queue is detected as a positive image by the management platform, then the user preference information corresponding to the storage queue is determined to be a pay attention preference; if the review label indicates that the target image corresponding to the storage queue is detected as a false alarm image by the management platform, then the user preference information corresponding to the storage queue is determined to be a don't pay attention preference; if the review label indicates that the target image corresponding to the storage queue is not detected by the management platform, then the user preference information corresponding to the storage queue is determined to be a don't pay attention preference.

[0036] Step 103 : For a target image with an abnormal event captured by a target camera, a target storage queue corresponding to the target image is selected from all storage queues corresponding to the target camera.

[0037] Step 104: Determine whether to filter the target image based on the image filtering parameters corresponding to the target storage queue. If yes, proceed to step 105; if not, proceed to step 106.

[0038] Exemplarily, determining whether to filter the target image based on the image filtering parameter corresponding to the target storage queue may include, but is not limited to: if the image filtering parameter corresponding to the target storage queue is a first filtering probability value, generating a first numerical interval and a second numerical interval based on the first filtering probability value; generating a first random number for the target image, and if the first random number matches the first numerical interval, filtering the target image; if the first random number matches the second numerical interval, not filtering the target image. Alternatively, if the image filtering parameter corresponding to the target storage queue is a second filtering probability value, generating a third numerical interval and a fourth numerical interval based on the second filtering probability value; generating a second random number for the target image, and if the second random number matches the third numerical interval, filtering the target image; if the second random number matches the fourth numerical interval, not filtering the target image. The length of the third numerical interval may be greater than the length of the first numerical interval; and the length of the fourth numerical interval may be less than the length of the second numerical interval.

[0039] Step 105: Filter the target image, that is, prohibit the target image from being sent to the management platform.

[0040] Step 106: Send the target image to the management platform, which performs abnormal event detection based on the target image. There is no restriction on the abnormal event detection process.

[0041] Exemplarily, if the number of target images remaining after filtering is multiple, sending the target image to the management platform (i.e., sending multiple target images to the management platform) may include but is not limited to: for each target image, determining the sending order of the target image based on the image features of the target image, and sending multiple target images to the management platform in sequence based on the sending order of each target image.

[0042] For example, if the similarity between the image features of the target image and the image features of the positive image is greater than the first similarity threshold, the target image is stored in the first sending queue; if the similarity between the image features of the target image and the image features of the false alarm image is greater than the second similarity threshold, the target image is stored in the second sending queue; if the similarity between the image features of the target image and the image features of the positive image is not greater than the first similarity threshold, and the similarity between the image features of the target image and the image features of the false alarm image is not greater than the second similarity threshold, the target image is stored in the third sending queue; on this basis, the target images in the first sending queue are first sent to the management platform, and then the target images in the third sending queue are sent to the management platform, and then the target images in the second sending queue are sent to the management platform.

[0043] As can be seen from the above technical solutions, in the embodiment of the present application, for the target image with abnormal events captured by the target camera, the target storage queue corresponding to the target image can be selected, and the target image can be filtered based on the image filtering parameters corresponding to the target storage queue. If yes, the target image is filtered, that is, the target image is not sent to the management platform. If not, the target image is sent to the management platform, so that the truly valuable target images can be sent to the management platform, rather than a large number of target images with abnormal events being sent to the management platform, thereby avoiding the existence of a large number of redundant target images on the management platform. The management personnel can review a small number of target images, saving human resources and review time. In the above method, the image filtering parameters corresponding to the storage queue can be determined based on user preference information, and the target image can be filtered based on the image filtering parameters, thereby meeting the management personnel's review habits, sending the target images that meet the management personnel's review habits to the management platform, avoiding sending redundant target images to the management platform, facilitating the management personnel to quickly obtain the target images with abnormal events, effectively solving the problems of redundant review and a large number of invalid events, improving the accuracy of abnormal event push, reducing the amount of invalid review by the management personnel, avoiding the management personnel from frequently visiting the site for inspection, and providing the management personnel with a better user experience.

[0044] The following describes the technical solutions of the embodiments of the present application in conjunction with specific application scenarios.

[0045] See also Figure 2 As shown, multiple cameras (such as network cameras or analog cameras, etc.) can be deployed in the target scene (i.e., the area to be managed). These cameras can capture the initial image of the target scene and send the initial image to the abnormal event detection device. Each camera can be called a point.

[0046] The abnormal event detection device can be a backend device (corresponding to a frontend device such as a camera), such as a storage device, and there is no restriction on the type of the abnormal event detection device. After receiving the initial image sent by the camera, the abnormal event detection device can store the initial image in a storage medium.

[0047] After receiving the initial image, the abnormal event detection device can also detect whether there are abnormal events in the initial image, such as illegal stalls, illegal parking, fighting, and other abnormal events. This embodiment does not limit the type of this abnormal event, which can be various phenomena or behaviors caused by human or natural factors.

[0048] For example, the abnormal event detection device includes a trained abnormal event detection model (i.e., a machine learning model, such as a deep learning model, a neural network model, etc., with no limitation on the type of abnormal event detection model). The abnormal event detection model is used to detect whether an abnormal event exists in an initial image. For example, the abnormal event detection device may input the initial image into the abnormal event detection model, which then performs abnormal event detection on the initial image to obtain a detection result indicating whether an abnormal event exists.

[0049] If the detection result determines that the initial image contains an abnormal event, the abnormal event detection device may send the initial image to the management platform. If the detection result determines that the initial image contains no abnormal event, the abnormal event detection device will not send the initial image to the management platform.

[0050] The management platform can be a device used by users (e.g., managers or staff), such as a personal computer, server, laptop, or smart terminal. There are no restrictions on the type of management platform. After receiving the initial image sent by the abnormal event detection device, the management platform displays the initial image to the user, who then views the initial image on the management platform, reviews it, and confirms whether an abnormal event has occurred. The user can then visit the site (i.e., where the camera is located) to verify whether an abnormal event has occurred.

[0051] However, in the above method, the abnormal event detection device sends all initial images with abnormal events to the management platform, and may send a large number of initial images with abnormal events to the management platform, while users usually only review a small number of initial images, resulting in a large number of redundant initial images. It is difficult for users to find truly valuable initial images from a large number of initial images for review, which causes users to frequently go to the site to check, which requires a lot of human resources, consumes a lot of time, and wastes workload.

[0052] In response to the above findings, an embodiment of the present application proposes an abnormal event detection method that can learn user review habits. By analyzing user review habits, abnormal event images can be filtered, and abnormal event images that match the user's review habits can be sent to the management platform, so that truly valuable images can be sent to the management platform, avoiding sending a large number of redundant images to the management platform, making it convenient for users to quickly obtain images with abnormal events, effectively solving problems such as review redundancy and a large number of invalid events, improving the accuracy of abnormal event push, reducing the amount of invalid reviews by users, and avoiding users from frequently visiting the site to check.

[0053] In the embodiment of the present application, a filtering and recommendation system for abnormal events is proposed. Figure 3 Figure 2 shows a schematic diagram of a filtering and recommendation system for abnormal events. The filtering and recommendation system may include a user model management module, an event model management module, and a recommendation module. The user model management module may collect user preference information (such as configured preferences proposed by users based on their own needs, or audit labels implicitly analyzed through the collection of user audit data), determine image filtering parameters based on the user preference information, and send the image filtering parameters to the recommendation module. The recommendation module then filters the images based on the image filtering parameters and sends the remaining images to the management platform, thus avoiding sending a large number of images to the management platform.

[0054] The event model management module can capture images from cameras and detect whether any abnormal events are present. For images with abnormal events, image features can be extracted and sent to the recommendation module, which then filters the images based on these features and sends the remaining images to the management platform, thus avoiding the need to send a large number of images to the platform.

[0055] For the recommendation module, the recommendation module can obtain image filtering parameters from the user model management module and image features from the event model management module, and filter images with abnormal events based on the image filtering parameters and image features, and send images that meet the user's review habits to the management platform, that is, send truly valuable images to the management platform, avoid sending a large number of redundant images to the management platform, reduce the amount of invalid reviews by users, and effectively solve problems such as redundant reviews and a large number of invalid events.

[0056] See also Figure 4 Figure 2 shows another schematic diagram of a filtering and recommendation system for abnormal events. The filtering and recommendation system may include an event model management module, a user model management module, and a recommendation module. The event model management module may include an event detection module and a feature extraction module. The user model management module may include a user preference information collection module and a user habit management module. The recommendation module may include a filtering module and a ranking module.

[0057] Of course, the above modules are just examples for the convenience of description in the embodiments of this application. There is no limitation on these modules as long as the functions of these modules can be realized. The functions of the above modules are explained below.

[0058] First, the event detection module. Multiple cameras can be deployed in the target scene (each camera can be called a point). These cameras can capture the initial image of the target scene and send the initial image to the event detection module. For each initial image, the event detection module can detect whether there are abnormal events in the initial image, such as illegal street stalls, illegal parking, fighting, and other abnormal events. There is no restriction on the type of abnormal event. For initial images with abnormal events, they can be referred to as abnormal event images. The event detection module can mark the abnormal area on the abnormal event image (i.e., the area where the abnormal event is located, such as a marked rectangular box) and give the category of the abnormal event, such as illegal street stalls, illegal parking, fighting, etc., and send the abnormal event image to the feature extraction module, and the subsequent modules will analyze and process based on the abnormal event image. For initial images without abnormal events, the event detection module will not send the initial image to the feature extraction module, and the subsequent modules will not analyze and process based on the initial image.

[0059] For example, the event detection module can use the trained abnormal event detection model to detect whether there is an abnormal event in the initial image. Figure 5A Figure 1 shows a schematic diagram of detecting whether an initial image contains an abnormal event. For each initial image, the event detection module can input the initial image into the abnormal event detection model. The abnormal event detection model then performs abnormal event detection on the initial image to obtain a detection result indicating whether an abnormal event exists. For initial images containing abnormal events, the abnormal event detection model can mark the abnormal area in the abnormal event image and assign the abnormal event category.

[0060] The abnormal event detection model is a model trained using machine learning techniques, and there are no restrictions on the training process of this abnormal event detection model. The abnormal event detection model can be a machine learning model, such as a deep learning model, a CNN (Convolutional Neural Network) model, a Transformer (Fully Self-Attention Network) model, etc. There are no restrictions on the type of abnormal event detection model, as long as the abnormal event detection model can detect whether an abnormal event exists in the initial image.

[0061] Second, the feature extraction module. After obtaining the abnormal event image (i.e., the initial image containing the abnormal event), the feature extraction module obtains the image features corresponding to the abnormal event image, such as the image features corresponding to the abnormal area in the abnormal event image, and stores the abnormal event image in the storage queue based on the image features.

[0062] For example, the feature extraction module can use a trained feature extraction model to extract image features from abnormal event images. Figure 5B FIG2 is a schematic diagram of extracting image features from an abnormal event image. For each abnormal event image, the feature extraction module can input the abnormal event image into the feature extraction model, and the feature extraction model extracts image features from the abnormal area of ​​the abnormal event image.

[0063] The feature extraction model is a model trained using machine learning techniques. There are no restrictions on the training process for this feature extraction model. The feature extraction model can be a machine learning model, such as a deep learning model or a CNN model. There are no restrictions on the type of feature extraction model, as long as it can extract image features from abnormal event images. For example, if the feature extraction model is a CNN model, the CNN model can be a general-purpose model, such as Resnet50 trained using ImageNet.

[0064] Exemplarily, a set of storage queues can be maintained for each camera, with at least one storage queue. Each storage queue includes abnormal event images captured by the camera. Abnormal event images in the same storage queue have similar features, and abnormal event images in different storage queues have dissimilar features.

[0065] Exemplarily, the feature extraction module uses a feature extraction model to extract image features from abnormal event images, which is used to store abnormal event images with similar features in the same storage queue, and store abnormal event images with dissimilar features in different storage queues, thereby ensuring that abnormal event images of the same type are sufficiently close in the feature space, and ensuring that abnormal event images of different types are sufficiently far apart in the feature space. In this way, when constructing storage queues, abnormal event images with similar features can be stored in the same storage queue, and abnormal event images with dissimilar features can be stored in different storage queues.

[0066] In one possible implementation, after obtaining an abnormal event image captured by a camera, if the camera does not have a corresponding storage queue, a storage queue is created for the camera and the abnormal event image is stored in the storage queue. If the camera has a corresponding storage queue, and the similarity between the image features of the abnormal event image and the image features stored in the storage queue corresponding to the camera is greater than a preset similarity threshold, the abnormal event image is stored in the storage queue. If the similarity between the image features of the abnormal event image and the image features stored in all storage queues corresponding to the camera is not greater than the preset similarity threshold, a new storage queue is created for the camera and the abnormal event image is stored in the new storage queue.

[0067] See also Figure 5C , which is a schematic diagram of the storage queues corresponding to each camera. Camera 1 corresponds to storage queue 11 , storage queue 12 and storage queue 13 , and camera 2 corresponds to storage queue 21 and storage queue 22 .

[0068] After obtaining the abnormal event image s captured by camera 2, the similarity between the image features of the abnormal event image s (i.e., the image features obtained by the feature extraction model) and the image features stored in the storage queue 21 is calculated. If the similarity is greater than a preset similarity threshold (which can be configured based on experience and is not limited to this preset similarity threshold), the abnormal event image s can be stored in the storage queue 21.

[0069] Among them, the stored image features can be the image features corresponding to any abnormal event image in the storage queue 21, or the average value of the image features corresponding to all abnormal event images in the storage queue 21, or the maximum value of the image features corresponding to all abnormal event images in the storage queue 21, or the minimum value of the image features corresponding to all abnormal event images in the storage queue 21, and there is no restriction on this.

[0070] Among them, after the abnormal event image s is stored in the storage queue 21, the stored image features of the storage queue 21 can also be updated, for example, the average value (or maximum value, minimum value) of the image features corresponding to all abnormal event images can be recalculated, and the recalculated average value can be used as the updated stored image features.

[0071] If the similarity between the image features of abnormal event image s and the features of images stored in storage queue 21 is not greater than a preset similarity threshold, the similarity between the image features of abnormal event image s and the features of images stored in storage queue 22 is calculated. If the similarity is greater than the preset similarity threshold, abnormal event image s may be stored in storage queue 22. If the similarity is not greater than the preset similarity threshold, a new storage queue 23 is created for camera 2, and abnormal event image s is stored in storage queue 23.

[0072] To sum up, based on the similarity between abnormal event images, similar abnormal event images can be classified into one storage queue, and dissimilar abnormal event images can be classified into different storage queues. That is to say, the abnormal event images in a single storage queue have similar image features and are all abnormal event images captured by the same camera, which ensures that the similarity of the abnormal event images in a single storage queue is high enough.

[0073] Third, a user preference information collection module. The user preference information collection module is used to collect user preference information corresponding to the storage queue. The user preference information can be a focus preference or a non-focus preference. For example, if the user preference information corresponding to the storage queue is a focus preference, it means that the abnormal event images in the storage queue are in line with the user's review habits. If the user preference information corresponding to the storage queue is a non-focus preference, it means that the abnormal event images in the storage queue are not in line with the user's review habits. If the storage queue does not have corresponding user preference information, it is impossible to know whether the abnormal event images in the storage queue are in line with the user's review habits.

[0074] For example, to collect user preference information corresponding to the storage queue, see Figure 5D As shown in FIG, a schematic diagram of collecting user preference information is provided. User preference information can be collected in the following situations:

[0075] Case 1: Based on the user's needs, the configured preferences are displayed. The user preference information collection module obtains the configured preferences corresponding to the storage queue and determines the user preference information corresponding to the storage queue based on the configured preferences. For example, if the configured preference indicates paying attention to abnormal event images in the storage queue, the user preference information corresponding to the storage queue is determined to be a pay attention preference; if the configured preference indicates not paying attention to abnormal event images in the storage queue, the user preference information corresponding to the storage queue is determined to be a ignore preference.

[0076] For example, when a user is interested in abnormal event images of a specified category (such as illegal parking), he can express his own needs and indicate that he is interested in abnormal event images of the specified category, that is, the configured preference indicates that he is interested in abnormal event images in the storage queue. Based on this, the user preference information collection module can obtain the storage queue corresponding to the specified category, see Figure 5CAs shown, assuming that the storage queue 11 and the storage queue 22 correspond to the specified category (that is, the abnormal event category of the abnormal event images in the storage queue 11 and the storage queue 22 is the specified category), the user preference information collection module determines that the configured preference corresponding to the storage queue 11 indicates attention to the abnormal event images in the storage queue 11, and determines that the user preference information corresponding to the storage queue 11 is a attention preference, determines that the configured preference corresponding to the storage queue 22 indicates attention to the abnormal event images in the storage queue 22, and determines that the user preference information corresponding to the storage queue 22 is a attention preference.

[0077] For another example, when a user does not pay attention to abnormal event images of a specified category, they can display their own needs, indicating that they do not pay attention to abnormal event images of the specified category, that is, the configured preference indicates that they do not pay attention to abnormal event images in the storage queue. On this basis, the user preference information collection module can obtain the storage queues corresponding to the specified category, such as storage queue 11 and storage queue 22, determine that the configured preference corresponding to storage queue 11 indicates that they do not pay attention to abnormal event images in storage queue 11, that is, the user preference information corresponding to storage queue 11 is a "no attention preference", and determine that the configured preference corresponding to storage queue 22 indicates that they do not pay attention to abnormal event images in storage queue 22, that is, the user preference information corresponding to storage queue 22 is a "no attention preference".

[0078] For another example, when a user pays attention to an abnormal event image in the storage queue 21, he or she can express his or her own needs, indicating that he or she is paying attention to the abnormal event image in the storage queue 21. On this basis, the user preference information collection module can determine that the configured preference corresponding to the storage queue 21 indicates paying attention to the abnormal event image in the storage queue 21, that is, the user preference information corresponding to the storage queue 21 is the attention preference.

[0079] For another example, when a user does not pay attention to the abnormal event images in the storage queue 22, he or she can display his or her own needs, indicating that he or she does not pay attention to the abnormal event images in the storage queue 22. On this basis, the user preference information collection module can determine that the configured preference corresponding to the storage queue 22 indicates that he or she does not pay attention to the abnormal event images in the storage queue 22, that is, the user preference information corresponding to the storage queue 22 is a do not pay attention preference.

[0080] Of course, the above are just a few examples of configured preferences. There is no limitation on the configured preferences. As long as the configured preferences can be displayed according to the user's own needs, they will not be repeated here.

[0081] Case 2: By collecting user audit data and implicitly analyzing the audit label, the user preference information is determined based on the audit label. That is, the user preference information collection module obtains the audit label corresponding to the storage queue, and determines the user preference information corresponding to the storage queue based on the audit label. For example, if the audit label indicates that the target image corresponding to the storage queue is detected as a positive image by the management platform, then it can be determined that the user preference information corresponding to the storage queue is a focus preference; if the audit label indicates that the target image corresponding to the storage queue is detected as a false alarm image by the management platform, then it can be determined that the user preference information corresponding to the storage queue is a non-focus preference; if the audit label indicates that the target image corresponding to the storage queue is not detected by the management platform, then it can be determined that the user preference information corresponding to the storage queue is a non-focus preference.

[0082] For example, for multiple abnormal event images in the storage queue, the multiple abnormal event images in the storage queue can be sent to the management platform. The specific sending process is described in the subsequent embodiments. The abnormal event images sent to the management platform are recorded as target images. After the multiple target images in the storage queue are sent to the management platform, the processing methods for the target images on the management platform include:

[0083] For a target image, a user may review the target image on the management platform to verify whether the abnormal event corresponding to the target image actually exists, which means that the target image has been detected by the management platform. Alternatively, a user may not review the target image to verify whether the abnormal event corresponding to the target image actually exists, which means that the target image has not been detected by the management platform.

[0084] When the user verifies whether the abnormal event corresponding to the target image actually exists, they can visit the site to verify whether the abnormal event actually exists. There are no restrictions on the implementation method. After the verification is completed, if the abnormal event corresponding to the target image actually exists, the target image is marked as a positive image, that is, the target image is detected as a positive image by the management platform. If the abnormal event corresponding to the target image does not exist, the target image is marked as a false alarm image, that is, the target image is detected as a false alarm image by the management platform.

[0085] For a certain storage queue 11, after sending multiple target images corresponding to the storage queue 11 to the management platform, if the user reviews some of the target images on the management platform, and the review result is that the abnormal event corresponding to at least one target image actually exists, that is, at least one target image is detected as a positive image by the management platform, then the user preference information collection module obtains the review label corresponding to the storage queue 11, and the review label indicates that at least one target image corresponding to the storage queue 11 is detected as a positive image by the management platform. Therefore, it is determined that the user preference information corresponding to the storage queue 11 is a focus preference.

[0086] For the storage queue 12, after sending multiple target images corresponding to the storage queue 12 to the management platform, if the user reviews some of the target images on the management platform, and the review result is that the abnormal events corresponding to more than k target images do not exist, that is, more than k target images are detected as false alarm images by the management platform, then the review label corresponding to the storage queue 12 indicates that the target images corresponding to the storage queue 12 are detected as false alarm images by the management platform, and the user preference information corresponding to the storage queue 12 is determined to be a no-pay attention preference. In summary, if there are more than k false alarms, the user preference information is a no-pay attention preference. The value of k can be configured based on experience, such as the value of k is 2, 3, etc., and there is no restriction on this.

[0087] For the storage queue 13, after sending the multiple target images corresponding to the storage queue 13 to the management platform, if the user does not review the m consecutive target images corresponding to the storage queue 13 on the management platform, that is, the m consecutive target images corresponding to the storage queue 13 are not detected by the management platform, then the review label corresponding to the storage queue 13 indicates that the target image corresponding to the storage queue 13 is not detected by the management platform. Therefore, it can be determined that the user preference information corresponding to the storage queue 13 is a no-attention preference. In summary, if there are m (or more than m) consecutive target images that are not detected, the user preference information is a no-attention preference. The value of m can be configured based on experience, such as 20, 30, etc., and there is no restriction on this.

[0088] Of course, the above are just a few examples of review labels, and there is no limitation to this. As long as the review labels can be implicitly analyzed by collecting user review data and the user preference information can be determined based on the review labels, it will be sufficient.

[0089] Case 3: If the storage queue corresponds to a configured preference and the storage queue corresponds to an audit tag, then the user preference information corresponding to the storage queue is determined based on the configured preference and the audit tag.

[0090] For example, if the configured preference indicates paying attention to abnormal event images in the storage queue, and the review label indicates that the target image corresponding to the storage queue is detected as a positive alarm image by the management platform, then the user preference information is a pay attention preference. If the configured preference indicates paying attention to abnormal event images in the storage queue, and the review label indicates that the target image corresponding to the storage queue is detected as a false alarm image by the management platform, then the user preference information is a pay attention preference. If the configured preference indicates paying attention to abnormal event images in the storage queue, and the review label indicates that the target image corresponding to the storage queue has not been detected by the management platform, then the user preference information is a pay attention preference. If the configured preference indicates not paying attention to abnormal event images in the storage queue, and the review label indicates that the target image corresponding to the storage queue is detected as a positive alarm image by the management platform, then the user preference information is a don't pay attention preference. If the configured preference indicates not paying attention to abnormal event images in the storage queue, and the review label indicates that the target image corresponding to the storage queue is detected as a false alarm image by the management platform, then the user preference information is a don't pay attention preference. If the configured preference indicates not paying attention to abnormal event images in the storage queue, and the review label indicates that the target image corresponding to the storage queue has not been detected by the management platform, then the user preference information is a don't pay attention preference.

[0091] Case 4: If the storage queue does not correspond to the configured preference and the storage queue does not correspond to the review label, then the storage queue has no corresponding user preference information, that is, the user preference information is neither a pay attention preference nor a don't pay attention preference, and it is impossible to know whether the abnormal event images in the storage queue comply with the user's review habits.

[0092] In summary, for each storage queue, the user preference information collection module may collect the user preference information corresponding to the storage queue. The user preference information may be an attention preference or a non-attention preference.

[0093] Fourth, the user habit management module. For each storage queue, the user habit management module determines the image filtering parameters corresponding to the storage queue based on the user preference information corresponding to the storage queue. The image filtering parameters can be filtering probability values, indicating that the abnormal event images corresponding to the storage queue are filtered using the filtering probability value. The remaining images are sent to the management platform, and the filtered images are prohibited from being sent.

[0094] For example, for a certain storage queue, if the user preference information corresponding to the storage queue is a focus preference, it means that the abnormal event images corresponding to the storage queue meet the user's review habits, and the abnormal event images corresponding to the storage queue should be sent to the management platform. Based on this, it is determined that the image filtering parameter corresponding to the storage queue is a first filtering probability value. The first filtering probability value can be a relatively small probability value, indicating that only a small number of abnormal event images corresponding to the storage queue are filtered, and a large number of abnormal event images corresponding to the storage queue are sent to the management platform. For example, the first filtering probability value can be 5%, 2%, 1%, etc., or even the first filtering probability value can be 0, indicating that the abnormal event images corresponding to the storage queue are not filtered, and all abnormal event images corresponding to the storage queue are sent to the management platform.

[0095] For a certain storage queue, if the user preference information corresponding to the storage queue is a "no attention preference," it indicates that the abnormal event images corresponding to the storage queue do not conform to the user's audit habits and should not be sent to the management platform. Based on this, the image filtering parameter corresponding to the storage queue is determined to be a second filtering probability value. The second filtering probability value can be a relatively large probability value, indicating that a large number of abnormal event images corresponding to the storage queue are filtered, and a small number of abnormal event images corresponding to the storage queue are sent to the management platform. For example, the second filtering probability value can be 95%, 98%, 99%, etc., or even 100%, indicating that all abnormal event images corresponding to the storage queue are filtered, and each abnormal event image corresponding to the storage queue is prohibited from being sent to the management platform.

[0096] From the above, it can be seen that the second filtering probability value is a relatively large probability value, while the first filtering probability value is a relatively small probability value. Therefore, the second filtering probability value can be greater than the first filtering probability value. There is no restriction on the second filtering probability value and the first filtering probability value, as long as the second filtering probability value is greater than the first filtering probability value, such as the second filtering probability value is 99% and the first filtering probability value is 1%.

[0097] For a certain storage queue, the storage queue may not have corresponding user preference information, that is, the user preference information is neither an attention preference nor an unattention preference, and it is impossible to know whether the abnormal event images in the storage queue comply with the user's review habits. In this case, it can be determined that the image filtering parameter corresponding to the storage queue is the third filtering probability value. The third filtering probability value can be a relatively small probability value, indicating that only a small number of abnormal event images corresponding to the storage queue are filtered, and a large number of abnormal event images corresponding to the storage queue are sent to the management platform, and the user decides whether to review the abnormal event images.

[0098] Among them, the third filtering probability value can be equal to the first filtering probability value, and the second filtering probability value can be greater than the third filtering probability value. For example, the third filtering probability value can be 5%, 2%, 1%, etc., and even the third filtering probability value can be 0, indicating that the abnormal event image corresponding to the storage queue is not filtered, and all abnormal event images corresponding to the storage queue are sent to the management platform without restriction.

[0099] Alternatively, the third filtering probability value can be greater than the first filtering probability value, and the second filtering probability value can be greater than the third filtering probability value. For example, the third filtering probability value is 50%, 20%, 10%, etc. There is no restriction on the third filtering probability value, and the third filtering probability value can be between the first filtering probability value and the second filtering probability value.

[0100] To sum up, for each storage queue, the user habit management module can determine the image filtering parameter corresponding to the storage queue. The image filtering parameter can be the first filtering probability value, or the second filtering probability value, or the third filtering probability value, indicating that the filtering probability value is used to filter the abnormal event image.

[0101] In one possible implementation, since user review habits are constantly changing, that is, review labels and configured preferences are constantly updated, the user preference information corresponding to each storage queue can be periodically determined, and the image filtering parameters corresponding to each storage queue can be periodically determined, that is, the user preference information and image filtering parameters corresponding to each storage queue will be continuously updated as user review habits change.

[0102] In a possible implementation, each storage queue needs to have a continuous updating mechanism. For example, each storage queue only retains abnormal event images within a specific time period (such as 3 months or 6 months).

[0103] Fifth, the filtering module. After the event detection module obtains the abnormal event image, in addition to sending it to the feature extraction module, which then stores it in a storage queue, the event detection module can also send it to the filtering module. Alternatively, after storing it in a storage queue, the feature extraction module can also send it to the filtering module. In summary, the filtering module can obtain an abnormal event image, that is, an image containing an abnormal event. For ease of distinction, this image containing an abnormal event is referred to as the target image.

[0104] For each target image, the filtering module can select the target storage queue corresponding to the target image (i.e., the storage queue for storing the target image) from all storage queues corresponding to the target camera (i.e., the camera used to capture the target image). Based on the image filtering parameters corresponding to the target storage queue, it is determined whether to filter the target image. If so, the target image is filtered, that is, the target image is prohibited from being sent to the management platform. If not, the target image is sent to the management platform. When the target image is sent to the management platform, an abnormal area (such as a rectangular box) is marked on the target image, and the category of the abnormal event is given.

[0105] For example, if the image filtering parameter corresponding to the target storage queue is a first filtering probability value, and the first filtering probability value is a relatively small probability value, indicating that a large number of abnormal event images are sent to the management platform, such as 2%, then a first numerical interval and a second numerical interval are generated based on the first filtering probability value. The length of the first numerical interval is less than the length of the second numerical interval, and the ratio of the length of the first numerical interval to the total length (i.e., the sum of the length of the first numerical interval and the length of the second numerical interval) is the first filtering probability value, such as the first numerical interval is [1, 2] and the second numerical interval is [3, 100]. On this basis, for each target image, a first random number, such as a positive integer between 1 and 100, can be generated for the target image. If the first random number matches the first numerical interval, i.e., the first random number is in the first numerical interval [1, 2], it indicates that the target image needs to be filtered. If the first random number matches the second numerical interval, i.e., the first random number is in the second numerical interval [3, 100], it indicates that the target image does not need to be filtered. In summary, if the image filtering parameter corresponding to the target storage queue is the first filtering probability value, the target image is sent to the management platform with a higher probability.

[0106] If the image filtering parameter corresponding to the target storage queue is the second filtering probability value, and the second filtering probability value is a relatively large probability value, indicating that a small number of abnormal event images will be sent to the management platform, such as 98%, then a third numerical interval and a fourth numerical interval are generated based on the second filtering probability value. The length of the third numerical interval is greater than the length of the fourth numerical interval, and the ratio of the length of the third numerical interval to the total length (i.e., the sum of the length of the third numerical interval and the length of the fourth numerical interval) is the second filtering probability value. The length of the third numerical interval is greater than the length of the first numerical interval, and the length of the fourth numerical interval is less than the length of the second numerical interval. For example, the third numerical interval is [1, 98] and the fourth numerical interval is [99, 100]. On this basis, for each target image, a second random number is generated for the target image, such as a positive integer between 1 and 100. If the second random number matches the third numerical interval, that is, the second random number is in the third numerical interval [1, 98], it indicates that the target image needs to be filtered. If the second random number matches the fourth numerical interval, that is, the second random number is in the fourth numerical interval [99, 100], it indicates that the target image does not need to be filtered. In summary, if the image filtering parameter corresponding to the target storage queue is the second filtering probability value, the target image is sent to the management platform with a lower probability.

[0107] If the image filtering parameter corresponding to the target storage queue is a third filtering probability value, such as 40%, a fifth numerical interval and a sixth numerical interval are generated based on the third filtering probability value. The ratio of the length of the fifth numerical interval to the total length (i.e., the sum of the length of the fifth numerical interval and the length of the sixth numerical interval) is the third filtering probability value. The length of the fifth numerical interval is greater than the length of the first numerical interval and less than the length of the third numerical interval. The length of the sixth numerical interval is less than the length of the second numerical interval and greater than the length of the fourth numerical interval. For example, the fifth numerical interval is [1, 40] and the sixth numerical interval is [41, 100]. On this basis, for each target image, a third random number, such as a positive integer between 1 and 100, can be generated for the target image. If the third random number matches the fifth numerical interval, that is, the third random number is in the fifth numerical interval [1, 40], it indicates that the target image needs to be filtered. If the third random number matches the sixth numerical interval, that is, the third random number is in the sixth numerical interval [41, 100], it indicates that the target image does not need to be filtered.

[0108] In summary, if the image filtering parameter corresponding to the target storage queue is the first filtering probability value, the filtering module uses a higher probability to send the target image corresponding to the target storage queue to the management platform, so that the target image sent to the management platform conforms to the user's review habits and reduces the amount of invalid review. If the image filtering parameter corresponding to the target storage queue is the second filtering probability value, the filtering module uses a lower probability to send the target image corresponding to the target storage queue to the management platform, reducing the number of images sent and the amount of user review.

[0109] Sixth, the sorting module. After using the filtering module to filter the target images in all storage queues (i.e., the target objects have been filtered using the filtering probability value), the number of target images remaining after filtering may be multiple. These target images are already target images that need to be sent to the management platform. Different target images can correspond to target images in different storage queues or target images in the same storage queue. On this basis, when sending multiple target images to the management platform, for each target image, the sorting module can determine the sending order of the target image based on the image features of the target image. Based on the sending order of each target image, the sorting module can send the multiple target images to the management platform in sequence.

[0110] Exemplarily, for each target image, if the similarity between the image features of the target image and the image features of the positive image is greater than a first similarity threshold (which can be configured based on experience), the sorting module stores the target image in the first sending queue. If the similarity between the image features of the target image and the image features of the false alarm image is greater than a second similarity threshold (which can be configured based on experience), the sorting module stores the target image in the second sending queue. If the similarity between the image features of the target image and the image features of the positive image is not greater than the first similarity threshold, and the similarity between the image features of the target image and the image features of the false alarm image is not greater than the second similarity threshold, the sorting module stores the target image in the third sending queue. On this basis, the sorting module can first send the target images in the first sending queue to the management platform. After the sending is completed, the sorting module then sends the target images in the third sending queue to the management platform. After the sending is completed, the sorting module then sends the target images in the second sending queue to the management platform.

[0111] For example, for the remaining target images after filtering, the sorting module can also sort the target images according to the acquisition time corresponding to the target image (i.e., the time when the abnormal event occurred) and the user's preference. Figure 6 FIG2 is a schematic diagram of the target image sorting process. The sorting module can maintain a first sending queue (also called a favorite pool queue), a third sending queue (also called a general pool queue), and a second sending queue (also called a dislike pool queue). The first sending queue is used to store target images that are highly favored by users, and the second sending queue is used to store target images that are less favored by users.

[0112] For each target image remaining after filtering, if the similarity between the image features of the target image and the image features of the positive image is greater than a first similarity threshold, the sorting module stores the target image in the first sending queue. Referring to the above embodiment, when the user verifies whether the abnormal event corresponding to the target image actually exists, if the abnormal event corresponding to the target image actually exists, the target image is a positive image, and the sorting module can obtain the image features of the positive image. When the first sending queue includes multiple target images, these target images can be sorted according to the similarity between the image features of the target image and the image features of the positive image. The greater the similarity, the higher the position of the target image in the first sending queue.

[0113] For each target image remaining after filtering, if the similarity between the image features of the target image and the image features of the false alarm image is greater than a second similarity threshold, the sorting module stores the target image in the second sending queue. Referring to the above embodiment, when the user verifies whether the abnormal event corresponding to the target image actually exists, if the abnormal event corresponding to the target image does not exist, the target image is a false alarm image, and the sorting module can obtain the image features of the false alarm image. When the second sending queue includes multiple target images, these target images can be sorted according to the similarity between the image features of the target image and the image features of the false alarm image. The greater the similarity, the later the target image is positioned in the second sending queue.

[0114] For each target image remaining after filtering, if the similarity between the target image's image features and those of the positive image is no greater than a first similarity threshold, and the similarity between the target image's image features and those of the false positive image is no greater than a second similarity threshold, then the target image has no apparent preference, and the sorting module stores the target image in the third sending queue. If the third sending queue includes multiple target images, the target images may be sorted according to their corresponding acquisition times. For example, target images with earlier acquisition times are positioned higher in the second sending queue.

[0115] In summary, see Figure 6 As shown, the sorting module can first send the target images in the first sending queue to the management platform. When sending multiple target images in the first sending queue, the target images are sent sequentially in the order of these target images. After the target images in the first sending queue are sent, the sorting module sends the target images in the third sending queue to the management platform. When sending multiple target images in the third sending queue, the target images are sent sequentially in the order of these target images. After the target images in the third sending queue are sent, the sorting module sends the target images in the second sending queue to the management platform. When sending multiple target images in the second sending queue, the target images are sent sequentially in the order of these target images.

[0116] Through the above-mentioned sending method, users can process the target images they need (i.e., the target images with higher ranking) in a timely manner. For target images with lower ranking, users can selectively review them according to their own circumstances, thereby reducing the review workload, improving review efficiency, and providing users with a better user experience.

[0117] As can be seen from the above technical solutions, in the embodiment of the present application, it is possible to send truly valuable target images to the management platform, rather than sending a large number of target images with abnormal events to the management platform, thereby avoiding the existence of a large number of redundant target images on the management platform. Users can review a small number of target images, saving human resources and review time. The image filtering parameters corresponding to the storage queue can be determined based on user preference information, and whether to filter the target image can be determined based on the image filtering parameters, thereby meeting the user's review habits, sending target images that meet the user's review habits to the management platform, avoiding sending redundant target images to the management platform, and facilitating users to quickly obtain target images with abnormal events. This can effectively solve problems such as redundant review and a large number of invalid events, improve the accuracy of abnormal event push, reduce the amount of invalid review by users, avoid users from frequently visiting the site for review, and provide users with a better user experience.

[0118] Based on the same application concept as the above method, an abnormal event detection device is proposed in the embodiment of the present application, see Figure 7 FIG. 1 is a schematic diagram of the structure of the device, which may include:

[0119] An acquisition module 71 is configured to acquire at least one storage queue corresponding to a target camera, each storage queue including abnormal event images captured by the target camera; wherein abnormal event images in the same storage queue have similar features, and abnormal event images in different storage queues have dissimilar features;

[0120] a determination module 72 for determining, for each storage queue, user preference information corresponding to the storage queue, and determining image filtering parameters corresponding to the storage queue based on the user preference information;

[0121] The processing module 73 is used to select a target storage queue corresponding to the target image with an abnormal event captured by the target camera from all storage queues corresponding to the target camera, and determine whether to filter the target image based on the image filtering parameters corresponding to the target storage queue; if so, filter the target image; if not, send the target image to the management platform so that the management platform performs abnormal event detection based on the target image.

[0122] Exemplarily, when the acquisition module 71 acquires at least one storage queue corresponding to the target camera, it is specifically used to: after obtaining the abnormal event image captured by the target camera, if the similarity between the image features of the abnormal event image and the stored image features of the storage queue corresponding to the target camera is greater than a preset similarity threshold (which can be configured based on experience), then the abnormal event image is stored in the storage queue; or, if the similarity between the image features of the abnormal event image and the stored image features of all storage queues corresponding to the target camera is not greater than the preset similarity threshold, then a new storage queue is created for the target camera, and the abnormal event image is stored in the new storage queue.

[0123] Exemplarily, when the determination module 72 determines the image filtering parameter corresponding to the storage queue based on the user preference information, it is specifically used to: if the user preference information is a focus preference, determine that the image filtering parameter corresponding to the storage queue is a first filtering probability value; if the user preference information is a non-focus preference, determine that the image filtering parameter corresponding to the storage queue is a second filtering probability value; wherein, the second filtering probability value is greater than the first filtering probability value.

[0124] Exemplarily, when determining the user preference information corresponding to the storage queue, the determination module 72 is specifically used to: obtain the configured preference corresponding to the storage queue, and determine the user preference information corresponding to the storage queue based on the configured preference; wherein, if the configured preference indicates paying attention to the abnormal event image in the storage queue, then the user preference information is determined to be a attention preference; if the configured preference indicates not paying attention to the abnormal event image in the storage queue, then the user preference information is determined to be a non-attention preference; or, obtain the review label corresponding to the storage queue, and determine the user preference information corresponding to the storage queue based on the review label; wherein, if the review label indicates that the target image corresponding to the storage queue is detected as a positive image by the management platform, then the user preference information is determined to be an attention preference; if the review label indicates that the target image corresponding to the storage queue is detected as a false alarm image by the management platform, then the user preference information is determined to be a non-attention preference; if the review label indicates that the target image corresponding to the storage queue is not detected by the management platform, then the user preference information is determined to be a non-attention preference.

[0125] Exemplarily, when the processing module 73 determines whether to filter the target image based on the image filtering parameter corresponding to the target storage queue, it is specifically used to: if the image filtering parameter corresponding to the target storage queue is a first filtering probability value, generate a first numerical interval and a second numerical interval based on the first filtering probability value; generate a first random number for the target image, and if the first random number matches the first numerical interval, filter the target image; if the first random number matches the second numerical interval, do not filter the target image; if the image filtering parameter corresponding to the target storage queue is a second filtering probability value, generate a third numerical interval and a fourth numerical interval based on the second filtering probability value; generate a second random number for the target image, and if the second random number matches the third numerical interval, filter the target image; if the second random number matches the fourth numerical interval, do not filter the target image; wherein, the length of the third numerical interval is greater than the length of the first numerical interval; wherein, the length of the fourth numerical interval is less than the length of the second numerical interval.

[0126] Exemplarily, if the number of target images remaining after filtering is multiple, the processing module 73 is specifically used when sending the target images to the management platform: for each target image, determine the sending order of the target image based on the image features of the target image, and send the multiple target images to the management platform in sequence based on the sending order of each target image; wherein, the processing module is specifically used when determining the sending order of the target image based on the image features of the target image, and sending the multiple target images to the management platform in sequence based on the sending order of each target image: if the similarity between the image features of the target image and the image features of the positive image is greater than the first similarity threshold, then the target image is sent to the management platform. The target image is stored in the first sending queue; if the similarity between the image features of the target image and the image features of the false alarm image is greater than the second similarity threshold, the target image is stored in the second sending queue; if the similarity between the image features of the target image and the image features of the positive alarm image is not greater than the first similarity threshold, and the similarity between the image features of the target image and the image features of the false alarm image is not greater than the second similarity threshold, the target image is stored in the third sending queue; the target images in the first sending queue are first sent to the management platform, and then the target images in the third sending queue are sent to the management platform, and then the target images in the second sending queue are sent to the management platform.

[0127] Based on the same application concept as the above method, an abnormal event detection device is proposed in the embodiment of the present application, see Figure 8As shown, the abnormal event detection device includes a processor 81 and a machine-readable storage medium 82, and the machine-readable storage medium 82 stores machine-executable instructions that can be executed by the processor 81; the processor 81 is used to execute the machine-executable instructions to implement the abnormal event detection method disclosed in the above example of this application.

[0128] Based on the same application concept as the above method, an embodiment of the present application also provides a machine-readable storage medium, on which a number of computer instructions are stored. When the computer instructions are executed by a processor, the abnormal event detection method disclosed in the above example of the present application can be implemented.

[0129] The machine-readable storage medium may be any electronic, magnetic, optical, or other physical storage device that may contain or store information, such as executable instructions, data, and the like. For example, the machine-readable storage medium may be: RAM (Random Access Memory), volatile memory, non-volatile memory, flash memory, a storage drive (such as a hard disk drive), a solid-state drive, any type of storage disk (such as a CD, DVD, etc.), or similar storage media, or a combination thereof.

[0130] The systems, devices, modules, or units described in the above embodiments may be implemented by computer chips or entities, or by products having certain functions. A typical implementation device is a computer, which may be in the form of a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email transceiver, game console, tablet computer, wearable device, or any combination of these devices.

[0131] For the convenience of description, the above devices are described as being divided into various units according to their functions. Of course, when implementing this application, the functions of each unit can be implemented in the same or multiple software and / or hardware.

[0132] Those skilled in the art will appreciate that the embodiments of the present application can be provided as methods, systems, or computer program products. Therefore, the present application can adopt the form of a complete hardware embodiment, a complete software embodiment, or an embodiment in combination with software and hardware. Moreover, the embodiments of the present application can adopt the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) that contain computer-usable program code.

[0133] The present application is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or box in the flowchart and / or block diagram, as well as the combination of the processes and / or boxes in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the steps in the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0134] Furthermore, these computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0135] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for executing on the computer or other programmable device to implement the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0136] The foregoing is merely an embodiment of the present application and is not intended to limit the present application. For those skilled in the art, the present application may have various changes and variations. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should all be included within the scope of the claims of the present application.

Claims

1. A method for detecting abnormal events, characterized in that: The method comprises: Acquire at least one storage queue corresponding to a target camera, each storage queue including abnormal event images captured by the target camera; wherein the abnormal event images in the same storage queue have similar features, and the abnormal event images in different storage queues have dissimilar features; For each storage queue, determine the user preference information corresponding to the storage queue, and determine the image filtering parameters corresponding to the storage queue based on the user preference information; the user preference information is determined according to the review label corresponding to the storage queue; wherein, if the review label indicates that the target image corresponding to the storage queue is detected as a positive image by the management platform, then determine that the user preference information is a focus preference; if the review label indicates that the target image corresponding to the storage queue is detected as a false alarm image by the management platform, or the target image corresponding to the storage queue is not detected by the management platform, then determine that the user preference information is a non-focus preference; the review label indicates that the target image corresponding to the storage queue is detected as a positive image by the management platform by the following method: if the management platform's review result for at least one target image in the storage queue is that the abnormal event corresponding to the target image actually exists, then determine that the review label indicates that the target image corresponding to the storage queue is detected as a positive image by the management platform; For a target image with an abnormal event captured by a target camera, a target storage queue corresponding to the target image is selected from all storage queues corresponding to the target camera, and whether to filter the target image is determined based on an image filtering parameter corresponding to the target storage queue; If so, the target image is filtered.

2. The method according to claim 1, characterized in that The obtaining of at least one storage queue corresponding to the target camera includes: After obtaining the abnormal event image captured by the target camera, if the similarity between the image features of the abnormal event image and the stored image features of the storage queue corresponding to the target camera is greater than a preset similarity threshold, the abnormal event image is stored in the storage queue; If the similarity between the image features of the abnormal event image and the stored image features of all storage queues corresponding to the target camera is not greater than the preset similarity threshold, a new storage queue is created for the target camera, and the abnormal event image is stored in the new storage queue.

3. The method according to claim 1, characterized in that The determining the image filtering parameter corresponding to the storage queue based on the user preference information includes: If the user preference information is a focus preference, it is determined that the image filtering parameter corresponding to the storage queue is a first filtering probability value; if the user preference information is a non-focus preference, it is determined that the image filtering parameter corresponding to the storage queue is a second filtering probability value; the second filtering probability value is greater than the first filtering probability value.

4. The method according to claim 3, characterized in that The determining whether to filter the target image based on the image filtering parameter corresponding to the target storage queue includes: If the image filtering parameter corresponding to the target storage queue is a first filtering probability value, generating a first numerical interval and a second numerical interval based on the first filtering probability value; generating a first random number for the target image, filtering the target image if the first random number matches the first numerical interval, and not filtering the target image if the first random number matches the second numerical interval; If the image filtering parameter corresponding to the target storage queue is a second filtering probability value, generating a third numerical interval and a fourth numerical interval based on the second filtering probability value; generating a second random number for the target image, filtering the target image if the second random number matches the third numerical interval, and not filtering the target image if the second random number matches the fourth numerical interval; The length of the third numerical interval is greater than the length of the first numerical interval; The length of the fourth numerical interval is smaller than the length of the second numerical interval.

5. The method according to claim 1, wherein After determining whether to filter the target image based on the image filtering parameter corresponding to the target storage queue, the method further includes: If not, the target image is sent to a management platform, and the management platform performs abnormal event detection based on the target image; If the number of target images remaining after filtering is more than one, sending the target images to the management platform includes: For each target image, a sending order of the target image is determined based on the image features of the target image, and multiple target images are sequentially sent to the management platform based on the sending order of each target image.

6. The method according to claim 5, characterized in that The step of determining a sending order of the target images based on the image features of the target images, and sequentially sending the plurality of target images to the management platform based on the sending order of each target image, includes: If the similarity between the image features of the target image and the image features of the positive image is greater than a first similarity threshold, the target image is stored in a first sending queue; if the similarity between the image features of the target image and the image features of the false alarm image is greater than a second similarity threshold, the target image is stored in a second sending queue; if the similarity between the image features of the target image and the image features of the positive image is not greater than the first similarity threshold, and the similarity between the image features of the target image and the image features of the false alarm image is not greater than the second similarity threshold, the target image is stored in a third sending queue; First, the target images in the first sending queue are sent to the management platform, then the target images in the third sending queue are sent to the management platform, and then the target images in the second sending queue are sent to the management platform.

7. An abnormal event detection device, characterized in that: The device comprises: An acquisition module is configured to acquire at least one storage queue corresponding to a target camera, each storage queue including abnormal event images captured by the target camera; wherein abnormal event images in the same storage queue have similar features, and abnormal event images in different storage queues have dissimilar features; A determination module is used to determine, for each storage queue, user preference information corresponding to the storage queue, and determine image filtering parameters corresponding to the storage queue based on the user preference information; the user preference information is determined according to the review label corresponding to the storage queue; wherein, if the review label indicates that the target image corresponding to the storage queue is detected as a positive image by the management platform, then the user preference information is determined to be a focus preference; if the review label indicates that the target image corresponding to the storage queue is detected as a false alarm image by the management platform, or the target image corresponding to the storage queue is not detected by the management platform, then the user preference information is determined to be a non-focus preference; the review label indicating that the target image corresponding to the storage queue is detected as a positive image by the management platform is determined in the following manner: if the management platform's review result for at least one target image in the storage queue is that the abnormal event corresponding to the target image actually exists, then it is determined that the review label indicates that the target image corresponding to the storage queue is detected as a positive image by the management platform; A processing module is used to select a target storage queue corresponding to the target image with an abnormal event captured by the target camera from all storage queues corresponding to the target camera, and determine whether to filter the target image based on the image filtering parameters corresponding to the target storage queue; if so, filter the target image.

8. The device according to claim 7, characterized in that in, When the acquisition module acquires at least one storage queue corresponding to the target camera, it is specifically configured to: after obtaining the abnormal event image captured by the target camera, if the similarity between the image features of the abnormal event image and the stored image features of the storage queue corresponding to the target camera is greater than a preset similarity threshold, then store the abnormal event image in the storage queue; if the similarity between the image features of the abnormal event image and the stored image features of all storage queues corresponding to the target camera is not greater than the preset similarity threshold, then create a new storage queue for the target camera and store the abnormal event image in the new storage queue; The determining module determines the image filtering parameter corresponding to the storage queue based on the user preference information, specifically for: if the user preference information is a focus preference, determining that the image filtering parameter corresponding to the storage queue is a first filtering probability value; if the user preference information is a non-focus preference, determining that the image filtering parameter corresponding to the storage queue is a second filtering probability value; wherein the second filtering probability value is greater than the first filtering probability value; Wherein, when the processing module determines whether to filter the target image based on the image filtering parameter corresponding to the target storage queue, it is specifically used to: if the image filtering parameter corresponding to the target storage queue is a first filtering probability value, generate a first numerical interval and a second numerical interval based on the first filtering probability value; generate a first random number for the target image, and if the first random number matches the first numerical interval, filter the target image; if the first random number matches the second numerical interval, do not filter the target image; if the image filtering parameter corresponding to the target storage queue is a second filtering probability value, generate a third numerical interval and a fourth numerical interval based on the second filtering probability value; generate a second random number for the target image, and if the second random number matches the third numerical interval, filter the target image; if the second random number matches the fourth numerical interval, do not filter the target image; wherein the length of the third numerical interval is greater than the length of the first numerical interval; wherein the length of the fourth numerical interval is less than the length of the second numerical interval; Wherein, after determining whether to filter the target image based on the image filtering parameters corresponding to the target storage queue, the processing module is further used to: if not, send the target image to the management platform, and the management platform performs abnormal event detection based on the target image; if the number of target images remaining after filtering is multiple, the processing module is specifically used when sending the target image to the management platform: for each target image, determine the sending order of the target image based on the image features of the target image, and send the multiple target images to the management platform in sequence based on the sending order of each target image; wherein, the processing module determines the sending order of the target image based on the image features of the target image, and sends the multiple target images to the management platform in sequence based on the sending order of each target image Used for: if the similarity between the image features of the target image and the image features of the positive image is greater than a first similarity threshold, then storing the target image in the first sending queue; if the similarity between the image features of the target image and the image features of the false alarm image is greater than a second similarity threshold, then storing the target image in the second sending queue; if the similarity between the image features of the target image and the image features of the positive image is not greater than the first similarity threshold, and the similarity between the image features of the target image and the image features of the false alarm image is not greater than the second similarity threshold, then storing the target image in the third sending queue; first sending the target images in the first sending queue to the management platform, then sending the target images in the third sending queue to the management platform, and then sending the target images in the second sending queue to the management platform.

9. An abnormal event detection device, characterized in that: include: a processor and a machine-readable storage medium storing machine-executable instructions capable of being executed by the processor; The processor is configured to execute machine-executable instructions to implement any method step of claims 1-6.

Images