A scene change detection method and system based on video analysis

Abstract

The application discloses a scene change detection system and method based on video analysis, and a storage medium. The system comprises the following steps: acquiring video image data, processing the video image data, obtaining target elements, traversing each video frame according to the order in the video image data, determining the first scene similarity corresponding to the current video frame according to the target elements corresponding to the current video frame and the scene reference target, updating the scene reference target when the first scene similarity is less than the first threshold, and otherwise maintaining the scene reference target unchanged. The application extracts the depth features of the invariant elements in the scene based on deep learning, does not need to manually design the depth features, and improves the robustness of the scene description features. The application maintains the automatic update of the scene change judgment standard, is favorable for more accurately judging whether the video frame has the scene change, and is favorable for improving the automation degree of the scene change detection. The application is widely applied in the technical field of image processing.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, and in particular to a method and system for scene change detection based on video analysis. Background Technology

[0002] In recent years, surveillance systems have been widely used in transportation, military, airports, banks, shopping malls, communities, and other locations. In many fields, one of the crucial conditions for the normal operation of a surveillance system is obtaining stable, clear, and highly targeted video images.

[0003] The large-scale application of surveillance systems has created new demands for their operation and maintenance capabilities. Firstly, 24 / 7 operation increases the risk and probability of product malfunctions. Secondly, the dispersed deployment and large number of surveillance devices significantly increase the cost and difficulty of manual inspection. In practical applications, changes in the shooting angle of surveillance cameras due to external forces or human intervention can affect the target in the video. Therefore, research on automatic scene change detection using computer vision technology is crucial for quickly and accurately identifying scene changes in surveillance videos.

[0004] Existing methods for scene change detection based on computer vision technology generally employ manually designed scene description features. These features calculate the amount of feature change between adjacent images (frames) and compare it to a set threshold; images exceeding the threshold are detected as having scene changes. Scene description features mainly utilize pixel features, motion features, and feature points. Existing pixel feature-based methods, such as pixel difference features and histogram features, are sensitive to changes in lighting and moving targets within the scene. Among motion feature-based methods, reliable motion estimation algorithms consume significant computational resources. In feature point matching-based methods, the selection of feature points varies greatly across different scenes, and the computational complexity is substantial.

[0005] Real-world application scenarios are complex and diverse, and manually designed scene description features have low robustness and are easily affected by lighting and slight vibrations in the scene. Summary of the Invention

[0006] To address the technical problems of current scene change detection technologies, such as low robustness of scene description features and susceptibility to the effects of lighting and slight shaking in the scene, the present invention aims to provide a scene change detection method and system based on video analysis.

[0007] On one hand, embodiments of the present invention include a scene change detection method based on video analysis, comprising:

[0008] Acquire video image data; the video image data includes multiple video frames;

[0009] The video image data is processed to obtain the target elements corresponding to each video frame;

[0010] Based on the video image data, set the initial value of the scene reference object;

[0011] According to the order in the video image data, each video frame is traversed; for any current video frame during the traversal process, the first scene similarity corresponding to the current video frame is determined based on the target element corresponding to the current video frame and the scene reference object. If the first scene similarity is less than a first threshold, the scene reference object is updated; otherwise, the scene reference object is kept unchanged. The updated or maintained scene reference object is used to determine the first scene similarity corresponding to subsequent video frames.

[0012] Furthermore, the acquisition of video image data includes:

[0013] Obtain the raw video data file;

[0014] The original video data file is preprocessed to obtain the video image data.

[0015] Furthermore, the acquisition of the original video data file includes:

[0016] The original video data file is obtained by capturing images of a fixed scene using a camera installed in the scene.

[0017] or

[0018] Obtain the historical video data file saved by the camera in the scene to obtain the original video data file.

[0019] Further, the preprocessing of the original video data file includes:

[0020] Convert the image dimensions of the original video data file to standard dimensions;

[0021] Convert the image color space of the original video data file to a standard color space.

[0022] Further, setting the initial value of the scene reference object based on the video image data includes:

[0023] Define the specific category of the subject matter;

[0024] Select the first few video frames from the video image data;

[0025] Each of the selected video frames is used to delineate corresponding target elements; each of the delineated target elements represents a target of the specific category.

[0026] The intersection of all defined target elements is used as the initial value of the scene benchmark.

[0027] Further, the step of processing the video image data to obtain the target elements corresponding to each of the video frames includes:

[0028] A deep learning-based target detection algorithm is used to detect the video image data and obtain the target detection elements corresponding to each video frame.

[0029] The video image data is segmented using a deep learning-based target segmentation algorithm to obtain the target segmentation elements corresponding to each video frame.

[0030] Further, determining the first scene similarity corresponding to the current video frame based on the target element corresponding to the current video frame and the scene reference object includes:

[0031] According to the formula Perform the calculation;

[0032] Among them, score q Reg represents the first scene similarity corresponding to the q-th video frame of the current video frame. t and Rec o Reg indicates the area where the reference object in the scene is located. k Rec represents the region where the target detection feature is located in the q-th video frame. l Let n represent the region where the target segmentation feature is located in the q-th video frame, n represent the number of target detection features in the q-th video frame, and m represent the number of segmentation detection features in the q-th video frame.

[0033] Furthermore, updating the scene benchmark includes:

[0034] Set a sliding window; the sliding window is used to select a series of consecutive video frames at a given location;

[0035] Near the current video frame, the sliding window is moved; for any current position of the sliding window during the movement, corresponding target elements are delineated in a series of consecutive video frames determined by the sliding window at the current position, wherein the delineated target elements all represent objects of the specific category. The intersection of all delineated target elements is used as a temporary scene reference object. Based on the target elements corresponding to the current video frame and the temporary scene reference object, a second scene similarity is determined for the temporary scene reference object. When the second scene similarity is greater than a second threshold, the movement of the sliding window is stopped, and the temporary scene reference object is used as the new scene reference object; otherwise, the movement of the sliding window continues.

[0036] Further, determining the second scene similarity corresponding to the temporary scene reference object based on the target element corresponding to the current video frame and the temporary scene reference object includes:

[0037] According to the formula Perform the calculation;

[0038] Among them, score q Represents the similarity of the second scene, where Regt and Reco represent the regions where the temporary scene reference object is located. k Rec represents the region where the target detection feature is located in the q-th video frame, which is the current video frame. l Let n represent the region where the target segmentation feature is located in the q-th video frame, n represent the number of target detection features in the q-th video frame, and m represent the number of segmentation detection features in the q-th video frame.

[0039] On the other hand, embodiments of the present invention also include a scene change detection system based on video analysis, the scene change detection system based on video analysis comprising:

[0040] The first module is used to acquire video image data; the video image data includes multiple video frames.

[0041] The second module is used to set the initial value of the scene reference object based on the video image data;

[0042] The third module is used to process the video image data and obtain the target elements corresponding to each video frame.

[0043] The fourth module is used to traverse each video frame in the order of the video image data; for any current video frame during the traversal process, a first scene similarity is determined based on the target element corresponding to the current video frame and the scene reference object; if the first scene similarity is less than a first threshold, the scene reference object is updated, otherwise the scene reference object is kept unchanged; the updated or maintained scene reference object is used to determine the first scene similarity corresponding to subsequent video frames.

[0044] The beneficial effects of this invention are as follows: The scene change detection method based on video analysis in the embodiments is based on the deep features of invariant elements in the scene extracted by deep learning, which eliminates the need for manual design of deep features and improves the robustness of scene description features; when a scene change is detected in the current video frame, the scene reference calibration object is updated, which enables the scene reference calibration object to change with the scene change of the video frame, thereby maintaining the automatic updating of the scene change judgment standard. On the one hand, this is conducive to more accurate judgment of whether a scene change has occurred in the video frame, and on the other hand, it is conducive to improving the automation level of scene change detection. Attached Figure Description

[0045] Figure 1 This is a flowchart of the scene change detection method based on video analysis in the embodiment;

[0046] Figure 2 This is a schematic diagram of the scene change detection method based on video analysis in the embodiment.

[0047] Figure 3 This is a flowchart illustrating the process of updating the scene benchmark in this embodiment. Detailed Implementation

[0048] The scene change detection problem can be defined as follows: when the shooting angle of a surveillance camera changes due to external force or human intervention, affecting the normal monitoring function, the system can automatically detect this scene change in a timely manner and issue an early warning.

[0049] In this embodiment, refer to Figure 1 The scene change detection method based on video analysis includes the following steps:

[0050] S1. Acquire video image data;

[0051] S2. Process the video image data to obtain the target elements corresponding to each video frame;

[0052] S3. Based on the video image data, set the initial values ​​of the scene reference objects;

[0053] S4. Traverse each video frame in the order of the video image data; for any current video frame during the traversal, determine the first scene similarity corresponding to the current video frame based on the target element and the scene reference object corresponding to the current video frame. If the first scene similarity is less than the first threshold, update the scene reference object; otherwise, keep the scene reference object unchanged; the updated or maintained scene reference object is used to determine the first scene similarity corresponding to subsequent video frames.

[0054] In this embodiment, the principle of steps S1-S4 is as follows: Figure 2 As shown.

[0055] During step S1, a camera installed in the scene can be used to capture images of the fixed scene to obtain the original video data file, or historical video data files saved by the cameras in the scene can be retrieved to obtain the original video data file. The original video data file can be represented as {frame1, frame2…frame…} q}, where frame q This represents a video frame in the original video data file.

[0056] For the original video data file {frame1, frame2...frame q The image data undergoes image resizing and color space conversion to transform each frame of the original video data to a standard size and color space, resulting in video frames {frame1, frame2', ..., frame...}. q ′} represents video image data.

[0057] In this embodiment, when performing step S2, which is to process the video image data and obtain the target elements corresponding to each video frame, the following steps can be performed:

[0058] S201. Use a deep learning-based object detection algorithm to detect video image data and obtain the object detection elements corresponding to each video frame;

[0059] S202. Use a deep learning-based target segmentation algorithm to segment the video image data and obtain the target segmentation elements corresponding to each video frame.

[0060] In step S201, for video image data {frame1, frame2'…frame...} q For each frame in the video, a deep learning-based object detection algorithm is used to detect the objects, obtaining the corresponding object detection features for each video frame. Taking the q-th video frame as an example... qFor example, using a deep learning-based object detection algorithm, we can obtain the category information Cla of m objects. m and regional information Rec m m, thus forming Det q ={{Cia1, Rec1}, {Cla2, Rec2}…{Cla m Rec m}}. Det q For the q-th video frame q The corresponding target detection elements.

[0061] In step S202, for video image data {frame1, frame2'…frame...} q Each frame in the video frame '} is segmented using a deep learning-based target segmentation algorithm to obtain the target segmentation features corresponding to each video frame. Taking the q-th video frame as an example... q For example, a deep learning-based object detection algorithm is used for segmentation to obtain the category information of n objects. n and regional information Reg n Thus forming Seg q ={{Lab1, Reg1}, {Lab2, Reg2}…{Lab n Reg n}}. Seg q For the q-th video frame q The corresponding target segmentation elements.

[0062] In this embodiment, the q-th video frame q Corresponding target detection element Det q Seq of target segmentation features q , composing target elements pet q Seg q The set of target elements in each video frame is represented as {{Det1, Seg1}, {Det2, Seg2}...pet q Seg q}}.

[0063] In this embodiment, when performing step S3, which is to set the initial value of the scene reference object based on the video image data, the following steps can be performed:

[0064] S301. Define the specific category of the subject matter;

[0065] S302. Select the first few video frames from the video image data;

[0066] S303. Delineate the corresponding target elements from the selected video frames respectively; the delineated target elements all represent objects of a specific category;

[0067] S304. The intersection of all defined target elements shall be used as the initial value of the scene reference object.

[0068] In step S301, taking a traffic scenario as an example, since the signs and markings in the road area are fixed, the signs and markings can be used as a significant feature of the traffic scenario, and thus as a standard for scenario change detection. Therefore, the specific category of the target is set as "signs and markings", which can be specifically set as {Cla = Road, Lab = Line}.

[0069] In step S302, the video image data {frame1, frame2'…frame...} can be obtained from the video image data. q Select the first i video frames from the given frame.

[0070] In step S303, for the i video frames selected in step S302, target elements that can represent a specific category of object (i.e., signs and markings according to step S301) are drawn from each video frame. Since the target elements in each video frame represent a specific category of object (signs and markings), and the specific category of object is constant in a specific scene (road area), the target elements in each video frame are invariant elements.

[0071] In step S304, the intersection Mark of all target elements defined in step S303 is calculated. init = {{Det1, Det2, ..., Det} x}, {Seg1, Seg2, ..., Seg y}}, thus obtaining the initial value Mark of the scene reference object. init .

[0072] The principle of step S4 is as follows: scene changes are judged for real-time continuous video image frames. In the scene change judgment, a scene similarity calculation method is proposed based on the reference calibration objects of scene invariant elements. The principle of scene similarity is based on the existence of fixed and unchanging elements in the scene, which can be used as reference calibration objects for position matching between continuous frames. When the scene does not change, the overlap between the calibration objects detected in real time in continuous frames and the reference calibration objects is high, which indicates that the scene similarity is high. When the scene changes, the overlap between the calibration objects detected in real time in continuous frames and the reference calibration objects is low, or even non-overlapping, which indicates that the scene similarity is low.

[0073] Based on the above principle, in step S4, for the video image data {frame1, frame2'…frame... q'}, starting from the first video frame, frame1, iterate through the data.

[0074] Initially, the current video frame is the first video frame, frame1. Based on the target elements and scene reference objects corresponding to the first video frame, frame1 (which are the initial values), according to the formula... Calculate the first scene similarity corresponding to the first video frame (frame1), where score1 represents the first scene similarity corresponding to the first video frame (the current video frame), and Reg t and Rec o Indicates the area where the scene reference object is located, Reg k Rec represents the region where the object detection feature is located in the first video frame (the current video frame). l Let represent the region where the target segmentation features are located in the first video frame (the current video frame), n represent the number of target detection features in the first video frame (the current video frame), and m represent the number of segmentation detection features in the first video frame (the current video frame). The first scene similarity score1 corresponding to the first video frame is compared with the first threshold thre1. If score1 > thre1, it means the scene has not changed (i.e., the scene is still a "road area"), and we can proceed to the next video frame, the second video frame, frame2. At this point, the second video frame, frame2, replaces the first video frame, frame1, as the current video frame. Using the unchanged scene reference object, the first scene similarity score2 is calculated, and then it is determined whether score2 > thre1 (the scene has not changed) or score2 ≤ thre1 (the scene has changed). If score2 > thre1 is still true (the scene has not changed), we can proceed to the next video frame, the third video frame, frame3, and continue the same process until all video image data {frame1, frame2'…frame3} has been traversed. q All video frames in '}.

[0075] In step S4, as long as any current video frame (e.g., the q-th video frame) q The score appeared q ≤thre1 indicates a change in the scenario, so refer to Figure 3 The process pauses before moving to the next video frame (the (q+1)th video frame), updates the scene reference objects, and then proceeds to the next video frame (the (q+1)th video frame). For subsequent video frames (e.g., the (q+1)th video frame), the updated scene reference objects are used to calculate the first scene similarity corresponding to the video frame.

[0076] In step S4, as long as any current video frame (e.g., the q-th video frame) q The score appeared q ≤thre1, refer to Figure 3 A sliding window can be set, which has a fixed size and can select j consecutive video frames. The position of the sliding window is variable, meaning that video image data {frame1, frame2'…frame} can be arbitrarily selected. q j consecutive video frames on ′}.

[0077] Initially, you can set the sliding window to the current video frame (the q-th video frame). q The location near the current video frame (the q-th video frame) is selected. q The j consecutive video frames (including the q-th video frame) q (This can be at any position within these j consecutive video frames). For these j consecutive video frames, define the target detection features {Det1, Det2, ..., Det} for each video frame. z} and target segmentation features {Seg1, Seg2, ... Seg r}, through the intersection Mark of all defined target elements temp = {{Det1, Det2, ..., Det} z}, {Seg1, Seg2, ..., Seq r}}, obtain the temporary scene reference object Mark temp .

[0078] In obtaining the temporary scene benchmark object Mark temp Subsequently, based on the temporary scenario benchmark object Mark temp With the current video frame (the q-th video frame) q For the corresponding target elements, calculate the temporary scene benchmark object Mark. temp The corresponding second scene similarity score q In this embodiment, the second scene similarity score is calculated. q The formula is In the formula, Reg t and Rec o Indicates the area where the temporary scene reference object is located, Reg k Rec represents the region where the object detection feature is located in the q-th video frame of the current video frame. l Let n represent the region where the target segmentation feature is located in the q-th video frame, n represent the number of target detection features in the q-th video frame, and m represent the number of segmentation detection features in the q-th video frame.

[0079] A second similarity threshold, thre2, is set based on different scenarios to determine whether the scenario stops changing. When the second scenario similarity score... q If the value is less than or equal to the second threshold thre2, it indicates that the scene has still changed, and the sliding window is moved to a new position (e.g., along the video image data {frame1, frame2…frame2}). q (Moving forward one frame along the timeline), the moved sliding window can select a new set of j consecutive video frames, thereby recalculating the second scene similarity score. q If the new second scene similarity score q If it is still less than or equal to the second threshold thre2, then continue to repeat the process of moving the sliding window to a new position.

[0080] As long as the second scene similarity score appears q If the second threshold thre2 indicates that the scene of the latest j video frames no longer changes, then the temporary scene reference object Mark can be calculated by intersecting all target elements of these latest j video frames. temp The updated scene benchmark object is determined and the update of the scene benchmark object is completed.

[0081] The scene change detection method based on video analysis in this embodiment is based on the deep features of invariant elements in the scene extracted by deep learning. It does not require manual design of deep features, thus improving the robustness of scene description features. When a scene change is detected in the current video frame, the scene reference calibration object is updated, which enables the scene reference calibration object to change with the scene changes of the video frame. This maintains the automatic updating of the scene change judgment standard, which is beneficial for more accurate judgment of whether a scene change has occurred in a video frame, and also helps to improve the automation of scene change detection.

[0082] The scene change detection method based on video analysis in this embodiment uses a deep feature method based on continuous frame video data to detect scene changes. It can accurately and quickly detect scene changes in real-time monitoring videos or video files, and can achieve scene change detection under various lighting and weather conditions in real-world scenarios, thus enhancing the robustness of the algorithm.

[0083] The scene change detection method based on video analysis in this embodiment uses scene element category information and spatial location information extracted from deep features to match with scene benchmark objects. Scene change detection is achieved based on the difference in positional similarity. At the same time, considering the method of updating scene benchmark objects after scene changes in practical applications, it is more conducive to practical applications.

[0084] This embodiment provides a scene change detection system based on video analysis, including:

[0085] The first module is used to acquire video image data; the video image data includes multiple video frames.

[0086] The second module is used to set the initial values ​​of the scene reference objects based on the video image data;

[0087] The third module is used to process video image data and obtain the target elements corresponding to each video frame.

[0088] The fourth module is used to traverse each video frame in the order of the video image data. For any current video frame during the traversal, the first scene similarity corresponding to the current video frame is determined based on the target element corresponding to the current video frame and the scene reference object. If the first scene similarity is less than a first threshold, the scene reference object is updated; otherwise, the scene reference object is kept unchanged. The updated or maintained scene reference object is used to determine the first scene similarity corresponding to subsequent video frames.

[0089] When the video analysis-based scene change detection system is running, the first module can execute step S1 in the video analysis-based scene change detection method, the second module can execute step S2, the third module can execute step S3, and the fourth module can execute step S4. Therefore, the video analysis-based scene change detection system can execute the video analysis-based scene change detection method, thereby achieving the technical effect of the video analysis-based scene change detection method.

[0090] A computer program that executes the scene change detection method based on video analysis in this embodiment can be written into a storage medium or computer device. When the computer program is read out and run, the scene change detection method based on video analysis in this embodiment is executed, thereby achieving the same technical effect as the scene change detection method based on video analysis in the embodiment.

[0091] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. Furthermore, the descriptions of "upper," "lower," "left," and "right" used in this disclosure are only relative to the relative positional relationships of the various components of this disclosure in the accompanying drawings. The singular forms "a," "described," and "the" used in this disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. Moreover, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this embodiment specification is only for describing particular embodiments and is not intended to limit the invention. The term "and / or" as used in this embodiment includes any combination of one or more of the associated listed items.

[0092] It should be understood that although the terms first, second, third, etc., may be used to describe various elements in this disclosure, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from each other. For example, a first element may also be referred to as a second element without departing from the scope of this disclosure, and similarly, a second element may also be referred to as a first element. The use of any and all instances or exemplary language (“e.g.,” “such as,” etc.) provided in this embodiment is intended only to better illustrate embodiments of the invention and, unless otherwise required, does not impose a limitation on the scope of the invention.

[0093] It should be recognized that embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can be implemented using standard programming techniques—including a non-transitory computer-readable storage medium configured with a computer program, wherein such a storage medium causes the computer to operate in a specific and predefined manner—according to the methods and drawings described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).

[0094] Furthermore, the procedures described in this embodiment can be performed in any suitable order unless otherwise indicated by this embodiment or clearly contradicted by the context. The procedures (or variations and / or combinations thereof) described in this embodiment can be executed under the control of one or more computer systems configured with executable instructions, and can be implemented by hardware or a combination thereof as code (e.g., executable instructions, one or more computer programs, or one or more applications) that commonly executes on one or more processors. The computer program includes a plurality of instructions executable by one or more processors.

[0095] Furthermore, the method can be implemented in any suitable type of computing platform, including but not limited to personal computers, minicomputers, mainframes, workstations, networked or distributed computing environments, standalone or integrated computer platforms, or in communication with charged particle tools or other imaging devices. Aspects of the invention can be implemented as machine-readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage medium, RAM, ROM, etc., such that it is readable by a programmable computer, and when the storage medium or device is read by the computer, it can be used to configure and operate the computer to perform the processes described herein. Furthermore, the machine-readable code, or portions thereof, can be transmitted via wired or wireless networks. The invention described in this embodiment includes these and other different types of non-transitory computer-readable storage media when such media comprises instructions or programs that implement the steps described above in conjunction with a microprocessor or other data processor. When programmed according to the methods and techniques described in the invention, the invention also includes the computer itself.

[0096] A computer program can be applied to input data to perform the functions described in this embodiment, thereby transforming the input data to generate output data stored in non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In a preferred embodiment of the invention, the transformed data represents physical and tangible objects, including specific visual depictions of physical and tangible objects generated on the display.

[0097] The above description is merely a preferred embodiment of the present invention. The present invention is not limited to the above-described embodiments. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention, as long as they achieve the technical effects of the present invention by the same means, should be included within the scope of protection of the present invention. Within the scope of protection of the present invention, the technical solutions and / or implementation methods can have various modifications and variations.

Claims

1. A scene change detection method based on video analysis, characterized in that, The video analysis-based scene change detection method includes: Acquire video image data; the video image data includes multiple video frames; The video image data is processed to obtain the target elements corresponding to each video frame; Based on the video image data, set the initial value of the scene reference object; According to the order in the video image data, each video frame is traversed; for any current video frame during the traversal process, the first scene similarity corresponding to the current video frame is determined based on the target element corresponding to the current video frame and the scene reference object; if the first scene similarity is less than a first threshold, the scene reference object is updated, otherwise the scene reference object is kept unchanged; the updated or maintained scene reference object is used to determine the first scene similarity corresponding to subsequent video frames. The step of processing the video image data to obtain the target elements corresponding to each video frame includes: A deep learning-based target detection algorithm is used to detect the video image data and obtain the target detection elements corresponding to each video frame. The video image data is segmented using a deep learning-based target segmentation algorithm to obtain the target segmentation elements corresponding to each video frame. The step of determining the first scene similarity corresponding to the current video frame based on the target element corresponding to the current video frame and the scene reference object includes: According to the formula Perform the calculation; in, Indicates the number of the current video frame. The similarity of the first scene to each video frame. and This indicates the area where the reference object in the scene is located. Indicates the first The region where the target detection elements are located in each video frame. Indicates the first The region where the target segmentation feature is located in each video frame. Indicates the first The number of target detection features in each video frame. Indicates the first The number of segmented and detected features in each video frame; The updating of the scenario benchmark object includes: Set a sliding window; the sliding window is used to select a series of consecutive video frames at a given location; Near the current video frame, the sliding window is moved. For any current position of the sliding window during the movement, corresponding target elements are delineated in a series of consecutive video frames determined by the sliding window at the current position. The delineated target elements all represent objects of a specific category. The intersection of all delineated target elements is used as a temporary scene reference object. Based on the target elements corresponding to the current video frame and the temporary scene reference object, a second scene similarity is determined. When the second scene similarity is greater than a second threshold, the movement of the sliding window is stopped, and the temporary scene reference object is used as the new scene reference object. Otherwise, the movement of the sliding window continues.

2. The scene change detection method based on video analysis according to claim 1, characterized in that, The acquisition of video image data includes: Obtain the raw video data file; The original video data file is preprocessed to obtain the video image data.

3. The scene change detection method based on video analysis according to claim 2, characterized in that, The process of obtaining the original video data file includes: The original video data file is obtained by capturing images of a fixed scene using a camera installed in the scene. or Obtain the historical video data file saved by the camera in the scene to obtain the original video data file.

4. The scene change detection method based on video analysis according to claim 2, characterized in that, The preprocessing of the original video data file includes: Convert the image dimensions of the original video data file to standard dimensions; Convert the image color space of the original video data file to a standard color space.

5. The scene change detection method based on video analysis according to any one of claims 1-3, characterized in that, The step of setting the initial value of the scene reference object based on the video image data includes: Define the specific category of the subject matter; Select the first few video frames from the video image data; Each of the selected video frames is used to delineate corresponding target elements; each of the delineated target elements represents a target of the specific category. The intersection of all defined target elements is used as the initial value of the scene benchmark.

6. The scene change detection method based on video analysis according to claim 1, characterized in that, The step of determining the second scene similarity corresponding to the temporary scene benchmark object based on the target element corresponding to the current video frame and the temporary scene benchmark object includes: According to the formula Perform the calculation; in, Indicates the similarity of the second scene. and This indicates the area where the temporary scene reference object is located. Indicates the number of frames in the current video frame. The region where the target detection elements are located in each video frame. Indicates the first The region where the target segmentation feature is located in each video frame. Indicates the first The number of target detection features in each video frame. Indicates the first The number of segmentation and detection features in each video frame.

7. A scene change detection system based on video analysis, characterized in that, The video analytics-based scene change detection system includes: The first module is used to acquire video image data; the video image data includes multiple video frames. The second module is used to set the initial value of the scene reference object based on the video image data; The third module is used to process the video image data and obtain the target elements corresponding to each video frame. The fourth module is used to traverse each video frame in the order of the video image data; for any current video frame during the traversal process, a first scene similarity is determined based on the target element corresponding to the current video frame and the scene reference object; if the first scene similarity is less than a first threshold, the scene reference object is updated, otherwise the scene reference object is kept unchanged; the updated or maintained scene reference object is used to determine the first scene similarity corresponding to subsequent video frames; The step of processing the video image data to obtain the target elements corresponding to each video frame includes: A deep learning-based target detection algorithm is used to detect the video image data and obtain the target detection elements corresponding to each video frame. The video image data is segmented using a deep learning-based target segmentation algorithm to obtain the target segmentation elements corresponding to each video frame. The step of determining the first scene similarity corresponding to the current video frame based on the target element corresponding to the current video frame and the scene reference object includes: According to the formula Perform the calculation; in, Indicates the number of the current video frame. The similarity of the first scene to each video frame. and This indicates the area where the reference object in the scene is located. Indicates the first The region where the target detection elements are located in each video frame. Indicates the first The region where the target segmentation feature is located in each video frame. Indicates the first The number of target detection features in each video frame. Indicates the first The number of segmented and detected features in each video frame; The updating of the scenario benchmark object includes: Set a sliding window; the sliding window is used to select a series of consecutive video frames at a given location; Near the current video frame, the sliding window is moved. For any current position of the sliding window during the movement, corresponding target elements are delineated in a series of consecutive video frames determined by the sliding window at the current position. The delineated target elements all represent objects of a specific category. The intersection of all delineated target elements is used as a temporary scene reference object. Based on the target elements corresponding to the current video frame and the temporary scene reference object, a second scene similarity is determined. When the second scene similarity is greater than a second threshold, the movement of the sliding window is stopped, and the temporary scene reference object is used as the new scene reference object. Otherwise, the movement of the sliding window continues.

Images