CN115909219A - Scene change detection method and system based on video analysis - Google Patents

Abstract

The invention discloses a scene change detection system, a method, a device and a storage medium based on video analysis, which comprises the steps of acquiring video image data, processing the video image data, acquiring target elements, traversing each video frame according to the sequence 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 a scene reference object, updating the scene reference object when the first scene similarity is smaller than a first threshold value, and otherwise, keeping the scene reference object unchanged. The depth features of the invariant elements in the scene extracted based on the deep learning do not need to be designed manually, so that the robustness of the scene description features is improved; the automatic updating of the scene change judgment standard is maintained, so that whether the video frame has a scene change or not is judged more accurately, and the automation degree of scene change detection is improved. The invention is widely applied to the technical field of image processing.

Description

A method and system for scene change detection based on video analysis

technical field

The invention relates to the technical field of image processing, in particular to a scene change detection method and system based on video analysis.

Background technique

In recent years, monitoring systems have been widely used in transportation, military, airports, banks, shopping malls, communities and other places. In many fields, one of the important conditions for the normal operation of the monitoring system is to obtain stable, clear and targeted monitoring video image content.

With the large-scale application of the monitoring system, people put forward new requirements for the operation and maintenance capabilities of the monitoring system. First of all, 24-hour uninterrupted use increases the risk and probability of monitoring product failures; secondly, due to the scattered deployment of monitoring equipment in practical applications and the large number of uses, the cost and difficulty of manual inspection have greatly increased. In practical applications, the change of the shooting angle of the surveillance camera due to external force or human action will affect the target of the surveillance video. Therefore, in order to quickly and accurately identify the scene changes in the surveillance video, computer vision technology is used to Research on automatic detection of scene changes in surveillance video is crucial.

The existing scene change detection method based on computer vision technology generally adopts artificially designed scene description features, calculates the feature change amount between adjacent images (between frames) and compares it with the set threshold, and detects when the threshold exceeds the threshold. Change the image for the scene. Scene description features mainly use pixel features, motion features, feature points, etc. Existing methods based on pixel features, such as pixel difference features and histogram features, are sensitive to illumination changes and moving objects in the scene; methods based on motion features are reliable The motion estimation algorithm consumes more computing resources; in the method based on feature point matching, the selection of feature points in different scenes is very different and the amount of calculation is complex.

The actual application scenarios are complex and diverse, and the artificially designed scene description features have low robustness and are easily affected by lighting and slight jitter in the scene.

Contents of the invention

In view of the technical problems existing in the current scene change detection technology, such as low robustness of scene description features and being easily affected by lighting and slight shaking in the scene, the purpose of the present invention is to provide a scene change detection method and system based on video analysis.

On the one hand, the embodiment of the present invention includes a scene change detection method based on video analysis, including:

Obtain video image data; the video image data includes a plurality of video frames;

Processing the video image data to obtain target elements corresponding to each of the video frames;

According to the video image data, setting the initial value of the scene benchmark object;

According to the order in the video image data, traverse each of the video frames; for any current video frame in the traversal process, determine according to the target element corresponding to the current video frame and the scene benchmark object The first scene similarity corresponding to the current video frame, when the first scene similarity is less than the first threshold, update the scene reference object, otherwise keep the scene reference object unchanged; after updating or maintaining The scene benchmark object is used to determine the first scene similarity corresponding to subsequent video frames.

Further, the acquisition of video image data includes:

Obtain the original video data file;

Preprocessing the original video data file to obtain the video image data.

Further, said obtaining the original video data file includes:

Obtaining the original video data file by shooting a fixed scene through a camera installed in the scene;

or

The historical video data files saved by the cameras in the scene are obtained, and the original video data files are obtained.

Further, the preprocessing of the original video data file includes:

converting the image size of the original video data file into a standard size;

Converting the image color space of the original video data file into a standard color space.

Further, setting the initial value of the scene benchmark object according to the video image data includes:

Set a specific category of subject matter;

selecting the first several video frames from the video image data;

Respectively delineating corresponding target elements from the selected several video frames; the delimited target elements all represent the subject matter of the specific category;

The intersection of all demarcated target elements is used as the initial value of the scene benchmark object.

Further, the processing of the video image data to obtain the respective target elements corresponding to each of the video frames includes:

Using a target detection algorithm based on deep learning to detect the video image data and obtain target detection elements corresponding to each of the video frames;

The video image data is segmented by using a target segmentation algorithm based on deep learning, and target segmentation elements corresponding to each of the video frames are obtained.

Further, the determining the first scene similarity corresponding to the current video frame according to the target element corresponding to the current video frame and the scene reference object includes:

执行计算；According to the formula

perform calculations;

Wherein, score _q represents the first scene similarity corresponding to the qth video frame as the current video frame, Reg _t and Rec _o represent the area where the scene benchmark object is located, and Reg _k represents the qth video frame The region where the target detection element is located, Rec _l represents the region where the target segmentation element is located in the qth video frame, n represents the number of the target detection element in the qth video frame, and m represents the area in the qth video frame The number of elements detected by the split.

Further, said updating said scene reference object includes:

Setting a sliding window; the sliding window is used to select a number of consecutive video frames at the location;

In the vicinity of the current video frame, move the sliding window; for any current position of the sliding window in the moving process, respectively, the sliding window is defined in several consecutive video frames determined by the current position. Corresponding target elements are delineated, wherein the delineated target elements all represent the targets of the specific category, and the intersection of all delineated target elements is used as the temporary scene reference target, according to the current video frame corresponding The target element and the temporary scene reference object, determine the second scene similarity corresponding to the temporary scene reference object, when the second scene similarity is greater than a second threshold, stop moving the sliding window, The temporary scene reference object is used as the new scene reference object, otherwise, the sliding window is continuously moved.

Further, the determining the second scene similarity corresponding to the temporary scene reference object according to the target element corresponding to the current video frame and the temporary scene reference object includes:

执行计算；According to the formula

perform calculations;

Wherein, score _q represents the similarity of the second scene, Regt and Reco represent the region where the temporary scene benchmark object is located, Reg _k represents the region where the target detection element is located in the qth video frame as the current video frame, and Rec _l represents the area where the target segmentation elements are located in the qth video frame, n represents the number of the target detection elements in the qth video frame, and m represents the number of the segmentation detection elements in the qth video frame.

On the other hand, the embodiment of the present invention also includes a scene change detection system based on video analysis, and the scene change detection system based on video analysis includes:

The first module is used to acquire video image data; the video image data includes a plurality of video frames;

The second module is used to set the initial value of the scene reference object according to the video image data;

The third module is used to process the video image data, and acquire target elements corresponding to each of the video frames;

The fourth module is used to traverse each of the video frames according to the order in the video image data; for any current video frame in the traversal process, according to the target element corresponding to the current video frame and the Scene reference object, determine the first scene similarity corresponding to the current video frame, when the first scene similarity is less than the first threshold, update the scene reference object, otherwise maintain the scene reference object change; the updated or maintained scene reference object is used to determine the first scene similarity corresponding to subsequent video frames.

The beneficial effects of the present invention are: the scene change detection method based on video analysis in the embodiment, based on the depth features of the invariant elements in the scene extracted by deep learning, does not need to manually design the depth features, and improves the robustness of scene description features ; Updating the scene reference calibration object when detecting a scene change in the current video frame can make the scene reference calibration object follow the scene change of the video frame and change, thereby maintaining the automatic update of the scene change judgment standard, which is conducive to more accurate judgment on the one hand Whether the video frame has a scene change, on the other hand, is conducive to improving the automation of scene change detection.

Description of drawings

Fig. 1 is the flowchart of the scene change detection method based on video analysis in the embodiment;

Fig. 2 is the schematic diagram of the scene change detection method based on video analysis in the embodiment;

Fig. 3 is a flow chart of updating scene reference objects in an embodiment.

Detailed ways

The scene change detection problem can be defined as: when the surveillance camera changes its shooting angle due to external force or human action, which affects the normal monitoring function, the system can automatically detect the scene change in time and give an early warning prompt.

In this embodiment, with reference to Fig. 1, the scene change detection method based on video analysis includes the following steps:

S1. Obtain video image data;

S2. Process the video image data to obtain the corresponding target elements of each video frame;

S3. According to the video image data, an initial value of the scene reference object is set;

S4. Traverse each video frame according to the order in the video image data; for any current video frame in the traversal process, determine the first scene corresponding to the current video frame according to the target element corresponding to the current video frame and the scene benchmark object Similarity, when the similarity of the first scene is less than the first threshold, the scene reference object is updated, otherwise, the scene reference object remains unchanged; the updated or maintained scene reference object is used to determine the first scene corresponding to the subsequent video frame. scene similarity.

In this embodiment, the principles of steps S1-S4 are shown in FIG. 2 .

When step S1 is executed, the camera installed in the scene can shoot the fixed scene to obtain the original video data file, or obtain the historical video data file saved by the camera in the scene to obtain the original video data file. The original video data file can be expressed as {frame ₁ , frame ₂ ... frame _q }, where frame _q represents a video frame in the original video data file.

Perform image size conversion and image color space conversion on the original video data file {frame ₁ , frame ₂ ... frame _q }, convert each frame in the original video data file to the standard size and standard excitation color space, and obtain Each video frame {frame ₁ , frame ₂ ′...frame _q ′} is the video image data.

In this embodiment, when performing step S2, that is, processing the video image data and obtaining the corresponding target elements of each video frame, the following steps can be specifically performed:

S201. Use a target detection algorithm based on deep learning to detect video image data, and obtain target detection elements corresponding to each video frame;

S202. Using a deep learning-based target segmentation algorithm, segment the video image data, and acquire target segmentation elements corresponding to each video frame.

In step S201, for each frame in the video image data {frame ₁ , frame ₂ '...frame _q '}, use the target detection algorithm based on deep learning to detect respectively, and obtain the target detection elements corresponding to each video frame. Taking the qth video frame frame _q as an example, use the target detection algorithm based on deep learning to detect, and obtain the category information Cla _m and area information Rec _m m of m targets, so as to form Det _q = {{Cia ₁ , Rec ₁ }, {Cla ₂ , Rec ₂ }...{Cla _m , Rec _m }}. Det _q is the target detection element corresponding to the qth video frame frame _q .

In step S202, for each frame in the video image data {frame ₁ , frame ₂ ′...frame _q ′}, use the object segmentation algorithm based on deep learning to segment, and obtain the corresponding object segmentation elements of each video frame. Taking the qth video frame frame _q as an example, use the target detection algorithm based on deep learning to segment, and obtain the category information Lab _n and region information Reg _n of n targets, so as to form Seg _q = {{Lab ₁ , Reg ₁ } , {Lab ₂ , Reg ₂ }...{Lab _n , Reg _n }}. Seg _q is the target segmentation element corresponding to the qth video frame frame _q .

In this embodiment, the target detection element Det _q and the target segmentation element Seq _q corresponding to the qth video frame frame _q form the target element pet _q , Seg _q }. The set of target elements of each video frame is denoted as {{Det ₁ , Seg ₁ }, {Det ₂ , Seg ₂ }…pet _q , Seg _q }}.

In this embodiment, when performing step S3, that is, setting the initial value of the scene reference object according to the video image data, the following steps may be specifically performed:

S301. Setting a specific category of the subject matter;

S302. Select the first several video frames from the video image data;

S303. Respectively delineate corresponding target elements from several selected video frames; the delineated target elements all represent specific types of objects;

S304. Use the intersection of all demarcated target elements as the initial value of the scene benchmark object.

In step S301, taking the traffic scene as an example, since the markings and markings in the road area are fixed, the markings and markings can be used as the salient features of the traffic scene, so as to be used as the standard for scene change detection, so the The specific category is set as "marking line", which can be specifically set as {Cla=Road, Lab=Line}.

In step S302, the first i video frames may be selected from the video image data {frame ₁ , frame ₂ ′...frame _q ′}.

In step S303 , for the i video frames selected in step S302 , draw target elements that can represent specific types of objects (according to step S301 , ie marking lines) from each video frame. Since the target elements in each video frame represent a specific category of objects (marking lines), and the specific category of objects is invariant in a specific scene (road area), the target in each video frame Elements are invariant elements.

In step S304, calculate the intersection set Mark _init ={{Det ₁ , Det ₂ ,...Det _x }, {Seg ₁ , Seg ₂ ,...Seg _y }} for all target elements delineated in step S303, and obtain the scene reference target The initial value of Mark _init .

The principle of step S4 is: to judge the scene change on the real-time continuous video image frame, in the scene change judgment, based on the reference calibration object of the scene invariant element, a scene similarity calculation method is proposed, the principle of the scene similarity is based on the There are fixed elements, and they can be used as a reference calibration object for position matching between consecutive frames. When the scene does not change, the positions of the calibration objects detected in real time by consecutive frames and the reference calibration object have a high degree of coincidence, which means that the scenes are similar When the scene changes, the coincidence degree of the calibration object and the reference calibration object detected in real time by continuous frames is low, or even non-coincident, which means that the scene similarity is low.

Based on the above principle, in step S4, for the video image data {frame ₁ , frame ₂ ′...frame _q ′}, traverse from the first video frame frame ₁ .

计算第1个视频帧frame₁对应的第一场景相似度，式中score₁表示第1个视频帧(当前视频帧)对应的第一场景相似度，Reg_t和Rec_o表示场景基准标的物所在区域，Reg_k表示第1个视频帧(当前视频帧)中目标检测要素所在区域，Rec_l表示第1个视频帧(当前视频帧)中目标分割要素所在区域，n表示第1个视频帧(当前视频帧)中目标检测要素的数量，m表示第1个视频帧(当前视频帧)中分割检测要素的数量；将第1个视频帧对应的第一场景相似度score₁与第一阈值thre₁进行比较，如果score₁>thre₁，表示场景未发生改变(即此时场景仍然是“道路区域”)，即可以转入下一个视频帧即第2个视频帧ftame₂，此时第2个视频帧frame₂取代第1个视频帧frame₁成为当前视频帧，以不变的场景基准标的物，计算第一场景相似度score₂，再判断score₂>thre₁成立(场景未发生改变)还是score₂≤thre₁成立(场景发生改变)，如果仍为score₂>thre₁成立(场景未发生改变)，即可以转入下一个视频帧即第3个视频帧frame₃，继续相同的过程，直至遍历完视频图像数据{frame₁，frame₂′…frame_q′}中全部的视频帧。At the beginning, the current video frame is the first video frame frame ₁ , according to the target element corresponding to the first video frame frame ₁ and the scene reference object (the initial value at this time), according to the formula

Calculate the similarity of the first scene corresponding to the first video frame frame ₁ , where score ₁ represents the similarity of the first scene corresponding to the first video frame (current video frame), and Reg _t and Rec _o represent the location of the scene reference object Region, Reg _k represents the area where the target detection element is located in the first video frame (current video frame), Rec _l represents the area where the target segmentation element is located in the first video frame (current video frame), n represents the first video frame ( The number of target detection elements in the current video frame), m represents the number of segmentation detection elements in the first video frame (current video frame); the first scene similarity score ₁ corresponding to the first video frame and the first threshold value thre ₁ for comparison, if score ₁ >thre ₁ , it means that the scene has not changed (that is, the scene is still "road area" at this time), that is, it can be transferred to the next video frame, which is the second video frame ftame ₂ , at this time the second video frame The first video frame, frame _2, replaces the first video frame, frame ₁ , as the current video frame, and calculates the first scene similarity score ₂ with the same scene reference object, and then judges that score ₂ >thre ₁ holds true (the scene has not changed) Still score ₂ ≤thre ₁ is established (scene changes), if score ₂ >thre ₁ is still established (scene has not changed), that is, you can transfer to the next video frame, which is the third video frame frame ₃ , and continue the same process , until all the video frames in the video image data {frame ₁ , frame ₂ ′… frame _q ′} are traversed.

In step S4, as long as score _{q≤thre 1} occurs in any current video frame (for example, the qth video frame frame _q ), indicating that the scene has changed, then referring to Figure 3, pause and enter the next video frame (q+1th video frame frame), update the scene reference object, and then enter the next video frame (q+1th video frame) after updating the scene reference object. For subsequent video frames (for example, the q+1th video frame), the first scene similarity corresponding to the video frames is calculated using the updated scene reference object.

In step S4, as long as score _q ≤ _{thre 1} occurs in any current video frame (for example, the qth video frame frame _q ), referring to Fig. 3, a sliding window can be set. This sliding window has a fixed size and can select continuous The j video frames of , and the position of the sliding window is variable, that is, the continuous j video frames on the video image data {frame ₁ , frame ₂ ′…frame _q ′} can be selected arbitrarily.

At the beginning, the sliding window can be set at the position near the current video frame (the qth video frame frame _q ), so as to select the continuous j including the current video frame (the qth video frame frame _q ) video frames (the qth video frame frame _q can be anywhere in this continuous j video frames). For these j consecutive video frames, delineate the respective target detection elements {Det ₁ , Der ₂ , ... Det _z } and target segmentation elements {Seg ₁ , Seg ₂ , ... Seg _r } for each video frame, by delineating The intersection of all target elements Mark _temp = {{Det ₁ , Det ₂ , ... Det _z }, {Seg ₁ , Seg ₂ , ... Seq _r }} to obtain the temporary scene reference target Mark _temp .

式中，Reg_t和Rec_o表示临时场景基准标的物所在区域，Reg_k表示作为当前视频帧的第q个视频帧中目标检测要素所在区域，Rec_l表示第q个视频帧中目标分割要素所在区域，n表示第q个视频帧中目标检测要素的数量，m表示第q个视频帧中分割检测要素的数量。After obtaining the temporary scene reference target Mark _temp , according to the target elements corresponding to the temporary scene reference target Mark _temp and the current video frame (the qth video frame frame _q ), calculate the second scene corresponding to the temporary scene reference target Mark _temp Similarity score _q . In this embodiment, the formula for calculating the second scene similarity score _q is

In the formula, Reg _t and Rec _o represent the area where the temporary scene benchmark object is located, Reg _k represents the area where the target detection element is located in the qth video frame as the current video frame, and Rec _l represents the location of the target segmentation element in the qth video frame area, n represents the number of target detection elements in the qth video frame, and m represents the number of segmentation detection elements in the qth video frame.

The second threshold of similarity thre ₂ is set according to different scenes to determine whether the scene stops changing. When the second scene similarity score _q ≤ the second threshold thre ₂ , it means that the scene still changes, and the sliding window is moved to a new position (for example, along the time axis of the video image data {frame ₁ , frame ₂ ... frame _q } direction forward one frame), the moved sliding window can select a new set of continuous j video frames, thereby recalculating the second scene similarity score _q , if the new second scene similarity score _q is still less than or equal to the first If the second threshold value is thre ₂ , the process of moving the sliding window to a new position and so on will be repeated.

As long as the second scene similarity score _q > the second threshold thre ₂ appears, it means that the scene of the latest j video frames will no longer change, then the temporary The scene reference target Mark _temp is determined as the updated scene reference target, and the update of the scene reference target is completed.

The scene change detection method based on video analysis in this embodiment is based on the depth features of the invariant elements in the scene extracted by deep learning, without manual design of depth features, which improves the robustness of scene description features; Updating the scene reference calibration object when the scene changes in the current video frame can make the scene reference calibration object follow the scene change of the video frame and change, thereby maintaining the automatic update of the scene change judgment standard. On the one hand, it is conducive to more accurately judging whether the video frame is Scene changes occur, on the other hand, it is beneficial to improve the automation of scene change detection.

The scene change detection method based on video analysis in this embodiment is to detect scene changes based on continuous frame video data using deep features, which can accurately and quickly perform scene change detection on real-time monitoring videos or video files, and can be used in actual scenes. The scene change detection is realized under various lighting and weather conditions, which enhances the robustness of the algorithm.

In the scene change detection method based on video analysis in this embodiment, the scene element category information and spatial position information extracted from the depth feature are used to match the scene reference calibration object, and the scene change detection is realized according to the difference in position similarity, taking into account In practical applications, the method for updating the scene reference calibration object after the scene changes is more conducive to practical applications.

In this embodiment, a scene change detection system based on video analysis is provided, including:

The first module is used to acquire video image data; the video image data includes a plurality of video frames;

The second module is used to set the initial value of the scene benchmark object according to the video image data;

The third module is used to process the video image data to obtain the corresponding target elements of each video frame;

The fourth module is used to traverse each video frame according to the order in the video image data; for any current video frame in the traversal process, determine the corresponding The first scene similarity, when 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 subsequent video frame The corresponding first scene similarity.

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 Step S4 is executed, so the video analysis-based scene change detection system can execute the video analysis-based scene change detection method, thereby realizing the technical effect of the video analysis-based scene change detection method.

It is possible to write a computer program that executes the video analysis-based scene change detection method in this embodiment, write the computer program into a storage medium or a computer device, and when the computer program is read and run, execute the method in this embodiment. The scene change detection method based on video analysis achieves the same technical effect as the scene change detection method based on video analysis in the embodiment.

It should be noted that, unless otherwise specified, when a feature is called "fixed" or "connected" to another feature, it can be directly fixed and connected to another feature, or indirectly fixed and connected to another feature. on a feature. In addition, descriptions such as up, down, left, and right used in the present disclosure are only relative to the mutual positional relationship of the components of the present disclosure in the drawings. As used in this disclosure, the singular forms "a", "the", and "the" are intended to include the plural forms as well, unless the context clearly dictates otherwise. In addition, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as commonly understood by those skilled in the art. The terms used in the description of this embodiment are only for describing specific embodiments, not for limiting the present invention. The term "and/or" used in this embodiment includes any combination of one or more related listed items.

It should be understood that although the terms first, second, third etc. may be used in the present disclosure to describe various elements, these elements should not be limited by these terms. These terms are only used to distinguish elements of the same type from one another. For example, a first element could be termed a second element, and, similarly, a second element could be termed a first element, without departing from the scope of the present disclosure. The use of any and all examples, or exemplary language ("such as", "such as", etc.) provided in the examples is intended merely to better illuminate the examples of the invention and will not cast a shadow on the scope of the invention unless otherwise claimed impose restrictions.

It should be appreciated that embodiments of the invention may be realized or implemented by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer readable memory. The methods can be implemented in a computer program using standard programming techniques - including a non-transitory computer-readable storage medium configured with a computer program, where the storage medium so configured causes the computer to operate in a specific and predefined manner - according to the specific Methods and Figures described in the Examples. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, the programs can be implemented in assembly or machine language, if desired. In any case, the language may be a compiled or interpreted language. Furthermore, the program can be run on an application specific integrated circuit programmed for this purpose.

Furthermore, operations of processes described in this embodiment may be performed in any suitable order unless otherwise indicated by this embodiment or otherwise clearly contradicted by context. The processes described in this embodiment (or variants and/or combinations thereof) can be executed under the control of one or more computer systems configured with executable instructions, and can be executed as code jointly executed on one or more processors (eg, executable instructions, one or more computer programs, or one or more applications), hardware or a combination thereof. The computer program comprises a plurality of instructions executable by one or more processors.

Further, the method can be implemented in any type of computing platform operably connected to a suitable one, including but not limited to personal computer, minicomputer, main frame, workstation, network or distributed computing environment, stand-alone or integrated computer platform, or communicate with charged particle tools or other imaging devices, etc. 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, optically read and/or written storage medium, RAM, ROM, etc., such that they are readable by a programmable computer, when the storage medium or device is read by the computer, can be used to configure and operate the computer to perform the processes described herein. Additionally, the machine-readable code, or portions thereof, may be transmitted over a wired or wireless network. The invention described in this embodiment includes these and other different types of non-transitory computer-readable storage media when such media include instructions or programs for implementing the steps described above in conjunction with a microprocessor or other data processor. The invention also includes the computer itself when programmed according to the methods and techniques described herein.

Computer programs can be applied to input data to perform the functions described in this embodiment, thereby transforming the input data to generate output data stored to non-volatile memory. Output information may 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 produced on a display.

The above is only a preferred embodiment of the present invention, and the present invention is not limited to the above-mentioned implementation, as long as it achieves the technical effect of the present invention by the same means, within the spirit and principles of the present invention, any Any modification, equivalent replacement, improvement, etc., shall be included within the protection scope of the present invention. Various modifications and changes may be made to the technical solutions and/or implementations within the protection scope of the present invention.

Claims (10)

1. A scene change detection method based on video analysis, characterized in that, the scene change detection method based on video analysis comprises: Obtain video image data; the video image data includes a plurality of video frames; Processing the video image data to obtain target elements corresponding to each of the video frames; According to the video image data, setting the initial value of the scene benchmark object; According to the order in the video image data, traverse each of the video frames; for any current video frame in the traversal process, determine according to the target element corresponding to the current video frame and the scene benchmark object The first scene similarity corresponding to the current video frame, when the first scene similarity is less than the first threshold, update the scene reference object, otherwise keep the scene reference object unchanged; after updating or maintaining The scene benchmark object is used to determine the first scene similarity corresponding to subsequent video frames. 2. the scene change detection method based on video analysis according to claim 1, is characterized in that, described acquisition video image data comprises: Obtain the original video data file; Preprocessing the original video data file to obtain the video image data. 3. the scene change detection method based on video analysis according to claim 2, is characterized in that, described obtaining original video data file, comprises: Obtaining the original video data file by shooting a fixed scene through a camera installed in the scene; or The historical video data files saved by the cameras in the scene are obtained, and the original video data files are obtained. 4. The scene change detection method based on video analysis according to claim 2, wherein the preprocessing of the original video data file includes: converting the image size of the original video data file into a standard size; Converting the image color space of the original video data file into a standard color space. 5. The scene change detection method based on video analysis according to any one of claims 1-3, wherein said setting the initial value of the scene benchmark object according to the video image data includes: Set a specific category of subject matter; selecting the first several video frames from the video image data; Respectively delineating corresponding target elements from the selected several video frames; the delimited target elements all represent the subject matter of the specific category; The intersection of all demarcated target elements is used as the initial value of the scene benchmark object. 6. The scene change detection method based on video analysis according to claim 5, wherein the processing of the video image data to obtain the respective target elements corresponding to each of the video frames comprises: Using a target detection algorithm based on deep learning to detect the video image data and obtain target detection elements corresponding to each of the video frames; The video image data is segmented by using a target segmentation algorithm based on deep learning, and target segmentation elements corresponding to each of the video frames are obtained. 7. The scene change detection method based on video analysis according to claim 6, wherein, according to the target element corresponding to the current video frame and the scene benchmark object, the current video frame is determined The corresponding first scene similarity includes: According to the formula

perform calculations; Wherein, score _q represents the first scene similarity corresponding to the qth video frame as the current video frame, Reg _t and Rec _o represent the area where the scene benchmark object is located, and Reg _k represents the qth video frame The region where the target detection element is located, Rec _l represents the region where the target segmentation element is located in the qth video frame, n represents the number of the target detection element in the qth video frame, and m represents the area in the qth video frame The number of elements detected by the split. 8. The scene change detection method based on video analysis according to claim 7, wherein said updating said scene benchmark object comprises: Setting a sliding window; the sliding window is used to select a number of consecutive video frames at the location; In the vicinity of the current video frame, move the sliding window; for any current position of the sliding window in the moving process, respectively, the sliding window is defined in several consecutive video frames determined by the current position. Corresponding target elements are delineated, wherein the delineated target elements all represent the targets of the specific category, and the intersection of all delineated target elements is used as the temporary scene reference target, according to the current video frame corresponding The target element and the temporary scene reference object, determine the second scene similarity corresponding to the temporary scene reference object, when the second scene similarity is greater than a second threshold, stop moving the sliding window, The temporary scene reference object is used as the new scene reference object, otherwise, the sliding window continues to be moved. 9. The scene change detection method based on video analysis according to claim 8, wherein the temporary scene is determined according to the target element corresponding to the current video frame and the temporary scene benchmark object The similarity of the second scene corresponding to the benchmark object, including: According to the formula

perform calculations; Wherein, score' _q represents the similarity of the second scene, Reg' _t and Rec' _o represent the area where the temporary scene reference object is located, and Reg _k represents the target detection in the qth video frame as the current video frame The area where the element is located, Rec _l represents the area where the target segmentation element is located in the qth video frame, n represents the number of the target detection element in the qth video frame, and m represents the segmentation detection element in the qth video frame quantity. 10. A scene change detection system based on video analysis, characterized in that, the scene change detection system based on video analysis comprises: The first module is used to acquire video image data; the video image data includes a plurality of video frames; The second module is used to set the initial value of the scene benchmark object according to the video image data; The third module is used to process the video image data, and acquire target elements corresponding to each of the video frames; The fourth module is used to traverse each of the video frames according to the order in the video image data; for any current video frame in the traversal process, according to the target element corresponding to the current video frame and the Scene reference object, determine the first scene similarity corresponding to the current video frame, when the first scene similarity is less than the first threshold, update the scene reference object, otherwise maintain the scene reference object change; the updated or maintained scene reference object is used to determine the first scene similarity corresponding to subsequent video frames.

Images