A monitoring and analysis system and method based on artificial intelligence

Abstract

The present invention discloses a monitoring and analysis system and method based on artificial intelligence, and relates to the field of monitoring and analysis technology; the method comprises the following steps: establishing a template library, integrating video materials used as templates into the template library, and performing classification processing, wherein the template library comprises a plurality of category libraries, each category library corresponds to at least one template video segment, and each category library comprises a plurality of sub-video segments, and the sub-video segments of the same category library are all provided with serial number codes, and the serial number codes are sorted based on the occurrence order of an event; obtaining monitoring videos, performing multi-segment processing on the monitoring videos, and obtaining segmented videos; matching the video contents of the plurality of segmented videos with the template video contents of the template video segments, respectively, and obtaining target video segments with high matching degree. The present invention performs intelligent segmentation processing, template comparison, and multi-level matching analysis on the monitoring videos, and the system can efficiently and accurately complete the monitoring event analysis tasks in complex scenes in a relatively short time.

Description

Technical Field

[0001] The present invention relates to the field of monitoring and analysis technology, and in particular to a monitoring and analysis system and method based on artificial intelligence. Background Art

[0002] With the development of science and technology, artificial intelligence technology has been widely used in various fields, such as security field, etc. However, the existing monitoring analysis usually relies on manual operation, which has problems such as low efficiency and low accuracy. Therefore, how to use artificial intelligence technology to realize the automation and intelligent analysis of surveillance video has become a technical problem that needs to be solved urgently.

[0003] Although existing video analysis technology has certain analytical capabilities, it lacks the ability to accurately identify specific events or behavioral patterns, and its ability to analyze long videos is insufficient, and needs to be improved. Summary of the invention

[0004] The purpose of the present invention is to solve the shortcomings of the prior art and to propose a monitoring and analysis system and method based on artificial intelligence.

[0005] In order to achieve the above object, the present invention adopts the following technical solutions:

[0006] A monitoring and analysis system based on artificial intelligence, comprising:

[0007] Video preprocessing module, template library management module, video segmentation module, matching and comparison module, category library calling module, small paragraph re-segmentation module, multi-level matching module and result output module;

[0008] The video preprocessing module is responsible for performing noise filtering and image enhancement preprocessing operations on the acquired surveillance video;

[0009] The template library management module is used to establish and maintain the template library, and integrate video materials into template video segments of different categories;

[0010] The video segmentation module is used to process the surveillance video into multiple segments for subsequent matching and comparison;

[0011] The matching and comparison module is used to compare the video content of the segmented video with the template video content of the template video segment to find a target video segment with a high matching degree;

[0012] The category library calling module is used to call out the target category library corresponding to the target video segment to prepare for further analysis;

[0013] The small paragraph re-segmentation module is used to re-segment the target video segment with high matching degree into small paragraphs to refine the matching granularity;

[0014] The multi-level matching module is used to perform multi-level matching between the sub-target segments and the sub-video segments to find the best matching pair;

[0015] The result output module is used to obtain comprehensive matching results, sort the comprehensive matching results according to their matching degree, and output the final analysis results.

[0016] Preferably: the video preprocessing module performs preprocessing operations on the acquired surveillance video, including noise filtering and image enhancement; improving video quality and reducing noise interference; the preprocessing process also includes color adjustment and brightness contrast optimization operations;

[0017] The video preprocessing module eliminates Gaussian noise in the surveillance video image based on Gaussian filtering; the surveillance video image is smoothed by calculating the weighted average of each pixel, and the weight is determined by the Gaussian kernel, specifically:

[0018]

[0019] Among them, σ is the standard deviation of the Gaussian filter, which controls the degree of smoothness; x and y are the coordinates of the pixel points in the image of the surveillance video.

[0020] Preferably, the video preprocessing module adjusts the grayscale histogram of each frame image in the surveillance video through the histogram equalization technology to improve the global contrast of each frame image in the surveillance video. The specific calculation is:

[0021]

[0022] Wherein, L is the number of gray levels; pR(w) is the probability density function of the gray level; r is the gray value of the original image of the surveillance video; s is the gray value after equalization; k is a constant, ensuring that 0≤s≤L.

[0023] Preferably: the template library management module is used to establish and maintain the template library; integrate video materials into template video segments of different categories, and classify these template video segments; template video segments of the same category library are provided with serial number codes, and the serial number codes are sorted based on time sequence.

[0024] Preferably: the video segmentation module performs multi-segment processing on the surveillance video, dividing the surveillance video into a plurality of segmented videos;

[0025] The matching and comparison module compares the video content of the segmented video with the template video content of the template video segment, calculates the video similarity between the video content of the segmented video and the template video content of the template video segment, and selects the target video segment from the template video segment based on the video similarity.

[0026] Preferably: the category library calling module calls out the target category library corresponding to the target video segment according to the matching result; when a target video segment successfully matches a specific template video segment in the divided target video segment, the system automatically calls the category library to which the successfully matched template video segment belongs as the target category library corresponding to the target video segment;

[0027] The small segment re-segmentation module is based on the successfully matched target video segment, and further subdivides it into smaller video clips of the same length to obtain sub-target segments, further refining the analysis granularity.

[0028] Preferably: the multi-level matching module finds the best matching pairs through a matching algorithm, and then expands and matches other small paragraphs on both sides based on these best matching pairs.

[0029] Preferably: the specific matching method of the multi-level matching module is:

[0030] S11: extracting a first feature vector that can represent the video content from the sub-target segment, integrating a plurality of first feature vectors to obtain a first feature vector set, and using one or more methods including a color histogram, a gradient direction histogram, and a scale-invariant feature transform to capture edge, texture, and color information of the image;

[0031] S12: The sub-video segment corresponding to each template video segment in the template library is also subjected to the same feature extraction steps as S11 to obtain a corresponding second feature vector set;

[0032] S13: Calculate the similarity of feature vectors between the first feature vector set and the second feature vector set using cosine similarity or Euclidean distance;

[0033] S14: Matching is performed based on feature similarity through nearest neighbor matching or K nearest neighbor matching to determine the final best matching pair.

[0034] Preferably: also include:

[0035] The target tracking module is used to track a specific target in the video frames of the continuous surveillance video;

[0036] Through Mean-Shift, parameter-free kernel density estimation based on color histogram is used to achieve real-time tracking of the target. Specifically:

[0037]

[0038] Among them, h is the bandwidth, which controls the window size; g is the kernel function; is the weight function; is the initial position of the target; The new location of the target.

[0039] A monitoring and analysis method based on artificial intelligence comprises the following steps:

[0040] S1: Establish a template library, integrate the video materials used as templates into the template library, and perform classification processing. The template library includes multiple category libraries, each category library corresponds to at least one template video segment, and each category library includes multiple sub-video segments. The sub-video segments of the same category library are all set with serial number codes, and the serial number codes are sorted based on the occurrence order of an event;

[0041] S2: Obtain surveillance video, and segment the surveillance video into multiple segments to obtain segmented videos;

[0042] S3: Match the video contents of the multiple segmented videos with the template video contents of the template video segment respectively to obtain a target video segment with a high matching degree, and call out a target category library corresponding to the target video segment;

[0043] S4: The target video segment with a high matching degree is segmented again into small segments to obtain sub-target segments; the divided sub-target segments are compared with the sub-video segments of the corresponding category library to find the pair with the highest matching degree;

[0044] S5: Taking the sub-target segment and sub-video segment with the highest matching degree as the benchmark, perform expansion matching to both sides;

[0045] S6: Obtain comprehensive matching results of each sub-target segment and its matching sub-video segment, sort the matching results in descending order according to their matching degree, and finally output the analysis results.

[0046] The beneficial effects of the present invention are:

[0047] 1. The present invention performs intelligent segmentation processing, template comparison and multi-level matching analysis on surveillance videos, so that the system can efficiently and accurately complete the surveillance event analysis tasks in complex scenes in a relatively short time.

[0048] 2. The present invention establishes a structured template library that can effectively manage and retrieve template video segments of different categories, thereby enhancing the flexibility and applicability of the system; through reasonable template library design and classification management, the system can quickly locate and match corresponding monitoring events, thereby improving the overall analysis efficiency.

[0049] 3. The present invention adopts a multi-level matching strategy, and the system can analyze the surveillance video segments layer by layer, improving the detail capture capability and matching accuracy; the comprehensive matching score shows high stability and reliability in various test scenarios, proving the superiority of the multi-level matching strategy. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1A flowchart of a monitoring and analysis method based on artificial intelligence proposed by the present invention;

[0051] Figure 2 This is a flow chart of the collaboration of various modules in the monitoring and analysis method based on artificial intelligence proposed by the present invention. DETAILED DESCRIPTION

[0052] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0053] Embodiment 1:

[0054] An artificial intelligence-based monitoring and analysis system, such as Figure 2 As shown, it includes: video preprocessing module, template library management module, video segmentation module, matching and comparison module, category library calling module, small paragraph re-segmentation module, multi-level matching module and result output module;

[0055] The video preprocessing module is responsible for performing preprocessing operations such as noise filtering and image enhancement on the input surveillance video;

[0056] The template library management module is used to establish and maintain the template library and integrate video materials into template video segments of different categories;

[0057] The video segmentation module is used to process the surveillance video into multiple segments for subsequent matching and comparison;

[0058] The matching and comparison module is used to compare the segmented video content with the template video segment to find the video segments with high matching degree;

[0059] The category library calling module is used to call out the corresponding category library according to the matching results to prepare for further analysis;

[0060] The small segmentation module is used to segment the video segments with high matching degree into small segments again to refine the matching granularity;

[0061] The multi-level matching module is used to perform multi-level matching between sub-target segments and sub-video segments to find the best matching pair;

[0062] The result output module is used to obtain the matching results, sort them according to the matching degree, and output the final analysis results.

[0063] Specific:

[0064] ①Video preprocessing module:

[0065] The main function of the video preprocessing module is to perform preprocessing operations on the input surveillance video, including noise filtering, image enhancement, etc. Through these preprocessing steps, the video quality can be improved, noise interference can be reduced, and a good foundation can be laid for subsequent analysis. The preprocessing process can also include operations such as color adjustment, brightness and contrast optimization, making the video clearer and helping to improve matching accuracy;

[0066] Based on Gaussian filtering, Gaussian noise in the image is eliminated; the image is smoothed by calculating the weighted average of each pixel, and the weight is determined by the Gaussian kernel, specifically:

[0067]

[0068] Where σ is the standard deviation of the Gaussian filter, which controls the degree of smoothing; x and y are the coordinates of the pixel points in the image;

[0069] Through the histogram equalization technology, the grayscale histogram of each frame image in the surveillance video is adjusted to improve the global contrast of each frame image in the surveillance video. The specific calculation is:

[0070]

[0071] Where L is the number of gray levels; pR(w) is the probability density function of the gray level; r is the gray value of the original image; s is the gray value after equalization; k is a constant, ensuring that 0≤s≤L.

[0072] ②Template library management module:

[0073] The template library management module is used to establish and maintain the template library. It is responsible for integrating video materials into template video segments of different categories and classifying these video segments. The sub-video segments of the same category library are set with serial number codes and sorted based on the order of event occurrence. Through reasonable template library management, a reliable reference standard can be provided for subsequent matching analysis to ensure the accuracy and consistency of the analysis.

[0074] ③Video segmentation module:

[0075] The task of the video segmentation module is to process the surveillance video into multiple segments, dividing it into multiple smaller segments, such as Ai, Bi, Ci, etc. This process helps to extract key frames and important information, reduce the amount of data and computational complexity, and improve system processing efficiency. A reasonable segmentation strategy can significantly improve the effect of subsequent matching analysis.

[0076] ④Matching and comparison module:

[0077] The core function of the matching and comparison module is to compare the segmented video content with the template video segment to find the segments with high matching degree. By calculating the similarity or feature matching degree, the qualified video segments can be screened out. For example, the Ai and Ci segments may be judged to have a high matching degree. The accuracy of this module directly affects the performance and reliability of the entire system.

[0078] ⑤Category library calling module:

[0079] The category library calling module is used to call out the corresponding category library according to the matching results. When a video clip successfully matches a specific template video segment, the system automatically calls the category library to which the clip belongs. This step ensures the pertinence and effectiveness of subsequent analysis, allowing the monitoring system to be flexibly adjusted according to different scenarios and needs.

[0080] ⑥Small paragraph subdivision module:

[0081] The function of the small segment re-segmentation module is to subdivide the video segments with high matching degree into smaller segments, such as dividing Ai into A1, A2, etc., and dividing Ci into C1, C2, etc. Through this secondary segmentation processing, the analysis granularity can be further refined, more detailed information can be captured, and the matching accuracy and reliability can be improved. The design of this module takes into account the detail differences in complex scenes.

[0082] ⑦Multi-level matching module:

[0083] The multi-level matching module is the key part of multi-level matching between sub-target segments and sub-video segments. Through the matching algorithm, the best matching pair can be found. For example, A5 and Px have the highest matching degree, and C4 and Py have the highest matching degree. Then, based on these best matching pairs, other small segments are expanded to both sides. This multi-level matching strategy significantly improves the flexibility and accuracy of the system.

[0084] ⑧Result output module:

[0085] The result output module is responsible for calculating the comprehensive matching score, sorting in descending order according to the score size, and finally outputting the analysis results. For example, if the total matching score of Ai is higher than Ci, the category library category to which the Ai paragraph belongs will be displayed first during output. In this way, users can intuitively understand the analysis results of the surveillance video and promptly discover and handle abnormal situations. The design of this module focuses on user experience and practical application effects.

[0086] In addition, the system includes:

[0087] Target tracking module,The target tracking module is used to track specific targets, such as pedestrians, vehicles, etc., in continuous video frames;

[0088] Through Mean-Shift, parameter-free kernel density estimation based on color histogram is used to achieve real-time tracking of the target. Specifically:

[0089]

[0090] Among them, h is the bandwidth, which controls the window size; g is the kernel function; is the weight function; is the initial position; for the new location.

[0091] Experiment and verification:

[0092] ① Experimental environment construction

[0093] In order to verify the effectiveness and superiority of the monitoring and analysis method and system based on artificial intelligence proposed in this invention, an experimental environment including a variety of hardware devices and software platforms was constructed. The hardware includes a high-resolution camera, a high-performance computer, and a stable network connection device. In terms of the software platform, the TensorFlow framework was selected to train and test the deep learning model, OpenCV was used for image processing, and the MySQL database was combined for data storage and management. The experimental environment simulated a real monitoring scenario to ensure the accuracy and reliability of the experimental results.

[0094] ②Experimental data preparation

[0095] A large number of representative data sets were selected from actual surveillance videos, covering various typical surveillance scenes and event types. The data sets include surveillance images in different time periods and weather conditions, as well as various behavior patterns (such as normal walking, running, vehicle traffic, etc.). To ensure the diversity and comprehensiveness of the data, public surveillance video data sets (such as the UCSDAvenue data set) were introduced. All data were preprocessed by denoising and standardization before use.

[0096] ③ Experimental design

[0097] The experiment is divided into two parts: training and testing. In the training phase, the deep learning model is trained using annotated data sets to enable it to learn and recognize different categories of events and behavior patterns. In the testing phase, unlabeled surveillance videos are input into the system and the accuracy and efficiency of the system are verified by matching them with the pre-established template library. Specific test indicators include accuracy, recall, F1 score, and processing speed.

[0098] ④ Experimental steps and parameter settings

[0099] Training steps:

[0100] The data set is divided into a training set and a validation set.

[0101] Build and train neural network models using deep learning frameworks such as TensorFlow.

[0102] Adjust hyperparameters (such as learning rate, batch size, etc.) through cross-validation to find the optimal model.

[0103] Save the trained model weights for use in the testing phase.

[0104] Test steps:

[0105] Input the preprocessed surveillance video into the system.

[0106] Use the trained model to segment and extract features from the input video.

[0107] The corresponding category library is retrieved from the template library according to the feature matching results.

[0108] The video segments with high matching degree are re-segmented and multi-level matched.

[0109] Calculate the comprehensive matching score and sort in descending order, and output the final analysis results.

[0110] Parameter settings:

[0111] Learning rate: 0.001;

[0112] Batch size: 32

[0113] Optimizer: Adam;

[0114] Loss function: cross entropy loss;

[0115] Template library size: dynamically adjusted according to experimental requirements;

[0116] Matching threshold: 0.8 (set based on experience);

[0117] The formula for calculating the comprehensive matching score is: ∑ (matching score of a single paragraph × weight coefficient);

[0118] ⑤Data analysis and discussion of results

[0119] During the experiment, the performance of various indicators was recorded and analyzed. The experimental results show that the performance of this system is better than traditional methods in various monitoring scenarios. The following are the specific data analysis results:

[0120] Precision and recall: average precision 92.7%; average recall 89.4%;

[0121] F1 score: 91.0%

[0122] Processing speed: average number of frames per minute: 60 frames (at 1080p resolution);

[0123] Response time: within 500 milliseconds on average;

[0124] Comprehensive matching score: The highest comprehensive matching score is 95.7 (out of 100); the lowest comprehensive matching score is 78.4 (out of 100).

[0125] Through comparative experiments with traditional methods (such as rule-based monitoring systems), it is found that this system has higher accuracy and robustness when dealing with complex scenes and changing behavior patterns. Especially in night monitoring or low-light conditions, this system can still maintain good performance. In addition, due to the introduction of template libraries and small segment re-segmentation processing mechanisms, the system's ability to capture details is significantly enhanced, and it can more accurately identify and classify different types of events.

[0126] The above experimental results show that the artificial intelligence-based monitoring and analysis method and system proposed in the present invention have significant advantages and potential in practical applications.

[0127] Embodiment 2:

[0128] A monitoring and analysis method based on artificial intelligence, such as Figure 1 As shown, the following steps are included:

[0129] S1: Establish a template library, integrate the video materials used as templates into the template library, and perform classification processing. The template library includes multiple category libraries, each category library corresponds to at least one template video segment, and each category library includes multiple sub-video segments. The sub-video segments of the same category library are all set with serial number codes. The serial number codes are sorted based on the order of occurrence of an event, for example, divided into P1, P2, P3, ..., Pn in sequence;

[0130] S2: Process the surveillance video into multiple segments, and divide them into Ai, Bi, Ci, ..., Ni in sequence;

[0131] S3: Compare the content of each segmented video with the template video segment to find the video segments with high matching degree; for example, Ai and Ci are two videos with matching degree higher than a set value;

[0132] S4: after obtaining the video with a high matching degree, calling out the corresponding category library according to the matching result;

[0133] S5: Segment each video segment with a high matching degree into small segments again, and divide them into A1, A2, ..., An, C1, C2, ..., Cn in sequence;

[0134] S6: Compare the divided sub-target segments with the sub-video segments of the corresponding category library to find the pair with the highest matching degree. For example, A5 and a sub-video segment Px have the highest matching degree, and C4 and a sub-video segment M have the highest matching degree.

[0135] S7: Based on the pair with the highest matching degree, expand the matching to both sides, that is, A1 matches Px-4, A2 matches Px-3, A3 matches Px-2, A4 matches Px-1, A5 matches Px, A6 matches Px+1, ..., An matches Px+n-5;

[0136] S8: Calculate the comprehensive matching results between each sub-target segment and its matching sub-video segment to obtain a total matching score;

[0137] S9: Sort in descending order according to the total matching score and output the analysis results. For example, if the final total matching score of paragraph Ai is greater than the total matching score of paragraph Ci, then when outputting the results, the category of the category library corresponding to paragraph Ai is displayed, and its priority is higher than the category of the category library corresponding to paragraph Ci.

[0138] The above description is only a preferred specific implementation manner of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can make equivalent replacements or changes according to the technical scheme and inventive concept of the present invention within the technical scope disclosed by the present invention, which should be covered by the protection scope of the present invention.

Claims

1. A monitoring and analysis system based on artificial intelligence, characterized in that: It includes video preprocessing module, template library management module, video segmentation module, matching and comparison module, category library calling module, small paragraph re-segmentation module, multi-level matching module and result output module; The video preprocessing module is used to perform noise filtering and image enhancement preprocessing operations on the acquired surveillance video; The template library management module is used to establish and maintain the template library, and integrate video materials into template video segments of different categories; The video segmentation module is used to process the surveillance video into multiple segments; The matching and comparison module is used to compare the video content of the segmented video with the template video content of the template video segment to find a target video segment with a high matching degree; The category library calling module is used to call out the target category library corresponding to the target video segment to prepare for further analysis; The small paragraph re-segmentation module is used to re-segment the target video segment with high matching degree into small paragraphs to refine the matching granularity; The multi-level matching module is used to perform multi-level matching between the sub-target segments and the sub-video segments to find the best matching pair; The result output module is used to obtain comprehensive matching results, sort the comprehensive matching results according to their matching degree, and output the final analysis results; The multi-level matching module finds the best matching pairs through a matching algorithm, and then expands and matches other small paragraphs on both sides based on these best matching pairs; The specific matching method of the multi-level matching module is: S11: extracting first feature vectors that can represent video content from a plurality of sub-target segments, integrating the plurality of first feature vectors to obtain a first feature vector set, and capturing edge, texture, and color information of the image using one or more methods including a color histogram, a gradient direction histogram, and a scale-invariant feature transform; S12: The sub-video segment corresponding to each template video segment in the template library is also subjected to the same feature extraction steps as S11 to obtain a corresponding second feature vector set; S13: Calculate the feature similarity between the first feature vector set and the second feature vector set using cosine similarity or Euclidean distance; S14: Matching is performed based on feature similarity through nearest neighbor matching or K nearest neighbor matching to determine the final best matching pair.

2. The monitoring and analysis system based on artificial intelligence according to claim 1, characterized in that: The video preprocessing module performs preprocessing operations on the acquired surveillance video, including noise filtering and image enhancement; Improve video quality and reduce noise interference; The pre-processing process also includes color adjustment, brightness and contrast optimization operations; The video preprocessing module eliminates Gaussian noise in the surveillance video image based on Gaussian filtering; the surveillance video image is smoothed by calculating the weighted average of each pixel, and the weight is determined by the Gaussian kernel, specifically: ； Among them, σ is the standard deviation of the Gaussian filter, which controls the degree of smoothness; x and y are the coordinates of the pixel points in the image of the surveillance video.

3. The monitoring and analysis system based on artificial intelligence according to claim 2 is characterized in that: The video preprocessing module adjusts the grayscale histogram of each frame image in the surveillance video through the histogram equalization technology to improve the global contrast of each frame image in the surveillance video. The specific calculation is: ； Wherein, L is the number of gray levels; pR(w) is the probability density function of the gray level; r is the gray value of the original image of the surveillance video; s is the gray value after equalization; k is a constant, ensuring that 0≤s≤L.

4. The monitoring and analysis system based on artificial intelligence according to claim 1, characterized in that: The template library management module is used to establish and maintain the template library; integrate video materials into template video segments of different categories, and classify these template video segments; template video segments of the same category library are set with serial number codes, and the serial number codes are sorted based on time sequence.

5. The monitoring and analysis system based on artificial intelligence according to claim 1, characterized in that: The video segmentation module performs multi-segment processing on the surveillance video, dividing the surveillance video into a plurality of segmented videos; The matching and comparison module compares the video content of the segmented video with the template video content of the template video segment, calculates the video similarity between the video content of the segmented video and the template video content of the template video segment, and selects the target video segment from the template video segment based on the video similarity.

6. The monitoring and analysis system based on artificial intelligence according to claim 1, characterized in that: The category library calling module calls out the target category library corresponding to the target video segment according to the matching result; when a target video segment successfully matches the template video segment after being divided, the system automatically calls the category library to which the successfully matched template video segment belongs as the target category library corresponding to the target video segment; The small segment re-segmentation module is based on the successfully matched target video segment, and further subdivides it into smaller video clips of the same length to obtain sub-target segments, further refining the analysis granularity.

7. The monitoring and analysis system based on artificial intelligence according to claim 1 is characterized in that: Also includes: The target tracking module is used to track the target in the video frames of the continuous surveillance video; Through Mean-Shift, parameter-free kernel density estimation based on color histogram is used to achieve real-time tracking of the target. Specifically: ； Among them, h is the bandwidth, which controls the window size; g is the kernel function; is the weight function; is the initial position of the target; The new location of the target.

8. A monitoring and analysis method based on artificial intelligence, applied to a monitoring and analysis system based on artificial intelligence as claimed in any one of claims 1 to 7, characterized in that: The steps include: S1: Establish a template library, integrate the video materials used as templates into the template library, and perform classification processing. The template library includes multiple category libraries, each category library corresponds to at least one template video segment, and each category library includes multiple sub-video segments. The sub-video segments of the same category library are all set with serial number codes, and the serial number codes are sorted based on the occurrence order of an event; S2: Obtain surveillance video, and segment the surveillance video into multiple segments to obtain segmented videos; S3: Match the video contents of the multiple segmented videos with the template video contents of the template video segment respectively to obtain a target video segment with a high matching degree, and call out a target category library corresponding to the target video segment; S4: The target video segment with a high matching degree is segmented again into small segments to obtain sub-target segments; the divided sub-target segments are compared with the sub-video segments of the corresponding category library to find the pair with the highest matching degree; S5: Taking the sub-target segment and sub-video segment with the highest matching degree as the benchmark, perform expansion matching to both sides; S6: Obtain comprehensive matching results of each sub-target segment and its matching sub-video segment, sort the comprehensive matching results in descending order according to their matching degree, and finally output the analysis results.

Images