Security box video monitoring system and method with intelligent discrimination function

Abstract

The application discloses a security box video monitoring system and method with intelligent discrimination function, and relates to the technical field of video monitoring. The system can dynamically extract the trend of the image frame brightness gradient change rate set by continuously collecting images and constructing the image frame average brightness sequence and the image frame time sequence. The system uses the abnormal occlusion image frame data set as the input basis, combines the spatial change trend of the image frame brightness gradient change rate set and the optical flow displacement vector set, and outputs the occlusion behavior possibility score to realize the probability determination of the occlusion intention, avoid the misjudgment of the incidental non-human interference, performs the edge structure analysis in the optical flow disturbance concentrated area, generates the optical flow disturbance concentrated area boundary contour morphological feature data set, calculates the occlusion area proportion, and then fuses the occlusion behavior possibility score to form the optical flow disturbance concentrated area feature fusion input set. Finally, the intelligent classification is realized through the interference behavior discrimination label.

Description

Technical Field

[0001] The present invention relates to the technical field of video surveillance, and in particular to a security box video surveillance system and method with intelligent discrimination function. Background Art

[0002] As information technology and urban infrastructure continue to converge, video surveillance, a crucial component of intelligent security technology, is evolving from traditional image acquisition to intelligent processing capabilities, including perception, analysis, and judgment. In particular, within the security industry's niche sectors, edge security equipment, due to its flexibility and low deployment costs, is increasingly being deployed in unmanned, complex, or remote locations. Security boxes are a typical example of this type of device, typically integrating multiple system modules, including a camera module, power supply, and communications modules, offering ease of deployment, remote control, and automated monitoring.

[0003] While security boxes have significantly improved their deployability and monitoring coverage in unmanned environments, they still face numerous challenges in practical operation, particularly in terms of video capture accuracy and stability. A typical issue is monitoring failure or misjudgment caused by lighting variations and occlusion interference. Current mainstream video surveillance algorithms still rely heavily on brightness threshold sensing or fixed occlusion outline templates to detect anomalies, such as lens obstruction, overexposure, and strong nighttime light interference. However, these methods have significant limitations: they often cannot accurately distinguish between natural lighting changes, such as sunlight movement and cloud cover, and artificial occlusions, such as hands or foreign objects. This lack of intelligent discrimination makes monitoring strategies prone to false alarms, missed alarms, or delayed processing, reducing the overall practicality of security systems. Summary of the Invention

[0004] In view of the deficiencies in the prior art, the present invention provides a security box video monitoring system and method with intelligent identification function, which solves the problems mentioned in the background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: a security box video monitoring method with intelligent identification function, comprising the following steps:

[0006] S1. Continuously capture on-site image data through the security box camera module to form a video image frame sequence. After extracting the average brightness value, form an image frame average brightness sequence. Simultaneously, record the acquisition time of the image frame to form an image frame time sequence. Calculate the image frame brightness gradient change rate and output the image frame brightness gradient change rate set.

[0007] S2. Performing dynamic trend analysis on the image frame brightness gradient change rate set to identify fluctuation states in consecutive frames and constructing an abnormal occlusion image frame data set;

[0008] S3, processing the abnormal occlusion image frame data set frame by frame to generate an optical flow displacement vector set, and outputting an occlusion behavior possibility score based on the spatial distribution trend of the image frame brightness gradient change rate set;

[0009] S4, performing edge extraction processing on the optical flow disturbance concentrated area, outputting the occlusion area ratio, and then merging the occlusion behavior possibility score, the occlusion area boundary contour morphological feature dataset, and the occlusion area ratio into an occlusion area feature fusion input set;

[0010] S5. Execute multi-feature fusion judgment logic on the occlusion area feature fusion, output an interference behavior discrimination label, and perform a classification response action according to the judgment type corresponding to the interference behavior discrimination label.

[0011] Preferably, said S1 includes S11 and S12;

[0012] S11, using the camera module integrated in the security box, continuously collects real-time image data of the target monitoring area at preset time intervals, and stores it in chronological order to form a video image frame sequence;

[0013] For each frame of the video image frame sequence, performing an image brightness information extraction operation to obtain an average pixel brightness value of the image frame; combining the average brightness values ​​of all image frames in frame order to form an image frame average brightness sequence;

[0014] At the same time, during the image acquisition process, the actual acquisition time information corresponding to each frame of image is recorded to generate an image frame time series;

[0015] The image brightness information is extracted by using the image frame grayscale channel average value method.

[0016] Preferably, S12, based on the index relationship between the image frame average brightness sequence and the image frame time sequence, performing differential calculation on the brightness change values ​​between adjacent image frames, and combining the acquisition time interval to obtain the rate of image frame brightness change per unit time, representing the image frame brightness gradient change rate;

[0017] The brightness gradient change rates corresponding to all image frames are stored in time sequence to form a set of image frame brightness gradient change rates;

[0018] The index relationship specifically indicates that a one-to-one correspondence is established between the image frame average brightness sequence and the image frame time sequence through the image frame index numbers.

[0019] Preferably, S2 comprises S21 and S22;

[0020] S21, performing sequence trend analysis on the set of image frame luminance gradient change rates, using a sliding time window mechanism, jointly calculating the first-order difference of the image frame luminance gradient change rate in each local window and the local statistical variance, when the continuous image frames simultaneously satisfy the following two conditions in the sliding window:

[0021] Condition one: the first-order difference exceeds the preset mutation slope threshold;

[0022] Condition two: the local statistical variance exceeds the preset severe fluctuation threshold;

[0023] Then mark the continuous image frames as suspected occlusion interference image frames, and include them in the suspected occlusion interference image frame set;

[0024] Wherein, the first-order difference of the image frame luminance gradient change rate in each local window reflects the change slope; the local statistical variance of the image frame luminance gradient change rate in each local window reflects the fluctuation severity.

[0025] Preferably, S22, for each frame image in the suspected occlusion interference image frame set, extract the image content in each frame image in the video image frame sequence, and obtain the original gray image of the image content as the analysis target; based on the original gray image, according to the luminance difference between the frame and its adjacent frame in the average luminance sequence of the image frame, calculate the gray scale change amplitude of each pixel point, and form a pixel-level gray scale difference matrix;

[0026] Map the pixel-level gray scale difference matrix to the original image space to generate an image luminance change distribution map, divide the entire image into two structural blocks of edge region and center region according to the preset spatial grid, count the number of pixel points with luminance change intensity greater than the set minimum change threshold in each region, calculate the proportion of the number of pixel points relative to the area of the region, and take it as the edge region luminance change density and the center region luminance change density respectively, and then calculate the luminance change density ratio of the edge region luminance change density and the center region luminance change density;

[0027] When the luminance change density ratio exceeds the preset edge concentration change threshold, it is determined that the luminance mutation concentration occurs at the image edge, which meets the occlusion behavior characteristic;

[0028] Mark the image frames that meet the occlusion behavior characteristic condition as abnormal occlusion image frames, and aggregate and build an abnormal occlusion image frame data set.

[0029] Preferably, S3 comprises S31 and S32;

[0030] S31, for each frame image in the set of abnormal occlusion image frame data, combined with the previous frame and the next frame image in the sequence of video image frames, perform pixel-by-pixel dense optical flow calculation processing, the dense optical flow calculation processing is processed by using Farneback optical flow algorithm, calculate the displacement vector of each pixel between adjacent frames, represent the corresponding optical flow vector distribution of image frame in time sequence;

[0031] The displacement vector obtained from each frame image is arranged according to the frame number, and the optical flow displacement vector set is generated by summarizing. Based on the optical flow displacement vector set, the optical flow amplitude value and direction vector of each pixel point are extracted, and the optical flow intensity map and optical flow direction map of the whole image are constructed. The optical flow intensity map is segmented by threshold value, the low intensity optical flow area is removed, and the high amplitude area is reserved. Then, the direction vector of the pixel in the reserved area is executed regional angle consistency calculation, the regional angle consistency calculation includes using density clustering algorithm to calculate the regional angle consistency of the pixel group with consistent motion direction, and the local consistency feature and feature condition in the pixel group are matched and marked as candidate disturbance area. The candidate disturbance area is judged according to the following three feature conditions, and the area meeting the conditions is marked as the optical flow disturbance concentrated area;

[0032] Feature condition one: the optical flow intensity in the local area is greater than 1.5 times of the average intensity of the whole image;

[0033] Feature condition two: the optical flow direction angle variance within the region is less than the preset direction consistency threshold;

[0034] Feature condition three: the number of pixels in the region exceeds the minimum disturbance region proportion threshold of the total pixel area of the image;

[0035] When the candidate disturbance area meeting the feature condition one, the feature condition two and the feature condition three is determined as the optical flow disturbance concentrated area, and the space boundary and the center coordinates are recorded.

[0036] Preferably, S32, based on the optical flow disturbance concentrated area and the set of image frame brightness gradient change rate, a spatial position mapping operation is performed, the spatial position mapping operation is based on the image frame space coordinate pair, the boundary range of the optical flow disturbance concentrated area and the brightness mutation area in the set of image frame brightness gradient change rate are analyzed by pixel level coordinates, and the spatial intersection is determined;

[0037] According to the spatial intersection coincidence degree of the optical flow disturbance concentrated area and the brightness mutation area, combined with the area proportion of the optical flow disturbance area in the image frame, an occlusion behavior scoring function is constructed, and the occlusion behavior possibility score corresponding to the image frame is output, which is used to quantify the possibility of the occlusion behavior.

[0038] Preferably, the S4 comprises S41 and S42;

[0039] S41. For the optical flow disturbance concentrated area, corresponding image frames in the video image frame sequence are obtained, and a grayscale image of the image frame is extracted as an input image for edge detection. An image gradient edge extraction algorithm is used to process the grayscale image to generate an edge binary image corresponding to the optical flow disturbance concentrated area for identifying boundary structures.

[0040] In the edge binary image, a contour tracing operation is performed based on pixel connectivity to identify a closed boundary path. When a contour structure exists whose starting point and end point are connected and the number of continuous boundary pixels exceeds a set length threshold, the region where the optical flow perturbation is concentrated is marked as having boundary integrity, and a boundary integrity attribute flag is output, wherein the boundary integrity attribute flag includes a yes or no flag.

[0041] In the optical flow perturbation concentrated area marked with the boundary integrity attribute flag as yes, a pixel filling operation is performed inside the area based on the closed boundary, the number of pixels inside the boundary is calculated, and the area ratio is calculated in combination with the total number of pixels in the image frame, and the output is used as the shape area ratio indicator;

[0042] Then, the boundary integrity attribute flag and the optical flow disturbance concentrated area are integrated as an outline area ratio indicator to construct a boundary contour morphological feature dataset of the optical flow disturbance concentrated area;

[0043] S42, based on the boundary contour morphological feature dataset of the optical flow disturbance concentrated area, counting the number of pixels in the area surrounded by the closed contour structure, and combining the total number of pixels in the image frame, calculating the area ratio occupied by the optical flow disturbance concentrated area in the entire image, and outputting it as the occlusion area ratio;

[0044] The data is combined with the occlusion behavior possibility score and the optical flow disturbance concentrated area boundary contour morphological feature dataset to construct an optical flow disturbance concentrated area feature fusion input set.

[0045] Preferably, the S5 includes S51;

[0046] S51, inputting the occlusion area feature fusion input set into the interference behavior discrimination model, executing multi-feature fusion judgment logic, outputting an interference behavior discrimination label, and executing a classification response action according to the judgment type corresponding to the interference behavior discrimination label, where the interference behavior discrimination label is used to clarify the category of image interference in the current video image frame sequence, including: natural light changes, occasional occlusion behavior, and malicious occlusion behavior;

[0047] The interference behavior discrimination model is constructed using a supervised training method. The construction logic includes weight aggregation of multi-dimensional features, feature space normalization, and discrimination threshold comparison mechanism. The interference behavior discrimination label output by the interference behavior discrimination model is a classification discrimination result, which is used to identify the type of interference behavior in the current video image frame sequence;

[0048] The interference behavior identification label includes the following types of judgment: natural light changes, occasional occlusion behavior, and malicious occlusion behavior, which correspond to different sources of image interference and preset judgment types respectively;

[0049] Execute classification response actions based on the judgment type corresponding to the interference behavior discrimination label;

[0050] If the interference behavior identification tag is a natural light change identification type, the record is kept but no alarm is triggered;

[0051] If the interference behavior identification label is a determination type of occasional occlusion behavior, record and mark the image frame as a focus frame and retain it for later reference;

[0052] If the interference behavior identification label is a determination type of malicious occlusion behavior, the security response module is automatically triggered to execute local image encryption, remote alarm push and video encryption and storage response actions.

[0053] A security box video monitoring method with intelligent discrimination function, comprising an image frame acquisition module, an image frame analysis and extraction module, an image optical flow disturbance determination module, an image contour extraction module and a security response module;

[0054] The image frame acquisition module continuously acquires on-site image data through the security box camera module to form a video image frame sequence, extracts the average brightness value to form an image frame average brightness sequence; at the same time, it records the acquisition time of the image frame to form an image frame time sequence, calculates the image frame brightness gradient change rate, and outputs the image frame brightness gradient change rate set;

[0055] The image frame analysis and extraction module performs dynamic trend analysis on the image frame brightness gradient change rate set, identifies the fluctuation state in consecutive frames, and constructs an abnormal occlusion image frame data set;

[0056] The image optical flow disturbance determination module processes the abnormal occlusion image frame data set frame by frame, generates an optical flow displacement vector set, and outputs an occlusion behavior possibility score based on the spatial distribution trend of the image frame brightness gradient change rate set;

[0057] The image contour extraction module performs edge extraction processing on the optical flow disturbance concentrated area, outputs the occlusion area ratio, and then combines the occlusion behavior possibility score, the occlusion area boundary contour morphological feature dataset and the occlusion area ratio into an occlusion area feature fusion input set;

[0058] The security response module performs multi-feature fusion judgment logic on the occlusion area feature fusion, outputs an interference behavior discrimination label, and performs a classification response action according to the judgment type corresponding to the interference behavior discrimination label.

[0059] The present invention provides a security box video monitoring system and method with intelligent identification function, which has the following beneficial effects:

[0060] (1) By continuously acquiring images and constructing the image frame average brightness sequence and image frame time sequence, the dynamic trend of the image frame brightness gradient change rate set can be extracted, effectively capturing the characteristics of illumination mutations; further, using the abnormal occlusion image frame dataset as the input basis, combined with the optical flow displacement vector set and the spatial variation trend of the image frame brightness gradient change rate set, the occlusion behavior possibility score is output to achieve probabilistic judgment of occlusion intention and avoid misjudgment of occasional non-human interference. At the same time, edge structure analysis is performed in the optical flow perturbation concentrated area to generate the boundary contour morphological feature dataset of the optical flow perturbation concentrated area, and the occlusion area ratio is calculated. It is then fused with the occlusion behavior possibility score to form the optical flow perturbation concentrated area feature fusion input set, providing complete structure and intensity information support. Finally, intelligent classification is performed through the interference behavior discrimination label.

[0061] (2) By using regional angle consistency calculation and combining the three characteristic conditions of optical flow amplitude, direction consistency and pixel ratio of the disturbed area, the concentrated area of ​​optical flow disturbance is accurately screened out, and the behavioral driving characteristics of the occlusion area are extracted. On this basis, by spatially mapping the concentrated area of ​​optical flow disturbance with the set of image frame brightness gradient change rates, a pixel-level alignment analysis of the spatial overlap relationship between the occlusion behavior driving characteristics and brightness mutation is established, thereby constructing an occlusion behavior scoring function and outputting an occlusion behavior probability score to quantify the occurrence intensity of the occlusion behavior in the image frame. This not only improves the ability to discriminate the intention of malicious occlusion and other behaviors, but also provides a scoring basis with behavioral explanation for subsequent decision models, significantly enhancing the judgment accuracy of behavior type differentiation while maintaining structural integrity.

[0062] (3) By extracting the edge contour and distinguishing the structure of the concentrated area of ​​optical flow disturbance, the key transition from image motion information to visual boundary geometric structure is achieved. The closed state of the boundary is identified by the image gradient edge extraction algorithm, and the boundary integrity attribute mark and the proportion of the outer area are output in combination with the regional pixel statistics. The boundary contour morphological feature dataset of the concentrated area of ​​optical flow disturbance is further constructed. The structural feature is then normalized by the total pixel range of the image frame to generate the occlusion area ratio. The classification response action is automatically executed by the typing result of the interference behavior discrimination label, so that the system not only has the ability to make accurate judgments after image interference recognition, but also has the advantages of linkage control of instant response, layered processing and local encryption, which enhances the practical ability of security box video surveillance in resisting malicious occlusion, suppressing false alarms and preserving event evidence. BRIEF DESCRIPTION OF THE DRAWINGS

[0063] Figure 1 This is a schematic diagram of the steps of a security box video monitoring method with intelligent identification function according to the present invention;

[0064] Figure 2 This is a schematic diagram of the flow chart of a security box video monitoring system with intelligent identification function according to the present invention. DETAILED DESCRIPTION

[0065] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0066] Example 1

[0067] The present invention provides a security box video monitoring method with intelligent identification function, please refer to Figure 1 , including the following steps:

[0068] S1. Continuously capture on-site image data through the security box camera module to form a video image frame sequence. After extracting the average brightness value, form an image frame average brightness sequence. Simultaneously, record the acquisition time of the image frame to form an image frame time sequence. Calculate the image frame brightness gradient change rate and output the image frame brightness gradient change rate set.

[0069] S2. Performing dynamic trend analysis on the image frame brightness gradient change rate set to identify fluctuation states in consecutive frames and constructing an abnormal occlusion image frame data set;

[0070] S3, processing the abnormal occlusion image frame data set frame by frame to generate an optical flow displacement vector set, and outputting an occlusion behavior possibility score based on the spatial distribution trend of the image frame brightness gradient change rate set;

[0071] S4, performing edge extraction processing on the optical flow disturbance concentrated area, outputting the occlusion area ratio, and then merging the occlusion behavior possibility score, the occlusion area boundary contour morphological feature dataset, and the occlusion area ratio into an occlusion area feature fusion input set;

[0072] S5. Execute multi-feature fusion judgment logic on the occlusion area feature fusion, output an interference behavior discrimination label, and perform a classification response action according to the judgment type corresponding to the interference behavior discrimination label.

[0073] In this embodiment, by continuously capturing images and constructing a sequence of average brightness and time frames, dynamic trends can be extracted from the image frame brightness gradient change rate set, effectively capturing the characteristics of sudden illumination changes. Furthermore, using the abnormal occlusion image frame dataset as input, combined with the optical flow displacement vector set and the spatial variation trend of the image frame brightness gradient change rate set, an occlusion behavior probability score is output, enabling probabilistic determination of occlusion intent and avoiding misjudgment of occasional non-human interference. Simultaneously, edge structure analysis is performed within the optical flow perturbation concentrated region to generate a dataset of boundary contour morphological features. The occlusion area ratio is calculated and then fused with the occlusion behavior probability score to form a fused input set of features for the optical flow perturbation concentrated region, providing comprehensive structural and intensity information. Finally, intelligent classification is performed using interference behavior discrimination labels, accurately distinguishing between "natural light changes," "incidental occlusion," and "malicious occlusion." This effectively improves the system's image interference recognition accuracy and response decision-making capabilities without requiring hardware modifications, effectively improving the system's low-cost image interference recognition and response decision-making capabilities. This approach is particularly suitable for addressing practical issues such as the difficulty in distinguishing between light and shadow occlusion, frequent false alarms, and difficulty identifying vandalism, as discussed in the previous background introduction.

[0074] Example 2

[0075] This embodiment is explained in Example 1, please refer to Figure 1 , specifically: the S1 includes S11 and S12;

[0076] S11, using the camera module integrated in the security box, continuously collects real-time image data of the target monitoring area at preset time intervals, and stores it in chronological order to form a video image frame sequence;

[0077] For each frame of the video image frame sequence, performing an image brightness information extraction operation to obtain an average pixel brightness value of the image frame; combining the average brightness values ​​of all image frames in frame order to form an image frame average brightness sequence;

[0078] At the same time, during the image acquisition process, the actual acquisition time information corresponding to each frame of image is recorded to generate an image frame time series;

[0079] The image brightness information is extracted by using the image frame grayscale channel average value method.

[0080] S12. Based on the index relationship between the image frame average brightness sequence and the image frame time sequence, perform differential calculation on the brightness change values ​​between adjacent image frames, and combine the acquisition time interval to obtain the rate of image frame brightness change per unit time, which represents the image frame brightness gradient change rate;

[0081] The brightness gradient change rates corresponding to all image frames are stored in chronological order to form an image frame brightness gradient change rate set, which serves as an important input parameter set for illumination change pattern recognition and abnormal occlusion image frame data set screening in subsequent steps;

[0082] Among them, the index relationship specifically indicates that a one-to-one correspondence is established between the image frame average brightness sequence and the image frame time sequence through the image frame index number to ensure the precise pairing of brightness changes and acquisition time, thereby providing basic support for the time series calculation of the image frame brightness gradient change rate.

[0083] In this embodiment, an index relationship between the two is established using the image frame index number, achieving precise temporal calculation of the image frame brightness gradient change rate, and then outputting a set of image frame brightness gradient change rates, significantly enhancing the system's ability to model dynamic responses to illumination changes. Unlike traditional methods based on single-frame brightness perception or threshold setting, this method introduces the image frame grayscale channel average method to ensure the robustness of brightness feature extraction and the accuracy of inter-frame difference determination. It also completes the analysis of the brightness change rate per unit time by combining the acquisition time interval, thereby improving the sensitive recognition ability of natural light gradual changes, sudden changes, and periodic illumination interference. The image frame brightness gradient change rate set not only serves as the basic input for subsequent illumination pattern classification and screening of abnormal occlusion image frame data sets, but also establishes a dynamic feature channel between image content changes and environmental interference by maintaining temporal correlation, providing key support for achieving more accurate natural interference removal without relying on external illumination sensors.

[0084] Example 3

[0085] This embodiment is explained in Example 1, please refer to Figure 1 Specifically: S2 includes S21 and S22;

[0086] S21. Perform a sequential trend analysis on the image frame brightness gradient change rate set. Use a sliding time window mechanism to jointly calculate the first-order difference and local statistical variance of the image frame brightness gradient change rate within each local window. When consecutive image frames simultaneously meet the following two conditions within the sliding window:

[0087] Condition 1: The first-order difference exceeds the preset mutation slope threshold;

[0088] Condition 2: The local statistical variance exceeds the preset violent fluctuation threshold;

[0089] The continuous image frames are marked as suspected occlusion interference image frames and included in the set of suspected occlusion interference image frames as input for subsequent image verification;

[0090] Among them, the first-order difference of the image frame brightness gradient change rate in each local window reflects the change slope; the local statistical variance of the image frame brightness gradient change rate in each local window reflects the severity of the fluctuation.

[0091] S22. For each frame in the set of suspected occlusion interference image frames, extract the image content of each frame in the video image frame sequence, and obtain an original grayscale image of the image content as an analysis target; based on the original grayscale image, calculate the grayscale change amplitude of each pixel according to the brightness difference between the frame and its adjacent frames in the image frame average brightness sequence, and form a pixel-level grayscale difference matrix;

[0092] Mapping the pixel-level grayscale difference matrix to the original image space to generate an image brightness change distribution map, which represents the brightness fluctuation strength of the image frame in the spatial dimension. Dividing the entire image into two structural blocks, an edge area and a central area, according to a preset spatial grid, counting the number of pixels in each area whose brightness change intensity is greater than a set minimum change threshold, calculating the proportion of the number of pixels relative to the area of ​​the area, and using them as the brightness change density of the edge area and the brightness change density of the central area, respectively. Then, calculating the brightness change density ratio of the brightness change density of the edge area to the brightness change density of the central area;

[0093] When the brightness change density ratio exceeds the preset edge concentration change threshold, it is determined that the brightness mutation occurs concentrated at the edge of the image, which meets the occlusion behavior characteristics;

[0094] The image frames that meet the occlusion behavior characteristic conditions are marked as abnormal occlusion image frames, and a data set of abnormal occlusion image frames is constructed.

[0095] In this embodiment, a sliding time window mechanism is introduced to perform sequential trend analysis on the image frame brightness gradient change rate set, effectively capturing the brightness mutation characteristics of the image frames in the temporal dimension. Combined with the dual indicators of first-order differences within a local window and local statistical variance, this method significantly improves the dynamic accuracy of the initial screening of occlusion behavior. Compared to traditional methods that use a single threshold or global fitting model to process image changes, this method constructs a set of suspected occlusion interference image frames based on the short-term trend fluctuations of the image frame brightness gradient change rate set, avoiding the misjudgment of natural light fluctuations. Furthermore, by calculating the pixel-level grayscale difference matrix formed by the original grayscale image and mapping it to the original image space to form an image brightness change distribution map, combined with an image spatial structure partitioning strategy, quantitatively assesses the intensity of brightness disturbances in image edge regions. Leveraging the brightness change density ratio judgment rule, rapid identification of concentrated mutation regions at the image edge is achieved, effectively extracting spatial distribution features highly correlated with occlusion behavior, and ultimately constructing a dataset of abnormal occlusion image frames, establishing a precise screening prerequisite for subsequent multimodal image discrimination.

[0096] Example 4

[0097] This embodiment is explained in Example 1, please refer to Figure 1 , specifically: said S3 includes S31 and S32;

[0098] S31. For each frame of the abnormal occlusion image frame data set, combined with the previous and next frames in the video image frame sequence, perform pixel-by-pixel dense optical flow calculation processing, wherein the dense optical flow calculation processing adopts the Farneback optical flow algorithm to calculate the displacement vector of each pixel between adjacent frames, representing the distribution of optical flow vectors corresponding to the image frames in the time series;

[0099] The displacement vectors obtained from each frame of the image are sorted by frame number, and a set of optical flow displacement vectors is generated. Based on the set of optical flow displacement vectors, the optical flow amplitude value and direction vector of each pixel are extracted to construct an optical flow intensity map and an optical flow direction map of the entire image. A threshold segmentation method is used for the optical flow intensity map to remove low-intensity optical flow areas and retain high-amplitude areas. Then, a regional angle consistency calculation is performed on the direction vectors of the pixels in the retained area. The regional angle consistency calculation includes using a density clustering algorithm to calculate the regional angle consistency of a group of pixels with a consistent motion direction, matching the local consistency features in the pixel group with the feature conditions, and marking them as candidate disturbance areas. The candidate disturbance areas are then judged according to the following three feature conditions, and the areas that meet the conditions are marked as concentrated optical flow disturbance areas.

[0100] Characteristic condition 1: The optical flow intensity in the local area is greater than 1.5 times the average intensity of the entire image;

[0101] Feature condition 2: The variance of the optical flow direction angle within the region is less than the preset direction consistency threshold;

[0102] Feature condition three: the number of pixels in the region exceeds the minimum disturbance region ratio threshold of the total pixel area of ​​the image;

[0103] When the candidate disturbance area meets the characteristic conditions 1, 2 and 3, it is determined to be the optical flow disturbance concentration area, and the spatial boundary and center coordinates are recorded.

[0104] S32. Based on the optical flow disturbance concentrated area and the image frame brightness gradient change rate set, performing a spatial position mapping operation, wherein the spatial position mapping operation is based on the image frame spatial coordinate pair, performing pixel-level coordinate coincidence analysis on the boundary range of the optical flow disturbance concentrated area and the brightness mutation area in the image frame brightness gradient change rate set to determine a spatial intersection;

[0105] Based on the spatial intersection and overlap between the optical flow disturbance concentrated area and the brightness mutation area, and the area ratio of the optical flow disturbance area in the image frame, an occlusion behavior scoring function is constructed to output the occlusion behavior probability score corresponding to the image frame, which is used to quantify the possibility of occlusion behavior.

[0106] The occlusion behavior possibility score is obtained by constructing the following occlusion behavior scoring function:

[0107] ;

[0108] Where Pf represents the occlusion behavior probability score, Rci represents the spatial intersection overlap rate, Rsp represents the area proportion, α and β represent the weight coefficients of the spatial intersection overlap rate Rci and the area proportion Rsp, respectively, and α + β = 1. The specific value is set by the user.

[0109] Among them, the spatial intersection coincidence rate Rci is obtained by calculating the area ratio of the overlapping area of ​​the image frame after calculating the pixel-level spatial position mapping of the image frame brightness gradient change rate set and the optical flow perturbation concentrated area;

[0110] In this embodiment, by introducing a pixel-by-pixel dense optical flow calculation mechanism based on the Farneback algorithm, the pixel-level motion change trajectory of each image in the abnormal occlusion image frame data set between adjacent frames is effectively obtained, a set of optical flow displacement vectors is constructed, and further an optical flow intensity map and an optical flow direction map are formed, providing a fine-grained motion feature basis for modeling the dynamic disturbance of the occlusion behavior area. Compared with the traditional method based only on image content change or static edge recognition, this method uses regional angle consistency calculation, combines the three feature conditions of optical flow amplitude, direction consistency and disturbance region pixel proportion, accurately selects the optical flow disturbance concentrated area, and realizes the extraction of the behavior driving characteristics of the occlusion area. On this basis, by mapping the optical flow disturbance concentrated area and the image frame brightness gradient change rate set in space, the pixel-level alignment analysis of the spatial overlap relationship between the occlusion behavior driving characteristics and the brightness mutation is established, so as to construct the occlusion behavior scoring function and output the occlusion behavior possibility score, which is used to quantify the occurrence intensity of the occlusion behavior in the image frame. This method not only improves the discrimination ability of the malicious occlusion behavior intention recognition, but also provides a scoring basis with behavior explanation for the subsequent decision model, while maintaining the structural integrity, significantly enhancing the judgment accuracy of behavior type differentiation.

[0111] Embodiment 5

[0112] This embodiment is an explanation and description in Embodiment 1, please refer to Figure 1 , in particular: the S4 comprises S41 and S42;

[0113] S41, for the optical flow disturbance concentrated area, corresponding image frames in the video image frame sequence are acquired, and a gray-scale image of the image frame is extracted as an input image for edge detection, and an image gradient edge extraction algorithm is used for processing based on the gray-scale image, the image gradient edge extraction algorithm comprises a Canny algorithm, and an edge binary image corresponding to the optical flow disturbance concentrated area is generated, which is used for identifying boundary structure;

[0114] In the edge binary image, a contour tracking operation is performed based on pixel connectivity, a closed boundary path is identified, and when there is a contour structure with a starting point and an ending point connected and a continuous boundary pixel number exceeding a set length threshold, it is marked that the optical flow disturbance concentrated area has boundary integrity, and a boundary integrity attribute flag is output, the boundary integrity attribute flag comprises yes and no;

[0115] In the optical flow disturbance concentrated area marked with the boundary integrity attribute flag as yes, a region interior pixel filling operation is performed based on the closed boundary, the number of boundary interior pixels is calculated, and the area proportion is calculated in combination with the total number of image frame pixels, and an outline area proportion index is output;

[0116] Then, the boundary integrity attribute flag and the optical flow disturbance concentrated area are integrated as an outline area ratio indicator to construct a boundary contour morphological feature dataset of the optical flow disturbance concentrated area;

[0117] S42, based on the boundary contour morphological feature dataset of the optical flow disturbance concentrated area, counting the number of pixels in the area surrounded by the closed contour structure, and combining the total number of pixels in the image frame, calculating the area ratio occupied by the optical flow disturbance concentrated area in the entire image, and outputting it as the occlusion area ratio;

[0118] The data is combined with the occlusion behavior possibility score and the optical flow disturbance concentrated area boundary contour morphological feature dataset to construct an optical flow disturbance concentrated area feature fusion input set.

[0119] The S5 includes S51;

[0120] S51, inputting the occlusion area feature fusion input set into the interference behavior discrimination model, executing multi-feature fusion judgment logic, outputting an interference behavior discrimination label, and executing a classification response action according to the judgment type corresponding to the interference behavior discrimination label, where the interference behavior discrimination label is used to clarify the category of image interference in the current video image frame sequence, including: natural light changes, occasional occlusion behavior, and malicious occlusion behavior;

[0121] The interference behavior discrimination model is constructed using a supervised training method. The construction logic includes weight aggregation of multi-dimensional features, feature space normalization, and discrimination threshold comparison mechanism. The interference behavior discrimination label output by the interference behavior discrimination model is a classification discrimination result, which is used to identify the type of interference behavior in the current video image frame sequence;

[0122] The interference behavior identification label includes the following types of judgment: natural light changes, occasional occlusion behavior, and malicious occlusion behavior, which correspond to different sources of image interference and preset judgment types respectively;

[0123] Execute classification response actions based on the judgment type corresponding to the interference behavior discrimination label;

[0124] If the interference behavior identification tag is a natural light change identification type, the record is kept but no alarm is triggered;

[0125] If the interference behavior identification label is a determination type of occasional occlusion behavior, record and mark the image frame as a focus frame and retain it for later reference;

[0126] If the interference behavior identification label is a determination type of malicious occlusion behavior, the security response module is automatically triggered to execute local image encryption, remote alarm push and video encryption and storage response actions.

[0127] In this embodiment, by performing edge contour extraction and structural discrimination on the concentrated area of ​​optical flow disturbance, a key transition from image motion information to visual boundary geometric structure is achieved. The closed state of the boundary is identified by relying on the image gradient edge extraction algorithm, and the boundary integrity attribute flag and the proportion of the outer area are output in combination with the regional pixel statistics to further construct a boundary contour morphological feature dataset of the concentrated area of ​​optical flow disturbance. The structural feature is then normalized by the total pixel range of the image frame to generate the occlusion area ratio. The above three structural parameters together constitute the feature fusion input set of the concentrated area of ​​optical flow disturbance, which serves as a unified feature entry for the subsequent classification and discrimination of image interference behavior. The feature fusion set is input into the interference behavior discrimination model, and the feature weight aggregation and normalization mechanism constructed based on supervised training is used to output a clearly defined interference behavior discrimination label, which effectively classifies natural light changes, occasional occlusion behavior and malicious occlusion behavior into intelligent categories. By driving the automatic execution of classification response actions through the classification results of interference behavior discrimination labels, the system not only has the ability to make accurate judgments after image interference recognition, but also has the advantages of instant response, layered processing and local encryption linkage control, enhancing the practical capabilities of security box video surveillance in resisting malicious occlusion, suppressing false alarms and preserving event evidence.

[0128] Example 6

[0129] A security box video surveillance system with intelligent identification function, please refer to Figure 2 ,Specifically: including image frame acquisition module, image frame analysis and extraction module, image optical flow disturbance judgment module, image contour extraction module and security response module;

[0130] The image frame acquisition module continuously acquires on-site image data through the security box camera module to form a video image frame sequence, extracts the average brightness value to form an image frame average brightness sequence; at the same time, it records the acquisition time of the image frame to form an image frame time sequence, calculates the image frame brightness gradient change rate, and outputs the image frame brightness gradient change rate set;

[0131] The image frame analysis and extraction module performs dynamic trend analysis on the image frame brightness gradient change rate set, identifies the fluctuation state in consecutive frames, and constructs an abnormal occlusion image frame data set;

[0132] The image optical flow disturbance determination module processes the abnormal occlusion image frame data set frame by frame, generates an optical flow displacement vector set, and outputs an occlusion behavior possibility score based on the spatial distribution trend of the image frame brightness gradient change rate set;

[0133] The image contour extraction module performs edge extraction processing on the optical flow disturbance concentrated area, outputs the occlusion area ratio, and then combines the occlusion behavior possibility score, the occlusion area boundary contour morphological feature dataset and the occlusion area ratio into an occlusion area feature fusion input set;

[0134] The security response module performs multi-feature fusion judgment logic on the occlusion area feature fusion, outputs an interference behavior discrimination label, and performs a classification response action according to the judgment type corresponding to the interference behavior discrimination label.

[0135] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

Claims

1. A security box video monitoring method with intelligent identification function, characterized by: The following steps are involved: S1. Continuously capture on-site image data using a security box camera module to form a video image frame sequence. Perform an image brightness information extraction operation on each frame in the video image frame sequence to obtain the average pixel brightness value of the image frame to form an image frame average brightness sequence. Simultaneously, record the acquisition time of the image frame to form an image frame time sequence. Calculate the image frame brightness gradient change rate, and output a set of image frame brightness gradient change rates. S2. Performing dynamic trend analysis on the image frame brightness gradient change rate set to identify fluctuation states in consecutive frames and constructing an abnormal occlusion image frame data set; S3, processing the abnormal occlusion image frame data set frame by frame to generate an optical flow displacement vector set, and outputting an occlusion behavior possibility score based on the spatial distribution trend of the image frame brightness gradient change rate set; S4, performing edge extraction processing on the optical flow disturbance concentrated area, outputting the occlusion area ratio, and then merging the occlusion behavior possibility score, the occlusion area boundary contour morphological feature dataset, and the occlusion area ratio into an occlusion area feature fusion input set; S5. Execute multi-feature fusion judgment logic on the occlusion area feature fusion, output an interference behavior discrimination label, and perform a classification response action according to the judgment type corresponding to the interference behavior discrimination label; Said S2 includes S21 and S22; S21. Perform a sequential trend analysis on the image frame brightness gradient change rate set. Use a sliding time window mechanism to jointly calculate the first-order difference and local statistical variance of the image frame brightness gradient change rate within each local window. When consecutive image frames simultaneously meet the following two conditions within the sliding window: Condition 1: The first-order difference exceeds the preset mutation slope threshold; Condition 2: The local statistical variance exceeds the preset violent fluctuation threshold; The continuous image frames are marked as suspected occlusion interference image frames and included in the suspected occlusion interference image frame set; Among them, the first-order difference of the image frame brightness gradient change rate in each local window reflects the change slope; the local statistical variance of the image frame brightness gradient change rate in each local window reflects the intensity of the fluctuation; S22. For each frame in the set of suspected occlusion interference image frames, extract the image content of each frame in the video image frame sequence, and obtain an original grayscale image of the image content as an analysis target; based on the original grayscale image, calculate the grayscale change amplitude of each pixel according to the brightness difference between the frame and its adjacent frames in the image frame average brightness sequence, and form a pixel-level grayscale difference matrix; Mapping the pixel-level grayscale difference matrix to the original image space to generate an image brightness change distribution map, dividing the entire image into two structural blocks, an edge area and a central area, according to a preset spatial grid, counting the number of pixels in each area whose brightness change intensity is greater than a set minimum change threshold, calculating the proportion of the number of pixels relative to the area of ​​the area, and using them as the brightness change density of the edge area and the brightness change density of the central area, respectively, and then calculating the brightness change density ratio of the brightness change density of the edge area to the brightness change density of the central area; When the brightness change density ratio exceeds the preset edge concentration change threshold, it is determined that the brightness mutation occurs concentrated at the edge of the image, which meets the occlusion behavior characteristics; Mark the image frames that meet the occlusion behavior characteristic conditions as abnormal occlusion image frames, and summarize and construct a data set of abnormal occlusion image frames; Said S3 includes S31 and S32; S31. For each frame of the abnormal occlusion image frame data set, combined with the previous and next frames in the video image frame sequence, perform pixel-by-pixel dense optical flow calculation processing, wherein the dense optical flow calculation processing adopts the Farneback optical flow algorithm to calculate the displacement vector of each pixel between adjacent frames, representing the distribution of optical flow vectors corresponding to the image frames in the time series; The displacement vectors obtained from each frame of the image are sorted by frame number, and a set of optical flow displacement vectors is generated. Based on the set of optical flow displacement vectors, the optical flow amplitude value and direction vector of each pixel are extracted to construct an optical flow intensity map and an optical flow direction map of the entire image. A threshold segmentation method is used for the optical flow intensity map to remove low-intensity optical flow areas and retain high-amplitude areas. Then, a regional angle consistency calculation is performed on the direction vectors of the pixels in the retained area. The regional angle consistency calculation includes using a density clustering algorithm to calculate the regional angle consistency of a group of pixels with a consistent motion direction, matching the local consistency features in the pixel group with the feature conditions, and marking them as candidate disturbance areas. The candidate disturbance areas are then judged according to the following three feature conditions, and the areas that meet the conditions are marked as concentrated optical flow disturbance areas. Characteristic condition 1: The optical flow intensity in the local area is greater than 1.5 times the average intensity of the entire image; Feature condition 2: The variance of the optical flow direction angle within the region is less than the preset direction consistency threshold; Feature condition three: the number of pixels in the region exceeds the minimum disturbance region ratio threshold of the total pixel area of ​​the image; When the candidate disturbance area meets the characteristic conditions 1, 2 and 3, it is determined to be the optical flow disturbance concentration area, and the spatial boundary and center coordinates are recorded.

2. The security box video monitoring method with intelligent identification function according to claim 1, characterized in that: Said S1 includes S11 and S12; S11, using the camera module integrated in the security box, continuously collects real-time image data of the target monitoring area at preset time intervals, and stores it in chronological order to form a video image frame sequence; For each frame of the video image frame sequence, performing an image brightness information extraction operation to obtain an average pixel brightness value of the image frame; combining the average brightness values ​​of all image frames in frame order to form an image frame average brightness sequence; At the same time, during the image acquisition process, the actual acquisition time information corresponding to each frame of image is recorded to generate an image frame time series; The image brightness information is extracted by using the image frame grayscale channel average value method.

3. The security box video monitoring method with intelligent identification function according to claim 2, characterized in that: S12. Based on the index relationship between the image frame average brightness sequence and the image frame time sequence, perform differential calculation on the brightness change values ​​between adjacent image frames, and combine the acquisition time interval to obtain the rate of image frame brightness change per unit time, which represents the image frame brightness gradient change rate; The brightness gradient change rates corresponding to all image frames are stored in time sequence to form a set of image frame brightness gradient change rates; The index relationship specifically indicates that a one-to-one correspondence is established between the image frame average brightness sequence and the image frame time sequence through the image frame index numbers.

4. The security box video monitoring method with intelligent identification function according to claim 1, characterized in that: S32. Based on the optical flow disturbance concentrated area and the image frame brightness gradient change rate set, performing a spatial position mapping operation, wherein the spatial position mapping operation is based on the image frame spatial coordinate pair, performing pixel-level coordinate coincidence analysis on the boundary range of the optical flow disturbance concentrated area and the brightness mutation area in the image frame brightness gradient change rate set to determine a spatial intersection; Based on the degree of spatial intersection and overlap between the concentrated area of ​​optical flow disturbance and the area of ​​brightness mutation, combined with the area ratio of the optical flow disturbance area in the image frame, an occlusion behavior scoring function is constructed to output the occlusion behavior possibility score corresponding to the image frame, which is used to quantify the possibility of occlusion behavior.

5. The security box video monitoring method with intelligent identification function according to claim 4, characterized in that: Said S4 includes S41 and S42; S41. For the optical flow disturbance concentrated area, corresponding image frames in the video image frame sequence are obtained, and a grayscale image of the image frame is extracted as an input image for edge detection. An image gradient edge extraction algorithm is used to process the grayscale image to generate an edge binary image corresponding to the optical flow disturbance concentrated area for identifying boundary structures. In the edge binary image, a contour tracing operation is performed based on pixel connectivity to identify a closed boundary path. When a contour structure exists whose starting point and end point are connected and the number of continuous boundary pixels exceeds a set length threshold, the region where the optical flow perturbation is concentrated is marked as having boundary integrity, and a boundary integrity attribute flag is output, wherein the boundary integrity attribute flag includes a yes or no flag. In the optical flow perturbation concentrated area marked with the boundary integrity attribute flag as yes, a pixel filling operation is performed inside the area based on the closed boundary, the number of pixels inside the boundary is calculated, and the area ratio is calculated in combination with the total number of pixels in the image frame, and the output is used as the shape area ratio indicator; Then, the boundary integrity attribute flag and the optical flow disturbance concentrated area are integrated as an outline area ratio indicator to construct a boundary contour morphological feature dataset of the optical flow disturbance concentrated area; S42, based on the boundary contour morphological feature dataset of the optical flow disturbance concentrated area, counting the number of pixels in the area surrounded by the closed contour structure, and combining the total number of pixels in the image frame, calculating the area ratio occupied by the optical flow disturbance concentrated area in the entire image, and outputting it as the occlusion area ratio; The data is combined with the occlusion behavior possibility score and the optical flow disturbance concentrated area boundary contour morphological feature dataset to construct an optical flow disturbance concentrated area feature fusion input set.

6. The security box video monitoring method with intelligent identification function according to claim 5, characterized in that: Said S5 includes S51; S51, inputting the occlusion area feature fusion input set into the interference behavior discrimination model, executing multi-feature fusion judgment logic, outputting an interference behavior discrimination label, and executing a classification response action according to the judgment type corresponding to the interference behavior discrimination label, where the interference behavior discrimination label is used to clarify the category of image interference in the current video image frame sequence, including: natural light changes, occasional occlusion behavior, and malicious occlusion behavior; The interference behavior discrimination model is constructed using a supervised training method. The construction logic includes weight aggregation of multi-dimensional features, feature space normalization, and discrimination threshold comparison mechanism. The interference behavior discrimination label output by the interference behavior discrimination model is a classification discrimination result, which is used to identify the type of interference behavior in the current video image frame sequence; The interference behavior identification label includes the following types of judgment: natural light changes, occasional occlusion behavior, and malicious occlusion behavior, which correspond to different sources of image interference and preset judgment types respectively; Execute classification response actions based on the judgment type corresponding to the interference behavior discrimination label; If the interference behavior identification tag is a natural light change identification type, the record is kept but no alarm is triggered; If the interference behavior identification label is a determination type of occasional occlusion behavior, record and mark the image frame as a focus frame and retain it for later reference; If the interference behavior identification label is a determination type of malicious occlusion behavior, the security response module is automatically triggered to execute local image encryption, remote alarm push and video encryption and storage response actions.

7. A security box video monitoring system with intelligent identification function, comprising the security box video monitoring method with intelligent identification function according to any one of claims 1 to 6, characterized in that: It includes image frame acquisition module, image frame analysis and extraction module, image optical flow disturbance judgment module, image contour extraction module and security response module; The image frame acquisition module continuously acquires on-site image data through the security box camera module to form a video image frame sequence, extracts the average brightness value to form an image frame average brightness sequence; at the same time, it records the acquisition time of the image frame to form an image frame time sequence, calculates the image frame brightness gradient change rate, and outputs the image frame brightness gradient change rate set; The image frame analysis and extraction module performs dynamic trend analysis on the image frame brightness gradient change rate set, identifies the fluctuation state in consecutive frames, and constructs an abnormal occlusion image frame data set; The image optical flow disturbance determination module processes the abnormal occlusion image frame data set frame by frame, generates an optical flow displacement vector set, and outputs an occlusion behavior possibility score based on the spatial distribution trend of the image frame brightness gradient change rate set; The image contour extraction module performs edge extraction processing on the optical flow disturbance concentrated area, outputs the occlusion area ratio, and then combines the occlusion behavior possibility score, the occlusion area boundary contour morphological feature dataset and the occlusion area ratio into an occlusion area feature fusion input set; The security response module performs multi-feature fusion judgment logic on the occlusion area feature fusion, outputs an interference behavior discrimination label, and performs a classification response action according to the judgment type corresponding to the interference behavior discrimination label.

Images