A police unmanned aerial vehicle monitoring method, system, device and storage medium based on image segmentation

Abstract

The application discloses a police unmanned aerial vehicle monitoring method and system based on image segmentation, and a storage medium, relates to the technical field of unmanned aerial vehicle monitoring and image processing, and comprises the following steps: acquiring a monitoring video sequence based on an unmanned aerial vehicle, performing texture feature analysis on a background region to construct a background change model and generate a dynamic background suppression region, and segmenting a target region; performing region matching and overlap degree calculation on the target segmentation region in continuous video frames, performing time sequence reconstruction through multi-frame target region information fusion, and generating a target segmentation result; analyzing a target activity mode, identifying an abnormal activity behavior, and generating early warning information. The method disclosed by the application generates a dynamic background suppression region by constructing a background change model, reduces dynamic background texture interference, realizes target region segmentation, and improves target recognition stability; and analyzes a target activity mode based on a target activity feature sequence, realizes abnormal activity behavior identification and early warning.

Description

Technical Field

[0001] This invention relates to the field of unmanned aerial vehicle (UAV) monitoring and image processing technology, specifically to a method, system, device, and storage medium for monitoring police UAVs based on image segmentation. Background Technology

[0002] With the continuous development of drone technology, computer vision technology, and intelligent monitoring technology, drones are being used more and more widely in patrol monitoring, regional supervision, on-site law enforcement, and security. Compared with traditional fixed monitoring equipment, drones have the advantages of high mobility, wide monitoring range, and flexible deployment, enabling real-time patrols of large monitoring areas and acquiring video data by carrying high-definition cameras. In recent years, target detection and image segmentation methods based on deep learning and image processing technologies have been gradually applied to drone monitoring systems. By analyzing the video images collected by drones, automatic identification and monitoring of vehicles, personnel, or other targets can be achieved, thereby improving the intelligence level of drone monitoring.

[0003] However, in real-world monitoring scenarios, due to factors such as changes in lighting, background disturbances, and environmental reflections, the background region typically exhibits complex and constantly changing image features, which can easily interfere with image segmentation results, leading to unstable target region identification. Monitoring videos are usually continuous video sequences; if target segmentation is performed only on single frames, discontinuous segmentation results or unstable target region changes can easily occur, making it difficult to accurately reflect the target's true activity status within the monitored area. Furthermore, existing technologies for analyzing the behavior of monitored targets are mostly based on simple rule judgments, lacking comprehensive analysis of information such as target region changes and movement trajectories, making it difficult to effectively identify target activity patterns and thus affecting the accuracy of abnormal behavior identification. Therefore, it is necessary to propose a police drone monitoring method based on image segmentation. This method models background changes and generates dynamic background suppression regions to achieve stable segmentation of target regions in video frames. It then combines multi-frame target region information for temporal reconstruction and activity feature analysis to achieve the identification and early warning of abnormal activity behaviors. Summary of the Invention

[0004] In view of the above-mentioned problems, the present invention is proposed.

[0005] Therefore, the technical problem solved by this invention is that existing UAV video surveillance and target recognition methods suffer from problems such as large dynamic background interference, poor stability of target region segmentation in complex scenes, insufficient consistency of target recognition results in continuous video frames, and difficulty in effectively analyzing the activity patterns of monitored targets and identifying abnormal behavior based on target motion trajectory and regional change information. Therefore, this invention proposes a UAV surveillance method based on image segmentation to achieve stable segmentation of target regions in surveillance videos, temporal reconstruction of target information in continuous frames, and identification and early warning of abnormal activity behavior.

[0006] To address the aforementioned technical problems, this invention provides the following technical solution: a police drone monitoring method based on image segmentation, comprising: analyzing the texture features of the background region in the video frames based on the monitoring video sequence acquired by the drone, constructing a background change model, and generating a dynamic background suppression region; segmenting the video frames into target regions based on the dynamic background suppression region to obtain the target segmentation region in each video frame; performing region matching and overlap calculation on the target segmentation regions in consecutive video frames, and temporally reconstructing the target regions based on the fusion of multi-frame target region information to generate a target segmentation result; constructing a target activity feature sequence based on the target segmentation result, analyzing the target's activity pattern within the monitoring area, identifying abnormal activity behaviors, and generating early warning information.

[0007] As a preferred embodiment of the police drone monitoring method based on image segmentation described in this invention, the step of performing texture feature analysis on the background region in the video frame includes extracting texture periodic features from the background region in consecutive video frames of the monitoring video sequence, calculating the background texture change amplitude between adjacent video frames, and establishing a background change model based on the background texture change amplitude.

[0008] As a preferred embodiment of the police drone monitoring method based on image segmentation described in this invention, the generation of dynamic background suppression regions includes: identifying high-frequency dynamic texture regions in video frames according to a background change model, marking the high-frequency dynamic texture regions as background suppression regions, and generating a dynamic background suppression mask.

[0009] As a preferred embodiment of the police drone monitoring method based on image segmentation described in this invention, the step of segmenting the video frame into a target region based on a dynamic background suppression region includes: when performing target region segmentation, using a dynamic background suppression mask to suppress background features in the background region, assigning background constraint weights to pixels in the dynamic background region, and maintaining the original feature response of pixels in the non-background region.

[0010] As a preferred embodiment of the police drone monitoring method based on image segmentation described in this invention, the step of performing region matching and overlap calculation on the target segmented regions in consecutive video frames includes calculating the region overlap index for the target segmented regions in adjacent video frames and establishing a correspondence between target regions based on the region overlap index.

[0011] The regional overlap index is determined by the ratio of the intersection area to the union area of ​​two target regions.

[0012] As a preferred embodiment of the police drone monitoring method based on image segmentation described in this invention, the step of temporal reconstruction of the target region based on multi-frame target region information fusion includes: aligning the target segmentation regions in consecutive video frames in time sequence according to the correspondence between target regions, and performing multi-frame region fusion by combining the spatial position parameters of the target regions with the region contour information to generate a set of target segmentation regions across time sequences.

[0013] As a preferred embodiment of the image segmentation-based police drone monitoring method of the present invention, the analysis of the target's activity pattern within the monitoring area includes: extracting the area parameters, morphological parameters, and spatial coordinate parameters of the target's center point from the target segmentation region set; constructing a target activity feature sequence in chronological order; and analyzing the target's movement trajectory and regional changes within the monitoring area based on the target activity feature sequence.

[0014] The target's movement direction, continuity, and dwell time are calculated based on its movement trajectory over a continuous time period. The target's behavior is then assessed by combining the changes in the area and morphology of the target region. When abnormal activity is identified, a corresponding warning message is generated.

[0015] Another objective of this invention is to provide a police drone monitoring system based on image segmentation. This system can achieve stable segmentation of target areas and analysis of target activity patterns in surveillance videos through a collaborative processing scheme that combines background modeling and dynamic suppression, target area segmentation, target area temporal reconstruction, and abnormal behavior recognition and early warning. This solves the problems of large dynamic background interference, poor stability of target segmentation results, and difficulty in effectively identifying abnormal target activity in existing drone video surveillance technologies.

[0016] As a preferred embodiment of the police drone monitoring system based on image segmentation according to the present invention, it includes: a background modeling and dynamic suppression module, a target region segmentation module, a target region temporal reconstruction module, and an abnormal behavior recognition and early warning module; the background modeling and dynamic suppression module is used to perform texture feature analysis on the background region in the video frame, construct a background change model, and generate a dynamic background suppression region; the target region segmentation module is used to segment the target region of the video frame based on the dynamic background suppression region, and obtain the target segmentation region in each video frame; the target region temporal reconstruction module is used to perform region matching and overlap calculation on the target segmentation region in consecutive video frames, and generate a target segmentation result based on the fusion of target region information from multiple frames; the abnormal behavior recognition and early warning module is used to construct a target activity feature sequence based on the target segmentation result, analyze the target activity pattern, identify abnormal activity behavior, and generate early warning information.

[0017] Another object of the present invention is to provide a police drone monitoring device based on image segmentation, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement a police drone monitoring method based on image segmentation.

[0018] Another object of the present invention is to provide a police drone monitoring storage medium based on image segmentation, wherein a computer program is stored thereon, and when the computer program is executed by a processor, the steps of the police drone monitoring method based on image segmentation are implemented.

[0019] The beneficial effects of this invention are as follows: The image segmentation-based police drone monitoring method provided by this invention analyzes the texture features of the background region and constructs a background change model to generate a dynamic background suppression region, thereby reducing the impact of dynamic background texture on target recognition; by segmenting the video frame based on the dynamic background suppression region, the method effectively extracts the target region in the monitoring video; by performing region matching and multi-frame information fusion on the target segmented region in continuous video frames, the method realizes the temporal reconstruction of the target region, thereby improving the stability of the target recognition result; by constructing a target activity feature sequence based on the target segmentation result and analyzing the target activity pattern, the method identifies abnormal activity behavior within the monitoring area and generates early warning information. Attached Figure Description

[0020] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is an overall flowchart of a police drone monitoring method based on image segmentation provided in Embodiment 1 of the present invention.

[0022] Figure 2 This is a timing diagram of the early warning processing of a police drone monitoring method based on image segmentation provided in Embodiment 1 of the present invention. Detailed Implementation

[0023] To make the above-mentioned objects, features, and advantages of the present invention more apparent and understandable, specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, and not all of them. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the protection scope of the present invention.

[0024] Example 1, referring to Figures 1-2 As one embodiment of the present invention, a police drone surveillance method based on image segmentation is provided, comprising: S1: Based on the surveillance video sequence 100 acquired by the UAV, perform texture feature analysis on the background area in the video frame, construct a background change model 200, and generate a dynamic background suppression region 201.

[0025] Furthermore, texture feature analysis of the background region in the video frames includes extracting texture periodic features from the background region in 100 consecutive video frames of the surveillance video sequence, calculating the background texture variation amplitude between adjacent video frames, and establishing a background variation model 200 based on the background texture variation amplitude, such as... Figure 1 As shown.

[0026] It should be noted that, based on the surveillance video sequence 100 acquired by the drone, consecutive video frames are read frame by frame, and the background region is determined in each video frame. Texture features are extracted from the background region. By calculating the pixel grayscale distribution, texture gradient information, and local texture structure information in the background region, texture feature data of the background region is obtained. After obtaining the texture feature data, texture periodic features are extracted from the consecutive video frames in the surveillance video sequence 100. By analyzing the texture changes in the background region between adjacent video frames, the background texture change amplitude between adjacent video frames is calculated and expressed as: , in, This represents the variation in background texture between adjacent video frames. This represents the total number of pixels within the background area. This refers to the set of pixels corresponding to the determined background region in the current video frame. These are the pixel coordinates in the image. For the first The pixel coordinates in the frame video image are The pixel grayscale value at that location, For the first The pixel coordinates in the frame video image are The pixel grayscale value at that location.

[0027] A background change model 200 is established based on the variation amplitude of background texture to characterize the change patterns of dynamic backgrounds in a scene, within a time window of length [missing information]. Within consecutive video frames, statistical modeling of the background texture variation amplitude yields a background variation model 200, represented as follows: , in, for Background change model 200 at time, The time window length, for The change range of background texture between a frame and the previous frame can be statistically analyzed by analyzing the change range of background texture in multiple consecutive video frames to obtain the average change characteristics of the background region over a period of time, thereby forming a scene background change model 200 that characterizes the dynamic background change law of the scene.

[0028] It should also be noted that by introducing the calculation of texture change amplitude between adjacent video frames and time window statistical modeling, the dynamic background changes in the background region can be continuously described, so that the periodic texture changes generated by the dynamic background can be effectively modeled, and provide a basis for the subsequent generation of the dynamic background suppression region 201.

[0029] Furthermore, generating the dynamic background suppression region 201 includes identifying high-frequency dynamic texture regions in the video frame according to the background change model 200, marking the high-frequency dynamic texture regions as background suppression regions, and generating a dynamic background suppression mask.

[0030] It should be noted that, for a background region pixel in the video image of frame t at the current time, the texture change value of that pixel between adjacent video frames is calculated, and the texture change value is compared with the background change model 200 to determine whether the texture change of that pixel exceeds the background change range. When the texture change value of a pixel is greater than the change feature value corresponding to the background change model 200, the pixel is determined to belong to a high-frequency dynamic texture region; when the texture change value of a pixel is less than or equal to the change feature value corresponding to the background change model 200, the pixel is determined to belong to a stable background region.

[0031] High-frequency dynamic texture regions are aggregated to divide continuously distributed high-frequency dynamic texture pixels into several dynamic texture regions, which are then marked as background suppression regions to generate corresponding dynamic background suppression masks. The established background change model 200 is used to identify high-frequency dynamic texture regions generated by dynamic scene changes, thus providing a basis for suppressing dynamic backgrounds during subsequent target region segmentation.

[0032] It should also be noted that by performing region aggregation processing on the identified high-frequency dynamic texture pixels, the spatially continuously distributed high-frequency dynamic texture pixels are divided into dynamic texture regions and marked as background suppression regions, thereby generating corresponding dynamic background suppression masks. Compared with the method of background region recognition based on only a single frame image, by introducing a background change model 200 to model background texture changes in the time dimension, and combining pixel-level texture change judgment and region aggregation processing, high-frequency dynamic texture regions caused by background disturbance, illumination reflection, and environmental changes can be identified more accurately. This allows for suppression of dynamic backgrounds in subsequent target region segmentation, reducing the interference of background changes on target region recognition, and providing a more stable target region information foundation for subsequent target region temporal reconstruction and abnormal behavior recognition.

[0033] S2: Based on the dynamic background suppression region 201, the video frame is segmented into a target region to obtain the target segmentation region 300 in each video frame.

[0034] Furthermore, the target region segmentation of the video frame based on the dynamic background suppression region 201 includes, when performing target region segmentation, using the dynamic background suppression mask to suppress background features in the background region, assigning background constraint weights to the pixels in the dynamic background region, and maintaining the original feature response of the pixels in the non-background region.

[0035] It should be noted that after obtaining the dynamic background suppression mask, when performing target region segmentation processing on the current video frame, the video image of frame t is read, and the dynamic background suppression region 201 and the non-background region are determined according to the dynamic background suppression mask. For pixels belonging to the dynamic background suppression region 201, background feature suppression processing is performed on their pixel features during target region segmentation, that is, a background constraint weight is assigned to the pixel to reduce its response intensity in the target segmentation process; for pixels not belonging to the dynamic background suppression region 201, their original image feature response is maintained.

[0036] When performing target region segmentation, the pixel feature response values ​​in the video frame are weighted and processed. The feature response values ​​after background constraint processing are represented as follows: , in, These are the feature response values ​​after background constraint processing. The background constraint weight coefficients, For dynamic background suppression mask, where, This indicates that the pixel belongs to the dynamic background suppression region 201. This indicates that the pixel belongs to a non-background area.

[0037] Furthermore, the range of values ​​for the background constraint weight coefficients is as follows: This is used to control the intensity of feature suppression in dynamic background regions. When a pixel belongs to a dynamic background region, the feature response value is reduced, thus weakening it in subsequent object detection and region segmentation. When a pixel belongs to a non-background region, the feature response value remains unchanged.

[0038] After feature suppression processing, target region segmentation is performed on the weighted image features. For example, target regions are extracted from video frames using thresholding, region growing, or deep learning segmentation networks to obtain the target segmentation region 300 in the video frame. Since high-frequency dynamic texture regions generated in the scene have been suppressed by background constraint weights, the interference of dynamic backgrounds such as background disturbances, light spots, and environmental changes on the target detection results can be effectively reduced during target region segmentation. This makes the real target region stand out more in the segmentation results, thereby improving the accuracy and stability of target region extraction.

[0039] It should also be noted that by establishing a feature response weighted model, pixel features belonging to dynamic background regions can be quantitatively suppressed, reducing the contribution of dynamic background regions to target segmentation calculations while keeping the feature responses of real target regions stable. This effectively reduces the interference of complex dynamic textures such as background disturbances, environmental reflections, and light spots on the segmentation results during target region segmentation. Compared to traditional methods that directly segment targets based on image features, the introduction of a dynamic background suppression mask and the synergistic effect of background constraint weights makes the target region more prominent in the segmentation results, improving the accuracy and stability of target region extraction in surveillance videos, and providing a more reliable target region information foundation for subsequent temporal reconstruction of target regions and abnormal behavior recognition.

[0040] S3: Perform region matching and overlap calculation on the target segmentation region 300 in consecutive video frames, and perform temporal reconstruction of the target region based on the fusion of target region information from multiple frames to generate the target segmentation result.

[0041] Furthermore, the process of performing region matching and overlap calculation on the target segmented region 300 in consecutive video frames includes calculating the region overlap index for the target segmented region 300 in adjacent video frames and establishing the correspondence between target regions based on the region overlap index.

[0042] The regional overlap index is determined by the ratio of the intersection area to the union area of ​​two target regions.

[0043] It should be noted that after the target region segmentation in each video frame is completed, region matching and overlap calculation are performed on the target segmented regions 300 in consecutive video frames. Specifically, all target segmented regions 300 are obtained in the t-th frame video image and compared one by one with the target segmented regions 300 in the previous frame video image to determine the correspondence between target regions in adjacent video frames.

[0044] During region matching, the target segmentation region 300 in the t-th frame of the video image is read, and its spatial location range and pixel distribution information in the image are extracted. In the (t-1)-th frame of the video image, candidate target regions spatially adjacent to or close to the target region are searched, and the spatial overlap between the two target regions is statistically analyzed. By calculating the number of overlapping pixels between the two target regions in the image and combining this with the overall pixel range of the two target regions, the degree of regional overlap between the two target regions is obtained. When the regional overlap index between a candidate target region and the current target region is greater than or equal to a preset matching threshold, it is determined that the target region in the current frame and the candidate target region in the previous frame belong to the same target region in consecutive video frames, and a target region association relationship is established between the two frames. When the regional overlap index is less than the matching threshold, it is determined that the two target regions do not belong to the same target region.

[0045] Furthermore, a preferred scheme for the region matching threshold is 0.3 to 0.6. When the region overlap index is not less than this threshold, it is considered that there is a valid correspondence between the two target regions. When the region overlap index of all candidate regions is less than the threshold, the target region is determined to be a newly emerging target region and a new target identifier is assigned to it.

[0046] It should also be noted that by introducing a target region matching mechanism based on regional overlap between consecutive video frames, the ratio between the intersection area and the union area of ​​the target segmentation regions 300 is used as the regional overlap index to quantitatively match and judge the target regions in adjacent video frames. The correspondence between target regions is established based on the preset regional matching threshold, so that the spatial position change of the target in consecutive video frames can be stably identified and associated. Factors such as changes in the shape, position and size of the target region can be considered simultaneously, thereby improving the stability and accuracy of target region matching.

[0047] Furthermore, the temporal reconstruction of the target region based on multi-frame target region information fusion includes: aligning the target segmentation regions 300 in consecutive video frames in time sequence according to the correspondence between target regions; and performing multi-frame region fusion by combining the spatial position parameters and region contour information of the target regions to generate a set of target segmentation regions 300 across time sequences.

[0048] It should be noted that the spatial location parameters of the target segmentation region 300 are obtained in the t-th frame of the video image, including the center coordinates of the target region, the boundary range of the region, and the area information of the region. In the t-1 frame and earlier video frames, target regions corresponding to this target region are searched, and the spatial location parameters and contour information of the corresponding target regions are extracted. Based on the positional changes of the target regions between consecutive video frames, position alignment processing is performed on the target regions in different video frames, so that the regional positions of the same target in each video frame are uniformly mapped to the same reference coordinate system.

[0049] After completing the position alignment process, multi-frame region information fusion is performed on the corresponding target regions in consecutive video frames. Specifically, the spatial location range and region contour information of the corresponding target regions in each video frame are comprehensively processed, the target regions in multiple video frames are superimposed and fused, and the region boundaries are smoothed to obtain a set of 300 target segmentation regions that can reflect the continuous change characteristics of the target in the time series.

[0050] It should also be noted that during the fusion process, the target region is superimposed frame by frame according to the time sequence of the video frames, the pixel set of the target region in each video frame is integrated, and the boundary of the fused target region is corrected, thereby forming a set of target segmentation regions 300 across the time series. Through this multi-frame target region information fusion processing, the target region information in multiple video frames can be comprehensively utilized in the time dimension, effectively reducing the region jitter problem caused by single-frame segmentation error or image noise, making the generated target segmentation region 300 more stable in spatial position and region contour, and providing a more reliable target region data foundation for the subsequent construction of target activity feature sequence 400 and abnormal behavior recognition.

[0051] S4: Construct a target activity feature sequence 400 based on the target segmentation results, analyze the target's activity patterns within the monitoring area, identify abnormal activity behaviors, and generate early warning information 500.

[0052] Furthermore, the analysis of the target's activity patterns within the monitoring area includes extracting the area parameters, morphological parameters, and spatial coordinate parameters of the target's center point from the target segmentation region set 300, constructing a target activity feature sequence 400 in chronological order, and analyzing the target's movement trajectory and regional changes within the monitoring area based on the target activity feature sequence 400.

[0053] The target's movement direction, movement continuity, and dwell time are calculated based on its movement trajectory over a continuous time period. The target's behavior status is determined by combining the changes in the area and morphology of the target region. When abnormal activity is determined, a corresponding warning message 500 is generated.

[0054] It should be noted that the area parameter of the target region is obtained by counting the number of pixels contained in the target region and converting it according to the image pixel scale; the shape parameter of the region is characterized by calculating the boundary length, aspect ratio, and compactness of the target region to represent its shape changes; the spatial coordinate parameter of the target center point is obtained by calculating the geometric center of the pixel set of the target region to obtain the spatial coordinate position of the target in the current video frame. The area parameter, shape parameter, and center point spatial coordinate parameter of the target region are sequentially arranged according to the time order of the video frames to form the target activity feature sequence 400. Based on the activity feature sequence, the motion trajectory of the target in continuous time periods is calculated, and the displacement change of the target center point between adjacent video frames is statistically analyzed to obtain the change in the target's motion direction and the continuity of motion.

[0055] The motion state of a target is determined by statistically analyzing the displacement distance of its center point between adjacent video frames. When the displacement distance between the target's center point and adjacent video frames is less than 5 pixels, the target is considered to be essentially stationary. When the angle of change in the displacement direction of the target's center point within consecutive video frames is within 30°, the target is considered to be in stable motion. When the angle of change in the motion direction between consecutive video frames is greater than 60°, the target is considered to be exhibiting significant directional changes. The parameter range for motion state determination is determined based on statistical results of the target's movement speed and the resolution of the drone footage. In surveillance videos with common resolutions ranging from 1280×720 to 1920×1080, this parameter range effectively distinguishes between normal and abnormal target motion.

[0056] Statistical analysis is performed on the changes in the area of ​​the target region. When the change ratio of the target region area between consecutive video frames exceeds 30%, it is determined that there are abnormal morphological changes in the target region.

[0057] When a target exhibits any of the following behaviors: obvious change of direction, abnormal stay, or abnormal change of form, it is judged as abnormal activity.

[0058] When abnormal activity is identified, a corresponding abnormal behavior warning message 500 is generated. The warning message 500 includes the time of the abnormal behavior, the spatial coordinates of the target, and the type of abnormal behavior, thus providing real-time warning support for the monitoring system. Abnormal behavior identification and warning processing are as follows: Figure 2 As shown.

[0059] It should also be noted that by extracting the area parameters, region morphology parameters, and spatial coordinate parameters of the target center point from the target segmentation region 300, and constructing the target activity feature sequence 400 according to the time sequence of video frames, the spatial position changes, region morphology changes, and area changes of the target within the monitoring area can be continuously recorded in the time dimension. Based on the activity feature sequence, the target's motion trajectory, motion continuity, and dwell state are comprehensively analyzed, and the target's behavior state is determined in conjunction with changes in the target's region morphology and area. This enables the identification of abnormal motion behavior, abnormal dwelling behavior, and abnormal morphological change behavior of the target within the monitoring area. When abnormal activity behavior is determined, corresponding early warning information 500 is generated, thereby providing real-time early warning support for the monitoring system and improving the accuracy and reliability of the monitoring system's identification of abnormal behavior.

[0060] Example 2, an embodiment of the present invention, provides a police drone monitoring system based on image segmentation, including a background modeling and dynamic suppression module, a target region segmentation module, a target region temporal reconstruction module, and an abnormal behavior recognition and early warning module.

[0061] Among them, the background modeling and dynamic suppression module is used to perform texture feature analysis on the background region in the video frame, construct the background change model 200, and generate the dynamic background suppression region 201.

[0062] The target region segmentation module is used to segment video frames based on the dynamic background suppression region 201 to obtain the target segmentation region 300 in each video frame.

[0063] The target region temporal reconstruction module is used to perform region matching and overlap calculation on the target segmentation region 300 in consecutive video frames, and generate target segmentation results based on the fusion of target region information from multiple frames.

[0064] The abnormal behavior recognition and early warning module is used to construct a target activity feature sequence 400 based on the target segmentation results, analyze the target activity pattern and identify abnormal activity behavior, and generate early warning information 500.

[0065] This embodiment also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements a police drone monitoring system based on image segmentation as proposed in the above embodiment.

[0066] This embodiment also provides a computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements a police drone monitoring system based on image segmentation as proposed in the above embodiment.

[0067] If a function is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as a sequenced list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by, or in conjunction with, an instruction execution system, apparatus, or device (such as a computer-based system, a processor-including system, or other system that can fetch and execute instructions from, an instruction execution system, apparatus, or device). For the purposes of this specification, "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transmit programs for use by, or in conjunction with, an instruction execution system, apparatus, or device.

[0069] More specific examples of computer-readable media (a non-exhaustive list) include: electrical connections (electronic devices) having one or more wires, portable computer disk drives (magnetic devices), random access memory (RAM), read-only memory (ROM), erasable and editable read-only memory (EPROM or flash memory), fiber optic devices, and portable optical disc read-only memory (CDROM). Furthermore, computer-readable media can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other medium, followed by editing, interpreting, or otherwise processing as necessary, and then stored in computer memory.

[0070] It should be understood that various parts of the present invention can be implemented in hardware, software, firmware, or a combination thereof. In the above embodiments, multiple steps or methods can be implemented in software or firmware stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, it can be implemented using any one or a combination of the following techniques known in the art: discrete logic circuits having logic gates for implementing logical functions on data signals, application-specific integrated circuits (ASICs) having suitable combinational logic gates, programmable gate arrays (PGAs), field-programmable gate arrays (FPGAs), etc.

[0071] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A police drone surveillance method based on image segmentation, characterized in that, include: Based on the surveillance video sequence (100) acquired by the UAV, the texture features of the background region in the video frame are analyzed, a background change model (200) is constructed, and a dynamic background suppression region (201) is generated. The target region segmentation of the video frame is performed based on the dynamic background suppression region (201) to obtain the target segmentation region (300) in each video frame. Region matching and overlap calculation are performed on the target segmentation region (300) in consecutive video frames, and the target region is temporally reconstructed based on the fusion of target region information from multiple frames to generate the target segmentation result; Based on the target segmentation results, a target activity feature sequence (400) is constructed, and the target's activity pattern in the monitoring area is analyzed to identify abnormal activity behavior and generate early warning information (500).

2. The police drone surveillance method based on image segmentation as described in claim 1, characterized in that: The texture feature analysis of the background region in the video frame includes extracting the texture periodic features of the background region in the continuous video frames of the monitoring video sequence (100), calculating the background texture change amplitude between adjacent video frames, and establishing a background change model (200) based on the background texture change amplitude.

3. The police drone surveillance method based on image segmentation as described in claim 1 or 2, characterized in that: The generation of dynamic background suppression region (201) includes identifying high-frequency dynamic texture regions in the video frame according to the background change model (200), marking the high-frequency dynamic texture regions as background suppression regions, and generating a dynamic background suppression mask.

4. The police drone surveillance method based on image segmentation as described in claim 3, characterized in that: The target region segmentation of the video frame based on the dynamic background suppression region (201) includes using a dynamic background suppression mask to suppress background features in the background region when performing target region segmentation, assigning background constraint weights to the pixels in the dynamic background region, and maintaining the original feature response of the pixels in the non-background region.

5. The police drone surveillance method based on image segmentation as described in any one of claims 1, 2, and 4, characterized in that: The process of performing region matching and overlap calculation on the target segmentation region (300) in consecutive video frames includes calculating the region overlap index for the target segmentation region (300) in adjacent video frames and establishing the correspondence between target regions based on the region overlap index. The regional overlap index is determined by the ratio of the intersection area to the union area of ​​two target regions.

6. The police drone surveillance method based on image segmentation as described in claim 5, characterized in that: The temporal reconstruction of the target region based on multi-frame target region information fusion includes: aligning the target segmentation regions (300) in consecutive video frames in time sequence according to the correspondence between target regions; and performing multi-frame region fusion by combining the spatial position parameters of the target regions with the region contour information to generate a set of target segmentation regions (300) across time sequences.

7. The police drone surveillance method based on image segmentation as described in any one of claims 1, 2, 4, and 6, characterized in that: The analysis of the target's activity pattern within the monitoring area includes extracting the area parameters, morphological parameters, and spatial coordinate parameters of the target's center point from the target segmentation region (300) set, constructing a target activity feature sequence (400) in chronological order, and analyzing the target's movement trajectory and regional changes within the monitoring area based on the target activity feature sequence (400). The target's movement direction, movement continuity, and dwell time are calculated based on its movement trajectory over a continuous time period. The target's behavior status is determined by combining the changes in the area and morphology of the target region. When abnormal activity is determined, a corresponding warning message (500) is generated.

8. A police drone monitoring system based on image segmentation, employing the police drone monitoring method based on image segmentation as described in any one of claims 1 to 7, characterized in that: It includes a background modeling and dynamic suppression module, a target region segmentation module, a target region temporal reconstruction module, and an abnormal behavior recognition and early warning module; The background modeling and dynamic suppression module is used to perform texture feature analysis on the background region in the video frame, construct a background change model (200), and generate a dynamic background suppression region (201). The target region segmentation module is used to segment the video frame based on the dynamic background suppression region (201) to obtain the target segmentation region (300) in each video frame. The target region temporal reconstruction module is used to perform region matching and overlap calculation on the target segmentation region (300) in consecutive video frames, and generate target segmentation results based on the fusion of target region information from multiple frames; The abnormal behavior identification and early warning module is used to construct a target activity feature sequence (400) based on the target segmentation result, analyze the target activity pattern and identify abnormal activity behavior, and generate early warning information (500).

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the police drone monitoring method based on image segmentation as described in any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the police drone monitoring method based on image segmentation as described in any one of claims 1 to 7.

Images