An unmanned aerial vehicle priority interception method and system based on a dense optical flow algorithm

Abstract

This invention provides a method and system for priority interception of unmanned aerial vehicles (UAVs) based on a dense optical flow algorithm, relating to the field of UAV interception technology. The method includes: acquiring flight image data of UAVs within an interception area; preprocessing the flight image data to obtain a corresponding standardized image dataset; calculating the optical flow field distribution of pixels in adjacent frame image data based on a dense optical flow algorithm and the standardized image dataset; extracting motion feature parameters of the UAVs from the optical flow field distribution to form a motion feature dataset of the UAVs; calculating the threat level quantification value of all detected UAVs within the UAV flight area based on the motion feature dataset and the actual distance between the UAVs and the key protection area; generating a priority interception sequence based on the threat level quantification value; and generating corresponding priority interception strategies based on different levels of UAV targets in the priority interception sequence.

Description

Technical Field

[0001] This invention relates to the field of drone interception technology, specifically a drone priority interception method and system based on dense optical flow algorithm. Background Technology

[0002] With the rapid popularization of drone technology, drones are increasingly used in various fields, but at the same time, they also bring security risks of unauthorized intrusion into key protected areas, posing a serious threat to public safety and the protection of core facilities.

[0003] Existing drone interception technologies struggle to scientifically quantify threat levels and generate reasonable priority interception plans, thus affecting the targeting and effectiveness of interception strategies and failing to meet the actual needs of key protection areas for rapid and precise drone interception.

[0004] Therefore, this paper presents a method and system for priority interception of drones based on dense optical flow algorithm, which improves the accuracy of drone motion feature extraction and the scientific nature of threat level assessment in complex scenarios, and ensures the efficient conduct of interception operations. Summary of the Invention

[0005] To address the aforementioned technical problems, the present invention aims to provide a method and system for priority interception of unmanned aerial vehicles (UAVs) based on a dense optical flow algorithm.

[0006] To achieve the above objectives, the present invention provides the following technical solution: a method for priority interception of unmanned aerial vehicles based on a dense optical flow algorithm, the method comprising: Collect flight image data of the UAV within the interception area; perform preprocessing operations on the flight image data to obtain the corresponding standardized image dataset; Based on the dense optical flow algorithm and according to the standardized image dataset, the optical flow field distribution of pixels in adjacent frame image data is calculated; the motion feature parameters of the UAV are extracted from the optical flow field distribution to form the motion feature dataset of the UAV. Based on the motion feature dataset and the actual distance between the UAV and the key protection area, calculate the threat level quantification value of all detected UAVs within the UAV's flight area; and generate a priority interception sequence based on the threat level quantification value. Based on the different levels of UAV targets in the priority interception sequence, a corresponding priority interception strategy is generated.

[0007] Furthermore, the preprocessing operation for the flight image data includes: The images are denoised using an adaptive median filtering algorithm; the denoised flight image data is converted into grayscale images using a weighted average method, and the corresponding grayscale image size is standardized; the contrast of the grayscale images is enhanced based on a histogram equalization algorithm; and the preprocessed flight image data is recorded as a standardized image dataset.

[0008] Furthermore, the process of calculating the optical flow field distribution of pixels in adjacent frame image data based on the DeepFlow deep optical flow algorithm and according to the standardized image dataset includes: Based on the DeepFlow deep optical flow algorithm, a deep optical flow network model adapted to UAV scenarios was constructed and pre-trained. The real-time standardized image dataset is input into the pre-trained deep optical flow network model to obtain the corresponding initial optical flow field; The initial optical flow field is optimized to obtain the optimized optical flow field distribution.

[0009] Furthermore, the process of constructing a deep optical flow network model adapted to the drone scenario and optimizing the initial optical flow field includes: Set the number of layers in the multi-scale pyramid feature encoder and obtain the feature maps of the corresponding layers. Introduce a channel attention weight matrix into the feature map of each layer. A deep feature correlation matching layer is set up, and a feature correlation matching cost body is constructed based on the feature map output by the multi-scale pyramid feature encoder. Based on the feature correlation matching cost body, the matching information is fused and encoded through a convolutional layer to obtain the initial matching features, and then the corresponding initial sparse optical flow field is obtained. The upsampling iterative optical flow refinement decoder is configured to perform layer-by-layer upsampling on the multi-scale pyramid feature encoder, and combined with residual optical flow compensation, the initial sparse optical flow field is optimized into an initial dense optical flow field. A spatiotemporal consistency constraint layer for optical flow is set up, and the initial dense optical flow field is corrected by this constraint, thereby outputting an optimized optical flow field distribution; A combined loss function is constructed by using the endpoint error loss function and the smoothness constraint loss function. This function is used when the loss is continuous. When the combined loss function value of each training batch meets the requirements, the pre-trained weight parameters are saved.

[0010] Furthermore, the process of extracting the motion feature parameters of the UAV from the optical flow field distribution to form the UAV motion feature dataset includes: Pixel-level motion vector sets of UAV target regions are extracted based on optical flow field distribution; The instantaneous motion feature parameters of the UAV are calculated based on the pixel-level motion vector set; the instantaneous motion feature parameters include instantaneous motion speed, instantaneous motion direction, and motion trajectory curvature; then, a spatiotemporal UAV motion feature dataset is constructed and standardized.

[0011] Furthermore, the process of extracting pixel-level motion vector sets of the UAV target region based on the optical flow field distribution includes: The pixel region of the UAV target is selected from adjacent frame images using a target detection algorithm; the horizontal and vertical motion vectors corresponding to all pixels in the target region are extracted from the optical flow field distribution to construct a pixel-level motion vector set of the UAV target. For noise vectors existing in the target area, a statistical filtering method is used to remove outliers. The mean and standard deviation of each component in the pixel-level motion vector set are calculated, and a preset filtering threshold is set. Based on the mean, standard deviation and filtering threshold, the pixel-level motion vector set of the target area after denoising is obtained.

[0012] Furthermore, the process of calculating the instantaneous motion characteristic parameters of the UAV based on the pixel-level motion vector set includes: Using the acquisition time interval between adjacent frames as the time reference, the instantaneous motion characteristic parameters of the UAV are calculated based on the denoised pixel-level motion vector set; including instantaneous motion speed, instantaneous motion direction, instantaneous acceleration, and motion trajectory curvature.

[0013] Furthermore, the process of calculating the threat level quantification value of all detected drones within the drone's flight area, based on the aforementioned motion feature dataset and the actual distance between the drone and the key protection area, includes: By combining the motion characteristic parameters of UAVs and the actual distance between UAVs and key protection areas, a UAV threat level assessment index system is constructed. The UAV threat level assessment index system includes distance index, speed index, acceleration index, azimuth angle index, and trajectory curvature index. The weights of each core indicator were determined using the analytic hierarchy process (AHP), and the threat level of the drones was calculated using the fuzzy comprehensive evaluation method. Based on the calculated threat level values, the drones were classified into four threat levels. All detected drones within the monitoring area are sorted from highest to lowest threat level, and drones of the same threat level are sorted from closest to furthest from the key protection area to generate a priority interception sequence.

[0014] A second aspect of the present invention also provides a drone priority interception system based on a dense optical flow algorithm, comprising: a data acquisition module, a data preprocessing module, a feature analysis module, a threat level quantification module, and an intelligent strategy module; The data acquisition module is used to collect flight image data of the drone within the interception area; The data preprocessing module is used to preprocess flight image data to obtain the corresponding standardized image dataset. The feature analysis module, based on the dense optical flow algorithm and according to the standardized image dataset, calculates the optical flow field distribution of pixels in adjacent frame image data; extracts the motion feature parameters of the UAV from the optical flow field distribution to form the motion feature dataset of the UAV; The threat level quantification module is used to calculate the threat level quantification value of all detected drones within the drone's flight area based on the motion feature dataset and the actual distance between the drone and the key protection area; and to generate a priority interception sequence based on the threat level quantification value. The intelligent strategy module is used to generate corresponding priority interception strategies based on different levels of UAV targets in the priority interception sequence.

[0015] Compared with existing technologies, the beneficial effects of this invention are as follows: The UAV priority interception method and system based on dense optical flow algorithm provided by this invention effectively solves the problems of insufficient accuracy in UAV motion feature extraction, unscientific threat level assessment, and poor targeting of interception strategies in existing technologies. Through targeted image preprocessing, environmental interference factors can be eliminated, providing a high-quality data foundation for subsequent optical flow field calculations and ensuring data reliability. A network model adapted to UAV scenarios is constructed based on the improved DeepFlow deep optical flow algorithm. Combined with multi-dimensional optimization processing and combined loss function training, the robustness and accuracy of optical flow field estimation are significantly improved, enabling precise capture of pixel-level motion information of the UAV target area. Multi-dimensional instantaneous motion feature parameters of the UAV are extracted based on the optical flow field. A threat level assessment system is constructed by combining multiple indicators. Scientific weight determination and evaluation methods are used to quantify the threat level, objectively reflecting the UAV's intrusion intent and threat level, and generating a reasonable priority interception sequence. Through the collaborative work of various modules, the entire process of UAV monitoring, feature extraction, threat assessment, and interception strategy generation is optimized, improving the accuracy and efficiency of UAV interception. It can adapt to the key protection needs in different complex scenarios, providing strong support for the security protection of key areas. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0017] Figure 1 This is a schematic diagram illustrating the steps of a drone priority interception method based on a dense optical flow algorithm.

[0018] Figure 2 This is a schematic diagram of a module for a drone priority interception system based on a dense optical flow algorithm. Detailed Implementation

[0019] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this invention will be described in detail below. Obviously, the described embodiments are merely some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other implementation methods obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0020] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of this application described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0021] Example 1: like Figure 1 As shown, a method for priority interception of drones based on a dense optical flow algorithm includes the following steps: Step S1: Collect flight image data of the UAV within the interception area; perform preprocessing operations on the flight image data to obtain the corresponding standardized image dataset; In this embodiment, prioritizing the interception of drones refers to the instruction executed when the target drone flies to the corresponding interception area. When it exceeds this interception area, the drone interception system loses its interception effect. It should be emphasized that the interception area can be set according to actual needs, and the larger the interception area, the worse the accuracy of drone interception at the edge of the interception area will be.

[0022] It should be further noted that there may be one or more target drones in the interception area. This application analyzes the threat level of the corresponding target drones and then executes priority interception commands to reduce the risk of drone intrusion.

[0023] When collecting flight image data of drones within the interception area, it is necessary to select appropriate acquisition equipment. Based on the required scope, terrain, and monitoring distance of the key protection area, different models of high-definition network cameras, infrared thermal imaging cameras, and panoramic cameras can be selected to form a multi-source image acquisition network. Specifically, high-definition network cameras should have a resolution of at least 1920×1080 and a frame rate of at least 60fps to ensure image clarity in well-lit daytime environments; infrared thermal imaging cameras should have a detection range of at least 500m and a resolution of at least 384×288 to address image acquisition issues in low-visibility environments such as nighttime, heavy fog, and sandstorms; and panoramic cameras should have a field of view of at least 180° to achieve comprehensive coverage of the monitoring area and avoid blind spots.

[0024] According to the preset equipment deployment plan, the corresponding data acquisition equipment is deployed to collect flight image data of the UAV within the interception area. The specific deployment process will not be described in detail.

[0025] It should be noted that the flight image data includes, but is not limited to, high-definition color digital image data, infrared thermal imaging image data, and wide-view color panoramic stitched images.

[0026] In practical applications, to eliminate environmental interference factors such as illumination changes, atmospheric scattering, and image noise, and to adapt to the feature extraction requirements of the DeepFlow optical flow algorithm, thereby improving the accuracy of subsequent optical flow estimation and motion feature extraction, the specific steps for preprocessing the flight image data to obtain the corresponding standardized image dataset include: A1: The image is denoised using an adaptive median filtering algorithm. This algorithm dynamically adjusts the filtering window size (e.g., from 3×3 to 7×7) based on the noise type of each pixel, performs a weighted average on the noisy pixels, removes common noises such as Gaussian noise and salt-and-pepper noise, and preserves the edge details and texture features of the corresponding flight image data. It should be noted that edge details and texture features are the foundation for DeepFlow's deep optical flow algorithm to extract deep semantic features through convolutional neural networks, which can avoid the loss of key features of the drone target due to excessive noise reduction. The processing procedure is as follows: traverse each pixel in the flight image data and calculate the maximum, minimum, and mean values ​​in its neighborhood; then determine whether the current pixel is a noise point. If the pixel value exceeds the reasonable range of the maximum and minimum values, it is determined to be a noise point; finally, noise points are replaced with the neighborhood mean, while non-noise points retain their original values.

[0027] A2: Since the DeepFlow optical flow algorithm does not require additional color information from the input flight image data, and color images contain a lot of redundant color information, which would increase the computational complexity of the network, a weighted average method is used to convert the corresponding flight image data into grayscale images. The grayscale images can retain the brightness information and target contour features of the original image to the greatest extent, ensuring the contrast between the UAV target and the background, and providing a clear base image for the DeepFlow optical flow algorithm to accurately match the pixel motion relationship between adjacent frames.

[0028] A3: The DeepFlow optical flow algorithm's convolutional neural network has fixed requirements for the size of the input grayscale image. To unify the network input format, reduce model computation, and improve training and inference efficiency, a bilinear interpolation algorithm is used to normalize the grayscale image to a fixed size, such as 640×480 pixels. The target pixel value is determined by calculating the weighted values ​​of the four neighboring pixels around the target pixel. This method has the advantages of smooth interpolation and low image distortion, avoiding distortion of the UAV target shape during size scaling and ensuring the continuity of the target motion trajectory in adjacent frames. The processing steps are as follows: calculate the scaling ratio between the original grayscale image and the target grayscale image, then map the pixels of the original grayscale image to the positions of the target grayscale image according to the scaling ratio, and finally obtain the grayscale value of each pixel in the target grayscale image through weighted calculation.

[0029] A4: To address issues such as low contrast and blurred target features in grayscale images due to insufficient lighting and atmospheric scattering, a histogram equalization algorithm is employed to enhance the contrast of grayscale images. Enhancing grayscale image contrast improves the accuracy of the DeepFlow optical flow algorithm in matching pixel motion between adjacent frames. The processing involves adjusting the grayscale histogram distribution to ensure uniform distribution of grayscale values ​​across the entire grayscale range, thereby improving image clarity and target recognition. For grayscale images from infrared thermal imaging data, an adaptive histogram equalization algorithm is used. This involves processing the image in blocks and adaptively adjusting the histogram distribution of each block to avoid the loss of local details caused by global histogram equalization, thus preserving the texture and contour features of the UAV target and supporting the DeepFlow optical flow algorithm in extracting accurate deep features. The flight image data after the above preprocessing operations is denoted as a standardized image dataset.

[0030] It should be noted that, based on the training and inference requirements of the DeepFlow optical flow algorithm, the dataset is subject to targeted integrity verification to avoid interruption of feature extraction due to image corruption; the final output standardized image dataset must meet the requirements of continuous adjacent frames, complete target features, and uniform size.

[0031] Step S2: Based on the dense optical flow algorithm and according to the standardized image dataset, calculate the optical flow field distribution of pixels in adjacent frame image data; extract the motion feature parameters of the UAV from the optical flow field distribution to form the motion feature dataset of the UAV; In this embodiment, the dense optical flow algorithm used is based on the DeepFlow algorithm. The DeepFlow algorithm accurately estimates the motion (i.e., optical flow) of pixels between adjacent image frames through a deep learning network. The working principle includes: First step feature extraction: the encoder extracts deep features from two consecutive images to capture image details at different scales; Second step optical flow prediction: the decoder upsamples the extracted features to restore the original image size, and uses skip connections to supplement details, outputting the motion vector (i.e., optical flow field) of each pixel, that is, the displacement of the pixel in the horizontal and vertical directions; Third step optimization and correction: the preliminary optical flow field is smoothed by guided filtering to remove noise and outliers, and finally a precise and smooth optical flow field is obtained, which provides a foundation for subsequent extraction of UAV motion features (such as speed, direction, etc.).

[0032] In this example, the DeepFlow optical flow algorithm is used to calculate the optical flow field distribution of pixels in adjacent frame image data. Compared with traditional Farneback optical flow algorithms and Lucas-Kanade optical flow algorithms, the DeepFlow algorithm is based on a deep learning framework. Instead of directly matching the grayscale of image pixels, it matches the deep semantic features extracted by the CNN through a convolutional neural network (CNN), improving the robustness of matching by an order of magnitude. It solves the problem of pixel grayscale distortion in UAV images under low light and cloud cover, and has stronger environmental adaptability and optical flow estimation accuracy. Especially under conditions where the UAV target is small, the background is complex (such as cloud cover or building occlusion), or the light is low, it can more accurately capture the motion information of pixels, thereby improving the reliability of UAV motion feature extraction.

[0033] In practical applications, based on the DeepFlow optical flow algorithm and the standardized image dataset, the steps for calculating the optical flow field distribution of pixels in adjacent frame image data are as follows: B1: Based on the DeepFlow deep optical flow algorithm, construct a deep optical flow network model adapted to the drone scenario and complete pre-training; B2: Input the real-time standardized image dataset into the pre-trained deep optical flow network model to obtain the corresponding initial optical flow field; B3: The initial optical flow field is optimized to obtain the optimized optical flow field distribution.

[0034] In step B1, the steps for constructing a deep optical flow network model adapted to the drone scenario are as follows: It should be noted that the overall architecture of the deep optical flow network model adapted for drone scenarios, based on the DeepFlow deep optical flow algorithm, is as follows: ;in, This represents the overall architecture of the deep optical flow network model; A multi-scale pyramid feature encoder representing a deep optical flow network model is used for core feature extraction from UAVs. This represents a deep feature correlation matching layer in a deep optical flow network model, used to suppress interference from complex backgrounds. This represents an upsampled iterative optical flow refinement decoder for a deep optical flow network model, used to improve the accuracy of optical flow estimation; This represents the spatiotemporal consistency constraint layer for optical flow in a deep optical flow network model, used to eliminate erroneous optical flow vectors. This indicates a chain operation relationship between layers, where the output of the previous layer serves as the input of the next layer.

[0035] B11: The multi-scale pyramid feature encoder Pre-set The layer, and the corresponding feature map of the layer are: ; In the formula, , Indicates the corresponding frame; express Feature map of layer 0 of frame; express Feature map of layer 0 of frame; express Frame number Feature maps of the layer; express Frame number Feature maps of the layer; This is a downsampling convolution with a stride of 2, used to halve the image size and double the feature dimension; It is a 3×3 convolution operation, used to extract the receptive field and improve computational efficiency; For the first Layer convolution bias term; It is a ReLU activation function used to avoid gradient vanishing and is suitable for weak feature extraction from low-light drone images.

[0036] Introduce a channel attention weight matrix into each layer of feature map. ;in, For the first The number of layer feature channels; it should be noted that the deep semantic features of the UAV target are highlighted and the background noise features are suppressed through the attention mechanism.

[0037] To further explain, the channel attention weight matrix ; ; ;in, Global average pooling is used to achieve global information aggregation of feature channels; , Learnable weights and biases for the attention mechanism; For the weighted output Frame number Feature maps of the layer; This is a channel-by-channel, element-by-element multiplication operation; For the weighted output Frame number Feature map of the layer.

[0038] B12: The deep feature correlation matching layer , used according to , Construct a feature correlation matching cost body The similarity between feature points in adjacent frames is used to characterize the matching similarity, and the calculation formula is: ; In the formula, These are the pixel coordinates on the corresponding feature map. , where is the search offset of the feature point; in this application Take 4, that is, each Search in adjacent frames All pixels within the range meet the corresponding matching accuracy and calculation speed; The dot product operation is performed on the feature vectors. The closer the result is to 1, the higher the feature matching degree.

[0039] Based on feature correlation matching cost body The initial matching features are obtained by fusing and encoding the matching information through convolutional layers. Thus, the corresponding initial sparse optical flow field is obtained: .

[0040] B13: The upsampling iterative optical flow refinement decoder Used for multi-scale pyramid feature encoder Layer-by-layer upsampling, combined with residual optical flow compensation, optimizes the initial sparse optical flow field into an initial dense optical flow field. The recursive iterative formula is as follows: ; In the formula, Represents the initial sparse optical flow field of the corresponding layer. ,when At time , it represents the initial sparse optical flow field of the top-level input in the iteration, i.e., the multi-scale pyramid feature encoder. Top floor The initial sparse optical flow field, whose value is the output of the feature correlation matching layer mentioned above, is the unique starting point for all iterations of the decoder; when When, it represents the initial sparse optical flow field input to the intermediate layer of the iteration; Indicates the first Upsampling of the layer coarsely estimates the optical flow field; Upsampling for bilinear interpolation with a step size of 2; To restore the optical flow vector scale, compensate for the downsampling size scaling; For feature channel splicing operations, features from adjacent frames are fused with upsampled optical flow information; For the first The residual optical flow field of the layer is used to correct the optical flow error of the upsampling; This represents the initial dense optical flow field of the corresponding output.

[0041] B14: The optical flow spatiotemporal consistency constraint layer To address issues such as cloud occlusion, complex background textures, and low-light noise in drone images, this method corrects local abrupt changes in the optical flow field by removing erroneous optical flow vectors. The constraint formula is as follows: ; In the formula, For photometric consistency constraints, it characterizes the gray-level similarity of adjacent frame pixels after optical flow mapping. The smaller the gray-level difference, the more reliable the optical flow vector. For optical flow smoothness constraints: , The horizontal and vertical gradients of the optical flow field are used to constrain the local smoothness of the optical flow field and suppress abrupt changes in optical flow caused by noise. The value is 0.2, which is used as a smoothing coefficient in this implementation to balance the constraint strength and optical flow accuracy.

[0042] The initial dense optical flow field is corrected by this constraint, resulting in an optimized optical flow field distribution: ; B15: Use the Endpoint Error (EPE) loss function With smoothness constraint loss function Construct the combined loss function ; In this embodiment, endpoint error loss Smoothness constraint loss Combination loss function ;in, , For the loss weight coefficients, satisfying ; The total number of pixels in the corresponding image for a standardized image dataset; For the pixels predicted by the model Optical flow vector; To standardize the true optical flow vector in the image dataset; , These are the gradient operators for the horizontal and vertical directions, respectively.

[0043] When continuous The combined loss function value of each training batch satisfies When the model converges, the pre-trained weight parameters are saved, where, This is the preset convergence threshold.

[0044] In practical applications, the steps for extracting the motion feature parameters of the UAV from the optical flow field distribution to form the UAV's motion feature dataset are as follows: C1: Extracting pixel-level motion vector sets of UAV target regions based on optical flow field distribution; In step C1, the specific process of extracting the pixel-level motion vector set of the UAV target region based on the optical flow field distribution is as follows: The pixel region of the drone target is selected from adjacent frames of the image using a target detection algorithm, denoted as . ;in , , The dimensions of the preprocessed image; Extract the target region from the optical flow field distribution. All pixels The corresponding horizontal and vertical motion vectors are used to construct a pixel-level motion vector set for the UAV target. The pixel-level motion vector set ; For noise vectors present in the target area, statistical filtering methods are used to remove outliers and calculate pixel-level motion vector sets. The mean of each component , with standard deviation , The preset filter threshold is According to the mean , Standard deviation , and filter threshold This yields a pixel-level motion vector set for the denoised target region.

[0045] C2: Calculate the instantaneous motion characteristic parameters of the UAV based on the pixel-level motion vector set; the instantaneous motion characteristic parameters include instantaneous motion speed, instantaneous motion direction, and motion trajectory curvature; In step C2, the steps for calculating the instantaneous motion characteristic parameters of the UAV based on the pixel-level motion vector set are as follows: The acquisition time interval of adjacent frames Using the time reference, the instantaneous motion feature parameters of the UAV are calculated based on the denoised pixel-level motion vector set; C21: The specific process for calculating instantaneous velocity is as follows: Calculate the weighted average vector of all pixel motion vectors in the pixel-level motion vector set. The weight is the gray-level gradient value of the corresponding pixel within the target area. The larger the gradient value, the higher the weight, to highlight the motion characteristics of the core target area. The magnitude of the weighted average vector is the instantaneous velocity of the UAV. The calculation formula is: ; ; ; In the formula, The target pixel area after noise reduction; For pixels Gray gradient weights; , For weighted average level and vertical pixel displacement.

[0046] C22: The specific process for calculating the instantaneous motion direction is as follows: with the horizontal direction to the right of the image coordinate system as 0... The reference direction is calculated using the arctangent function to determine the direction angle of the average vector. , which is the instantaneous direction of motion of the drone, is calculated using the following formula: .

[0047] C23: The specific process for calculating instantaneous acceleration is as follows: For three consecutive frames of images... , , Calculate separately → and → Instantaneous velocity over a period of time , , at time intervals The step size is used to calculate the magnitude of the rate of change of velocity as the instantaneous acceleration. The formula is: .

[0048] C24: The specific process for calculating the curvature of the motion trajectory is as follows: Select continuous... The sequence of drone center position coordinates corresponding to the frame image The trajectory curve equation is constructed using a polynomial fitting method. The instantaneous curvature of the trajectory is calculated using the curvature formula. The formula is: ;in, , These are the first and second derivatives of the trajectory curve, respectively.

[0049] C3: Construct a spatiotemporal drone motion feature dataset and complete the standardization process.

[0050] In step C3, the instantaneous motion feature parameters calculated at different times are associated with the corresponding timestamps and the pixel coordinate range of the target area to form a structured feature data tuple. ,in, The effective frame sequence length is determined. Secondly, the feature parameters are standardized to eliminate the impact of dimensional differences on subsequent analysis. The standardized feature data is then sorted according to time series and stored in conjunction with the corresponding frame image identifiers and optical flow field distribution data to form a UAV motion feature dataset containing spatiotemporal correlation information. A data indexing mechanism is then established to enable rapid association and querying of feature parameters with original images and optical flow field data.

[0051] Step S3: Based on the motion feature dataset and the actual distance between the UAV and the key protection area, calculate the threat level quantification value of all detected UAVs within the UAV's flight area; generate a priority interception sequence based on the threat level quantification value; It should be noted that the actual distance between the drone and the key protection area refers to the actual straight-line distance between the drone and the boundary of the key protection area. The closer the distance, the higher the threat level.

[0052] In practical applications, a drone threat level assessment index system is constructed by combining the drone's motion characteristic parameters and the actual distance between the drone and key protection areas. This system specifically includes the following core indicators: 1. Distance Index: Based on the security control requirements of key protected areas, distances are divided into three levels: (greater than) (low-risk distance) ( (medium-risk distance) and (less than) (High-risk distance); specific distance thresholds can be set according to actual needs, and this application does not require them.

[0053] 2. Speed ​​Index: The drone's flight speed. The faster the speed, the shorter the time available for interception and response, and the higher the threat level. Based on the aforementioned motion feature dataset, speed is divided into three levels: (less than) (low speed) ( (medium speed) and (greater than) (High speed); the specific speed threshold can be set according to actual needs, and this application does not require it.

[0054] 3. Acceleration Index: The flight acceleration of a drone indicates its higher threat level; drones accelerating towards key protection areas pose a greater threat. Based on the aforementioned motion feature dataset, acceleration is divided into three levels: (less than) (Slow down flight, low risk) (equal (Uniform speed flight, medium risk) and (greater than) (Accelerated flight, high risk); the specific acceleration threshold can be set according to actual needs, and this application does not require it.

[0055] 4. Azimuth Angle Index: The angle between the drone's flight direction and its direction toward the key protection area. The smaller the angle, the higher the threat level. Based on the aforementioned motion feature dataset, the azimuth angle is divided into three levels: (greater than) (Fly away or sideways, low risk) ( (Oblique flight, medium risk) and (less than) (Decelerated flight, low risk) The specific azimuth angle threshold can be set according to actual needs, and this application does not require it.

[0056] 5. Trajectory Curvature Index: The curvature of the drone's trajectory; drones flying in a straight line towards key protection areas pose a higher threat level. Based on the aforementioned motion feature dataset, trajectory curvature is divided into three levels: (greater than) (Significant turn, low risk) ( (Slight turn, medium risk) and (less than) (Approximately straight line, high risk); the specific trajectory curvature threshold can be set according to actual needs, and this application does not require it.

[0057] The specific steps for determining the weights of each core indicator using the Analytic Hierarchy Process (AHP) and combining it with the fuzzy comprehensive evaluation method to calculate the quantitative value of the threat level of the UAV are as follows: A judgment matrix was constructed using the analytic hierarchy process (AHP). Experts in fields such as security protection and drone technology were invited to score the importance of each core indicator, and the weights of each core indicator were calculated. Let the weights of each core indicator be: distance indicator weight. Speed ​​index weight Acceleration index weight directional indicator weights Trajectory curvature index weight The weights of each indicator are such that their sum equals 1.

[0058] Each core indicator is assigned a corresponding score for different levels. In this embodiment, for the distance indicator, The corresponding rating is 1. The corresponding rating is 3 and The corresponding score is 5; in the speed index, The corresponding rating is 1. The corresponding rating is 3 and The corresponding score is 5; in the acceleration index, The corresponding rating is 1. The corresponding rating is 3 and The corresponding score is 5; in the orientation angle index, The corresponding rating is 1. The corresponding rating is 3 and The corresponding score is 5; in the trajectory curvature index, The corresponding rating is 1. The corresponding rating is 3 and The corresponding rating is 5.

[0059] The threat level quantification value of drones is calculated using a weighted summation method. The threat level quantification value ; in, , , , as well as These are the score values ​​for distance, speed, acceleration, direction, and trajectory curvature, respectively. Threat level quantification value. The range of values ​​is The higher the value, the higher the threat level of the drone.

[0060] Based on the calculated threat level quantification value Drones are classified into four threat levels, with the specific classification criteria as follows: 1. Level 1 Threat (Emergency Interception): These types of drones typically fly at high speed and accelerate towards key protected areas, and are relatively close, so they need to be intercepted immediately. 2. Level 2 Threat (Priority Interception): These types of drones typically fly at medium to high speeds, at a constant speed, either forward or at an angle, towards key protected areas at a moderate distance. They should be intercepted first after the first-level threat drones have been intercepted. 3. Level 3 Threat (Standard Interception): These types of drones typically fly at low speed, constant speed, or deceleration, and fly away from or to the side of key protection areas. They are at a relatively long distance and can be intercepted routinely after first- or second-level threat drones have been intercepted. 4. Level 4 Threat (Monitoring and Observation): Such drones do not have obvious intrusion intentions, their flight status is stable and they are far away from key protected areas, so there is no need to intercept them immediately; it is only necessary to continuously monitor their flight trajectory.

[0061] All detected drones within the monitoring area are sorted from highest to lowest threat level. Drones with the same threat level are then sorted from closest to furthest from the key protection area, generating a priority interception sequence. This priority interception sequence is pushed to the interception command system in real time, providing a basis for the formulation of subsequent interception strategies.

[0062] Step S4: Generate corresponding priority interception strategies based on the different levels of UAV targets in the priority interception sequence.

[0063] In practical applications, based on the different threat levels of UAV targets in the priority interception sequence, and combined with their motion characteristic parameters and flight trajectories, corresponding interception timing and location strategies are generated: 1. Level 1 Threat Drone Interception Strategy: Interception Timing: Initiate interception immediately upon target detection, ensuring interception is completed before the drone enters the core area of ​​the key protection zone, such as when it is ≥500m from the boundary; Interception Location: Select a location on the drone's flight path 1000-1500m from the boundary of the key protection zone as the interception point. This location ensures the effective range of the interception equipment while preventing drone debris from falling into the core area.

[0064] 2. Secondary Threat Drone Interception Strategy: Interception Timing: Initiate immediately after the primary threat drone is intercepted, ensuring that the interception is completed before the drone enters the boundary of the key protection area, such as when it is ≥1000m from the boundary; Interception Location: Select a location on the drone's flight path that is 2000-2500m from the boundary of the key protection area as the interception point.

[0065] 3. Level 3 Threat Drone Interception Strategy: Interception Timing: After the Level 1 and Level 2 threat drones are intercepted, the interception will be initiated at an opportune time based on changes in the drone's flight trajectory; Interception Location: Select a location on the drone's flight trajectory that is more than 3,000 meters away from the boundary of the key protection area as the interception point to avoid impacting the surrounding area during the interception process.

[0066] 4. Level 4 Threat Drone Interception Strategy: No need to activate the interception equipment. The background monitoring system continuously tracks its flight trajectory. If it detects a change in its flight status (such as turning towards a key protection area or accelerating flight), its threat level is recalculated, the priority interception sequence is adjusted, and a corresponding interception strategy is generated.

[0067] Example 2: like Figure 2 As shown, a drone priority interception system based on dense optical flow algorithm is disclosed. The system includes, but is not limited to, a data acquisition module, a data preprocessing module, a feature analysis module, a threat level quantification module, and an intelligent strategy module. The data acquisition module is used to collect flight image data of the drone within the interception area; The data preprocessing module is used to preprocess flight image data to obtain the corresponding standardized image dataset. The feature analysis module, based on the dense optical flow algorithm and according to the standardized image dataset, calculates the optical flow field distribution of pixels in adjacent frame image data; extracts the motion feature parameters of the UAV from the optical flow field distribution to form the motion feature dataset of the UAV; The threat level quantification module is used to calculate the threat level quantification value of all detected drones within the drone's flight area based on the motion feature dataset and the actual distance between the drone and the key protection area; and to generate a priority interception sequence based on the threat level quantification value. The intelligent strategy module is used to generate corresponding priority interception strategies based on different levels of UAV targets in the priority interception sequence.

[0068] Optionally, in this embodiment, those skilled in the art will understand that all or part of the steps in the various methods of the above embodiments can be implemented by a program instructing the hardware related to the terminal device. The program can be stored in a computer-readable storage medium, which may include: flash drive, read-only memory (ROM), random access memory (RAM), disk or optical disk, etc.

[0069] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0070] If the integrated units in the above embodiments are implemented as software functional units and sold or used as independent products, they can be stored in the aforementioned computer-readable storage medium. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, or all or 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 one or more electronic devices to execute all or part of the steps of the methods described in the various embodiments of this application.

[0071] In the above embodiments of this application, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0072] In the several embodiments provided in this application, it should be understood that the disclosed application can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of units is only a logical functional division, and in actual implementation, there may be other division methods. For example, multiple units or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces; the indirect coupling or communication connection of units or modules may be electrical or other forms.

[0073] The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the units can be selected to achieve the purpose of this embodiment according to actual needs.

[0074] Furthermore, the functional units in the various embodiments of this application can be integrated into one processing unit, or each unit can exist physically separately, or two or more units can be integrated into one unit. The integrated unit can be implemented in hardware or as a software functional unit.

[0075] The above description is only a preferred embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for priority interception of unmanned aerial vehicles (UAVs) based on a dense optical flow algorithm, characterized in that, The method includes: Collect flight image data of the UAV within the interception area; perform preprocessing operations on the flight image data to obtain the corresponding standardized image dataset; Based on the dense optical flow algorithm and according to the standardized image dataset, the optical flow field distribution of pixels in adjacent frame image data is calculated; the motion feature parameters of the UAV are extracted from the optical flow field distribution to form the motion feature dataset of the UAV. Based on the motion feature dataset and the actual distance between the UAV and the key protection area, calculate the threat level quantification value of all detected UAVs within the UAV's flight area; and generate a priority interception sequence based on the threat level quantification value. Based on the different levels of UAV targets in the priority interception sequence, a corresponding priority interception strategy is generated.

2. The method for priority interception of unmanned aerial vehicles based on dense optical flow algorithm according to claim 1, characterized in that, The process of preprocessing the flight image data includes: The images are denoised using an adaptive median filtering algorithm; the denoised flight image data is converted into grayscale images using a weighted average method, and the corresponding grayscale image size is standardized; the contrast of the grayscale images is enhanced based on a histogram equalization algorithm; and the preprocessed flight image data is recorded as a standardized image dataset.

3. The method for priority interception of unmanned aerial vehicles based on dense optical flow algorithm according to claim 2, characterized in that, The process of calculating the optical flow field distribution of pixels in adjacent frame image data based on the DeepFlow deep optical flow algorithm and the standardized image dataset includes: Based on the DeepFlow deep optical flow algorithm, a deep optical flow network model adapted to UAV scenarios was constructed and pre-trained. The real-time standardized image dataset is input into the pre-trained deep optical flow network model to obtain the corresponding initial optical flow field; The initial optical flow field is optimized to obtain the optimized optical flow field distribution.

4. The method for priority interception of unmanned aerial vehicles based on dense optical flow algorithm according to claim 3, characterized in that, The process of constructing a deep optical flow network model adapted to the drone scenario and optimizing the initial optical flow field includes: Set the number of layers in the multi-scale pyramid feature encoder and obtain the feature maps of the corresponding layers. Introduce a channel attention weight matrix into the feature map of each layer. A deep feature correlation matching layer is set up, and a feature correlation matching cost body is constructed based on the feature map output by the multi-scale pyramid feature encoder. Based on the feature correlation matching cost body, the matching information is fused and encoded through a convolutional layer to obtain the initial matching features, and then the corresponding initial sparse optical flow field is obtained. The upsampling iterative optical flow refinement decoder is configured to perform layer-by-layer upsampling on the multi-scale pyramid feature encoder, and combined with residual optical flow compensation, the initial sparse optical flow field is optimized into an initial dense optical flow field. A spatiotemporal consistency constraint layer for optical flow is set up, and the initial dense optical flow field is corrected by this constraint, thereby outputting an optimized optical flow field distribution; A combined loss function is constructed by using the endpoint error loss function and the smoothness constraint loss function. This function is used when the loss is continuous. When the combined loss function value of each training batch meets the requirements, the pre-trained weight parameters are saved.

5. The method for priority interception of unmanned aerial vehicles based on dense optical flow algorithm according to claim 4, characterized in that, The process of extracting motion feature parameters of the UAV from the optical flow field distribution to form a motion feature dataset of the UAV includes: Pixel-level motion vector sets of UAV target regions are extracted based on optical flow field distribution; The instantaneous motion feature parameters of the UAV are calculated based on the pixel-level motion vector set; the instantaneous motion feature parameters include instantaneous motion speed, instantaneous motion direction, and motion trajectory curvature; then, a spatiotemporal UAV motion feature dataset is constructed and standardized.

6. The method for priority interception of unmanned aerial vehicles based on dense optical flow algorithm according to claim 5, characterized in that, The process of extracting pixel-level motion vector sets of UAV target regions based on optical flow field distribution includes: The pixel region of the UAV target is selected from adjacent frame images using a target detection algorithm; the horizontal and vertical motion vectors corresponding to all pixels in the target region are extracted from the optical flow field distribution to construct a pixel-level motion vector set of the UAV target. For noise vectors existing in the target area, a statistical filtering method is used to remove outliers. The mean and standard deviation of each component in the pixel-level motion vector set are calculated, and a preset filtering threshold is set. Based on the mean, standard deviation and filtering threshold, the pixel-level motion vector set of the target area after denoising is obtained.

7. The method for priority interception of unmanned aerial vehicles based on dense optical flow algorithm according to claim 6, characterized in that, The process of calculating the instantaneous motion characteristic parameters of the UAV based on the pixel-level motion vector set includes: Using the acquisition time interval between adjacent frames as the time reference, the instantaneous motion characteristic parameters of the UAV are calculated based on the denoised pixel-level motion vector set; including instantaneous motion speed, instantaneous motion direction, instantaneous acceleration, and motion trajectory curvature.

8. The method for priority interception of unmanned aerial vehicles based on dense optical flow algorithm according to claim 7, characterized in that, The process of calculating the threat level quantification value of all detected drones within the drone's flight area, based on the aforementioned motion feature dataset and the actual distance between the drone and the key protection area, includes: By combining the motion characteristic parameters of UAVs and the actual distance between UAVs and key protection areas, a UAV threat level assessment index system is constructed. The UAV threat level assessment index system includes distance index, speed index, acceleration index, azimuth angle index, and trajectory curvature index. The weights of each core indicator were determined using the analytic hierarchy process (AHP), and the threat level of the drones was calculated using the fuzzy comprehensive evaluation method. Based on the calculated threat level values, the drones were classified into four threat levels. All detected drones within the monitoring area are sorted from highest to lowest threat level, and drones of the same threat level are sorted from closest to furthest from the key protection area to generate a priority interception sequence.

9. A drone priority interception system based on a dense optical flow algorithm, implementing the drone priority interception method based on a dense optical flow algorithm as described in any one of claims 1 to 8, characterized in that, include: The system includes a data acquisition module, a data preprocessing module, a feature analysis module, a threat level quantification module, and an intelligent strategy module. The data acquisition module is used to collect flight image data of the drone within the interception area; The data preprocessing module is used to preprocess flight image data to obtain the corresponding standardized image dataset. The feature analysis module, based on the dense optical flow algorithm and according to the standardized image dataset, calculates the optical flow field distribution of pixels in adjacent frame image data; extracts the motion feature parameters of the UAV from the optical flow field distribution to form the motion feature dataset of the UAV; The threat level quantification module is used to calculate the threat level quantification value of all detected drones within the drone's flight area based on the motion feature dataset and the actual distance between the drone and the key protection area. Based on the threat level quantification value, a priority interception sequence is generated; The intelligent strategy module is used to generate corresponding priority interception strategies based on different levels of UAV targets in the priority interception sequence.

Images