Target monitoring method based on visible range analysis of mass monitoring equipment in three-dimensional scene

Abstract

The invention belongs to the technical field of monitoring equipment visualization, and relates to a target monitoring method based on visible range analysis of mass monitoring equipment in a three-dimensional scene, which comprises the following steps: acquiring a three-dimensional map and position information of each monitoring equipment; mapping the monitoring equipment to a three-dimensional map according to the position information of the monitoring equipment; relevant parameters of each monitoring device are collected, a visual area of each monitoring device is drawn, and a view blind area of each monitoring device is constructed; each monitored visual area and each visual blind area are marked; target point position information is obtained, visual information of a user is determined, and a dragging track of the user on the three-dimensional map is predicted according to the visual information of the user; loading a visual area of the monitoring equipment in the target point position area to the user according to the dragging track, and performing real-time monitoring; according to the technical scheme, the visual field and the visual field blind area of the mass monitoring equipment are constructed, so that management personnel can be effectively helped to understand and master the monitoring range of the mass monitoring equipment.

Description

Technical Field

[0001] The present invention belongs to the technical field of monitoring equipment visualization, and in particular relates to a target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene. Background Art

[0002] The field of view (FOV) of surveillance equipment refers to the three-dimensional spatial range within which a camera, based on optical imaging principles, can effectively capture clear images. Defining parameters within this three-dimensional range involves multiple technical dimensions: first, lens optical parameters (including focal length, aperture, and distortion coefficient) directly influence the width of viewing angle; second, the physical characteristics of the image sensor (size and pixel density) determine spatial resolution accuracy; and third, device installation parameters (pitch, azimuth, and height above the ground) collectively form a spatial coordinate reference. Accurate FOV modeling is crucial for optimizing device deployment and improving security effectiveness.

[0003] The visual domain helps personnel understand the monitoring range of surveillance equipment and is crucial for its management. The following technical bottlenecks currently exist in the field of view management of large-scale video surveillance systems: 1. When the system integrates more than a thousand monitoring points, the 3D visualization platform must render massive spatial topological relationships in real time, which overloads the graphics processor and significantly increases interface response latency. This also creates visual information overload, impacting the operator's situational awareness. 2. Existing algorithms lack dynamic occlusion models for environmental factors such as building structures and vegetation growth, making it impossible for the system to automatically identify and warn of monitoring blind spots caused by scene changes, resulting in a security vulnerability risk of approximately 12-15%. Summary of the Invention

[0004] In order to solve the problems existing in the above-mentioned prior art, the present invention proposes a target monitoring method based on the visual field analysis of massive monitoring devices in a three-dimensional scene, the method comprising: obtaining a three-dimensional map and the position information of each monitoring device; mapping the monitoring device to the three-dimensional map according to the position information of the monitoring device; collecting relevant parameters of each monitoring device, and drawing the visualization area of ​​each monitoring device according to the relevant parameters; screening out obstructions in the visualization area of ​​each monitoring device, and constructing the blind spot of the field of view of each monitoring device; marking the visualization area and blind spot of the field of view of each monitoring device; obtaining the position information of the target point, determining the visual information of the user according to the target point position information, and predicting the dragging trajectory of the user on the three-dimensional map according to the visual information of the user; loading the visualization area of ​​the monitoring device within the target point position area to the user according to the dragging trajectory of the user; and monitoring the target in real time according to the visualization area.

[0005] Beneficial effects of the present invention:

[0006] This technical solution can effectively help managers understand and control the monitoring range of massive monitoring equipment by constructing the visual domain and blind spots of massive monitoring equipment. At the same time, relying on tile data, it can realize dynamic loading of visual domain data, increasing the loading speed and running smoothness of three-dimensional maps. In addition, it can also rely on the visual domain and blind spots for dynamic monitoring. It is rich in functions and has strong usability. This application also provides an improved 3D-Unet network structure. By adding an adaptive fusion module to each convolutional layer, the extracted features are enhanced, thereby more accurately predicting the user's drag trajectory and improving the model's response efficiency. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 This is a schematic diagram of the blind area of ​​the visual field of the present invention;

[0008] Figure 2 Each point map of the present invention;

[0009] Figure 3 It is the overall flow chart of the present invention;

[0010] Figure 4 This is the self-attention fusion network structure diagram of the present invention;

[0011] Figure 5 This is the improved 3D-Unet network structure diagram of the present invention. DETAILED DESCRIPTION

[0012] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0013] A target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene, such as Figure 3 As shown, the method includes: obtaining a three-dimensional map and position information of each monitoring device; mapping the monitoring device to the three-dimensional map according to the position information of the monitoring device; collecting relevant parameters of each monitoring device, and drawing the visualization area of ​​each monitoring device according to the relevant parameters; screening out obstructions in the visualization area of ​​each monitoring device, and constructing a blind spot of vision of each monitoring device; marking the visualization area and blind spot of vision of each monitoring device; obtaining target point position information, determining the user's visual information according to the target point position information, and predicting the user's dragging trajectory on the three-dimensional map according to the user's visual information; loading the visualization area of ​​the monitoring device within the target point position area to the user according to the user's dragging trajectory; and monitoring the target in real time according to the visualization area.

[0014] In this embodiment, the map reconstruction process includes: first, preparing three-dimensional map data (including high-precision terrain, high-precision tilt model and high-precision image data) and relevant parameters of the monitoring equipment (including the installation position of the monitoring equipment, the installation height of the monitoring equipment, the orientation of the monitoring equipment, the focal length of the monitoring equipment, the viewing angle of the monitoring equipment, the opening and closing angle and the pitch angle of the monitoring equipment); mapping the position of the monitoring equipment to the three-dimensional map data, and starting to create a visual area for each monitoring device.

[0015] Creating a visual domain for each monitoring device includes: taking the location of the monitoring device as the origin, drawing the coverage area of ​​the monitoring device with a fixed viewing angle, for example, if the fixed viewing angle is 75°, then drawing a coverage area of ​​a 75° viewing angle quadrangular pyramid; increasing the opening and closing angles and pitch angles brought by the monitoring device bracket, for example, if the opening and closing angles are 25° and the pitch angles are 15°, then a visual domain of a quadrangular pyramid with a left-right angle of 100° and a top-bottom angle of 90° is formed; filtering out obstructions in the visual domain, and using the projection surface of the obstructions to construct the blind area of ​​the monitoring device. The blind area of ​​the visual domain is as follows: Figure 1 shown.

[0016] In this embodiment, the specific process of drawing the visualization area of ​​each monitoring device according to relevant parameters includes: determining the basic parameters of the monitoring device, taking the mainstream Hikvision as an example. Its main parameters include the following:

[0017] (1) Location information: The coordinates (x, y, z) of the monitoring device in three-dimensional space, usually determined by the installation location.

[0018] (2) Visual parameters: horizontal viewing angle (FOV Horizontal): for example, 75°; vertical viewing angle (FOV Vertical): for example, 45°.

[0019] (3) Direction parameters: Pan Angle: The horizontal rotation angle of the monitoring device, for example, 25°; Tilt Angle: The vertical tilt angle of the monitoring device, for example, 15°.

[0020] (4) Monitoring distance: The maximum visual distance of the monitoring device, for example 50 meters.

[0021] (5) Resolution: The image resolution of the monitoring device affects the accuracy of the visual field.

[0022] In this embodiment, the geometric model of the viewing area pyramid is constructed as follows: the viewing area pyramid is a quadrangular pyramid, the vertex of which is the location of the monitoring device, and the bottom surface is the area covered by the monitoring device. The specific construction process is as follows:

[0023] Step 1: Determine the vertex of the pyramid, that is, take the position (x, y, z) of the monitoring device as the vertex of the pyramid.

[0024] Step 2: Determine the base of the pyramid. That is, calculate the four corner points of the pyramid base based on the horizontal field of view (FOV Horizontal) and the vertical field of view (FOV Vertical). Assuming the monitoring device is oriented in the initial direction (usually the positive direction of the Z axis), the four corner points of the base can be calculated using the following formula:

[0025] Horizontal range: [-FOV_Horizontal / 2,FOV_Horizontal / 2]

[0026] Vertical range: [-FOV_Vertical / 2,FOV_Vertical / 2]

[0027] The four corner points of the bottom surface are:

[0028] Upper left corner: (x+d*tan(FOV_H / 2),y+d*tan(FOV_V / 2),z+d)

[0029] Upper right corner: (xd*tan(FOV_H / 2),y+d*tan(FOV_V / 2),z+d)

[0030] Lower left corner: (x+d*tan(FOV_H / 2),yd*tan(FOV_V / 2),z+d)

[0031] Lower right corner: (xd*tan(FOV_H / 2),yd*tan(FOV_V / 2),z+d)

[0032] Where d is the monitoring distance.

[0033] Step 3: Adjust the direction of the pyramid; specifically, rotate the pyramid according to the pan angle and tilt angle. Use a three-dimensional rotation matrix to transform the bottom corners of the pyramid.

[0034] Horizontal rotation matrix (rotation around the Y axis, opening and closing angle):

[0035] [cos(θ),0,sin(θ)]

[0036] [0,1,0]

[0037] [-sin(θ),0,cos(θ)]

[0038] Vertical rotation matrix (rotation around the X axis, pitch angle):

[0039] [1,0,0]

[0040] [0,cos(φ),-sin(φ)]

[0041] [0,sin(φ),cos(φ)]

[0042] Among them, θ is the opening angle, and φ is the pitch angle.

[0043] Multiply the bottom corner points by the rotation matrix in sequence to obtain the adjusted pyramid bottom surface.

[0044] The blind spot of a single monitoring device is in the shape of a polygonal pyramid. For example, if the interface of the obstruction is a rectangle, then the blind spot is a quadrangular pyramid. The method for determining the blind spot is to determine the obstruction and the projection surface (and imaging surface) of the obstruction, wherein the line connecting the centroid of the projection surface of the obstruction and the monitoring device is perpendicular to the projection surface of the obstruction. Generally, objects such as terrain or buildings, which have an imaging area that occupies a larger than a preset threshold in the photo area, are considered obstacles. For example, if the obstacle is a small hill, its area in the imaging occupies more than 5% of the total area of ​​the photo. The contour is extracted using a variety of contour extraction operators, such as the Canny and Sobel operators, and the pixels inside the contour are used as the imaging area of ​​the obstacle. The total area of ​​the photo is a fixed value, such as 2 million pixels.

[0045] Sampling is performed on the contour of the obstruction's projection surface to obtain multiple sampling points. Sampling points can be sampled at a fixed angle, meaning the angle formed by the line connecting any two adjacent sampling points and the center of mass is a fixed value, such as 5°. This means that 72 sampling points are extracted from the contour. Lines are drawn connecting the monitoring device's location and the sampling points of the obstruction's projection surface, extending the line to the farthest imaging surface to form a blind spot. The maximum imaging distance of a typical camera is between 100 and 1000 meters, but the actual distance must be determined based on the specific situation.

[0046] In this embodiment, sampling the obstacle contour includes: based on existing contour extraction algorithms (such as Canny and Sobel), this article introduces an edge detection algorithm based on phase congruency and adds other influencing factors to improve the accuracy of contour sampling. The principle of phase congruency is to analyze the frequency domain information of the image and detect the phase-consistent areas in the image to extract the edges. The specific steps are as follows: input the projection surface image of the occluder; perform phase congruency detection on the projection surface image of the occluder, that is, calculate the phase congruency map and extract the occluder contour. After extracting the obstacle contour, the contour is sampled to determine the blind spot of the field of view.

[0047] Determining the blind area of ​​the field of view includes: taking the center of mass of the contour as the reference point, sampling the contour at fixed angle intervals. Specifically, let the center of mass of the contour be (xc ,y c ), the number of sampling points is N, the fixed angle interval is Δθ=360° / N; for the i-th sampling point, its angle is θ i =i*Δθ, the coordinates of the sampling points are:

[0048] x i =x c +r i *cos(θ i )

[0049] y i =y c +r i *sin(θ i )

[0050] Among them, r i is the distance from the i-th point on the contour to the centroid.

[0051] In this embodiment, the contour is sampled adaptively, dynamically adjusting the sampling density based on the contour's curvature. Areas with greater curvature have denser sampling points. Specifically, the curvature κ of each point on the contour is calculated. Sampling points are then allocated based on the curvature, with the interval Δθ between sampling points decreasing in areas with greater curvature.

[0052] Constructing the blind zone includes: after sampling is completed, connecting the position of the monitoring device with each sampling point and extending it to the farthest imaging surface to form the blind zone. Specifically, it includes: assuming the position of the monitoring device is (x0, y0, z0); for each sampling point (x i ,y i ) and calculate the equation of the line connecting it to the monitoring device. Extend the line to the farthest imaging plane (e.g., 1000 meters) to obtain the boundary points of the blind spot. Connect all boundary points to form a multi-sided pyramid-shaped blind spot. This method can further improve the accuracy of contour detection and sampling, and more accurately determine the blind spot.

[0053] In this embodiment, monitoring and management based on visual zones and blind spots includes marking the visual zones and blind spots of all monitoring devices, which can be marked with different colors. Overlapping visual zones can be marked with a different color. For any blind spot of a monitoring device, the portion that overlaps with the visual zones of other monitoring devices is marked as the blind spot of that monitoring device. Any portion of the blind spot that does not overlap with any other visual zones is marked as a system blind spot.

[0054] For example, visible or overlapping visual areas are marked in green, blind areas of any monitoring device are marked with a light red plus a number (device number), and system blind areas are marked in dark red. If a monitoring device is damaged during use, causing its blind area to become a system blind area, the newly added system blind area mark will be sent to the administrator. This process helps administrators intuitively understand the monitoring ranges and blind areas of a large number of monitoring devices, as well as changes in monitoring ranges during use.

[0055] In addition, if the visual range of monitoring device A just covers monitoring device B, the two will be paired. If there is a blind spot of monitoring device A during the monitoring process, its paired device can be quickly called to perform blind spot monitoring.

[0056] In this embodiment, visual domain loading management includes creating and marking visual domains for a large number of monitoring devices. Therefore, when loading map information, if all visual domains are loaded at once, it will cause the web page to load slowly or freeze. Therefore, this solution adopts the following loading management solution:

[0057] The high-precision terrain, high-precision tilt model and high-precision image data in this solution are stored in the server in the form of tile data. The server will load the tile data within the user's field of view into the web page.

[0058] In this solution, tile data is stored in the server (to facilitate third-party calls to the API interface, it is only stored in the server), and the viewport data can be stored in the server or locally.

[0059] The visual area data is associated with the tile data in advance. Specifically, if the coordinates of the monitoring device are within the range of the tile data, they are associated.

[0060] In this embodiment, in order to improve the user experience, this solution also provides a pre-loading solution for trajectory prediction, which specifically includes: predicting the drag trajectory based on the current visual range and the visual range before the previous n screen drags. Specifically: taking the five visual ranges V1, V2, V3, V4 and V5 as examples, extract the center points of V1, V2, V3, V4 and V5, which are C1, C2, C3, C4 and C5 respectively; connect the first center point C1 and the last center point C5 to obtain the displacement reference line; calculate the shortest straight-line distance S2, S3, S4 between C2, C3, C4 and the displacement reference line, and the positions of each point are as follows: Figure 2 shown.

[0061] In this embodiment, the average value A and variance σ of S2, S3, and S4 are calculated:

[0062]

[0063] Among them, A is the average value of the shortest straight line distance, n is the number of points, S i is the shortest straight-line distance of the i-th point, and σ is the variance of the shortest straight-line distance.

[0064] The variance σ reflects the uniformity of the shortest straight-line distances S2, S3, and S4, while the average reflects the distribution characteristics of the shortest straight-line distances S2, S3, and S4. Therefore, if the average value A is smaller and the variance is smaller, it means that C2, C3, and C4 are more closely aligned with the displacement reference line, and the overall trend of these five points is in the same direction. In this case, a drag trend is determined, and preloading is performed based on the displacement reference line.

[0065] Specifically, when A < α and σ < β, α is the displacement threshold, for example, 200 pixels, and β is the variance threshold, for example, 50. A dragging trend is determined to exist. At this point, one or more sets of tile data and viewport data along the displacement reference line are loaded. When loading tile data, if the tile data stores associated viewport data, the viewport data is also loaded from the server or local storage.

[0066] In this embodiment, predicting the dragging trajectory of the user on the three-dimensional map includes: using an inertial sliding model to predict the dragging trajectory of the user, specifically including: data collection: recording the screen coordinates (x t ,y t ) and timestamp t. Calculate speed and direction:

[0067]

[0068] Assuming that after dragging stops, the map will continue to slide for a distance, which is determined by the initial velocity vt and the deceleration a. The sliding distance s is:

[0069]

[0070] Where a is the deceleration, usually an empirical value is taken (such as a=0.5px / m 2 ).

[0071] According to the sliding distance s and direction θ t , calculate the future coordinates (x t+1 ,y t+1 ), whose expression is:

[0072] x t+1 =x t +S*cosθ t

[0073] y t+1 =yt +S*sinθ t

[0074] According to the predicted future coordinates (x t+1 ,y t+1 ), loading tile data and viewshed data near the area. This physical model's prediction is suitable for scenarios with high real-time requirements, is computationally simple, and conforms to the user's natural operating habits.

[0075] Example 1

[0076] Predicting a user's drag trajectory on a 3D map primarily combines prior knowledge of physical motion with data-driven user behavior correction, ensuring real-time performance while improving personalized prediction accuracy. Its core stages include "inertial prediction" and "data-driven correction." The following is a detailed technical breakdown and specific formulas:

[0077] Inertia prediction includes: assuming that the user's dragging operation in the 3D map can be modeled as damped inertial motion, its motion equation is as follows:

[0078] Position pt=(xt,yt,zt)∈R3,

[0079] Velocity vt = (vx, t, vy, t, vz, t)

[0080] acceleration at=(ax,t,ay,t,az,t)

[0081] Equation of motion: Using explicit Euler integration, time step Δt:

[0082]

[0083] Wherein, μ is the damping coefficient (such as μ = 0.1), simulating friction.

[0084] The initial velocity v0 is calculated from the position difference between the last two operations of the user:

[0085]

[0086] The data-driven correction phase includes: using machine learning models to predict the residual of the physical model (the difference between the actual trajectory and the physical prediction); specifically, the actual next position and physical predictions Deviation:

[0087]

[0088] Select LSTM network as the prediction network; set the historical position window {p t-k ,…,p t Physical prediction Map context feature f t Input into the prediction network and output the prediction result; the formula for the prediction network to process the data is:

[0089] h t =LSTM(h t-1 ,[p t ;v t ;f t ])

[0090]

[0091] Among them, h t is the hidden state, W o is the output layer weight.

[0092] The output of the hybrid model is the sum of the physical prediction and the residual, which is expressed as:

[0093]

[0094] Example 2

[0095] In this embodiment, an improved 3D-Unet network structure is designed. The improvement method includes setting an adaptive connection layer in each convolution layer in the 3D-Unet network structure; specifically, it includes: 4 three-dimensional convolution layers, 4 maximum pooling layers, 4 three-dimensional deconvolution layers, 4 feature splicing layers and 3 adaptive layers; the trained improved 3D-Unet network structure is used to predict the user's drag trajectory.

[0096] like Figure 5 As shown, training the improved 3D-Unet network structure includes:

[0097] S101: Obtain a user dragging behavior dataset and preprocess the user data in the dataset, including de-redundancy, completion, and enhancement.

[0098] S102: Input the enhanced data into a first three-dimensional convolutional layer for feature extraction to obtain a first feature map;

[0099] S103: Input a feature map into a second three-dimensional convolutional layer, and the second three-dimensional convolutional layer outputs a second feature map;

[0100] S104: Input the second feature map into a third three-dimensional convolutional layer for feature extraction to obtain a third feature map;

[0101] S105: Input the second feature map into a fourth three-dimensional convolutional layer for feature extraction to obtain a fourth feature map;

[0102] S106, inputting the first feature map and the second feature map into the first adaptive layer to obtain a first fusion feature;

[0103] S107, inputting the second feature map and the third feature map into the second adaptive layer to obtain a second fusion feature;

[0104] S108, inputting the third feature map and the fourth feature map into the third adaptive layer to obtain a third fusion feature;

[0105] S109: Input the fourth feature map and the third fused feature into the first deconvolution layer for fusion processing to obtain a fused feature A;

[0106] S110, inputting the fused feature A and the second fused feature into the second deconvolution layer for fusion processing to obtain the fused feature B;

[0107] S111, inputting the fused feature B and the first fused feature into the third deconvolution layer for fusion to obtain a fused feature C;

[0108] S112, inputting the fused feature C and the first feature map into the fourth deconvolution layer for fusion to obtain a fused feature D;

[0109] S113, inputting the fusion feature D into the classification network to obtain the final prediction result;

[0110] S114: Construct a loss function for the model based on the prediction results, adjust the model parameters, and complete the model training when the loss function converges. The loss function of the model adopts a cross entropy loss function.

[0111] In this embodiment, the data is fused in the deconvolution layer using a self-attention fusion network, such as Figure 4 As shown in the self-attention module framework diagram, the feature map x of the hidden layer in the middle of the network is converted into two feature spaces f(x)=W f x and g(x) = W g x, used to calculate attention. Specifically including:

[0112] Step 1: Convert the feature map x of the hidden layer in the middle of the network into two feature spaces. The calculation formula is:

[0113]

[0114] Where f(x) represents the feature space obtained after feature extraction of the input image; W f represents the learning weight matrix, which is represented by a 1×1×1 convolution; g(x) represents the feature space obtained after feature extraction of the input image; W g Represents the learning weight matrix, represented by a 1×1×1 convolution;

[0115] Step 2: Calculate the attention in the feature space. The calculation formula is:

[0116]

[0117] Among them, β j,i represents the attention of voxel j to voxel i, and N represents the total number of voxels in the input image;

[0118] Step 3: Calculate the third feature space h(x) = W h x∈R C×(h×w×d) , where W h Indicates that R C×(h×w×d) express;

[0119] Step 4: Calculate the attention feature map O of the third feature space j , its calculation expression is:

[0120]

[0121] h(x i )=W h x,v(x i )=W v x

[0122] Where v represents β j,i After matrix multiplication with the third feature space h(x), information is extracted from the fused features; β j,i represents the attention of voxel j to voxel i; W v Represents the learning weight matrix, represented by a 1×1×1 convolution;

[0123] Step 5: The fusion feature map finally output by the attention calculation module based on the three feature spaces is expressed as:

[0124] y j =αO j +x j

[0125] Here, α is a learnable scalar initialized to 0.

[0126] In this embodiment, the adaptive layer fuses the feature maps including: performing a cascade operation on the first feature map and the second feature map, halving the number of channels of the cascaded features through a 1×1 convolution layer to generate a four-dimensional tensor feature map; inputting the four-dimensional tensor feature map into four strip pooling branches for pooling operations, wherein the four strip pooling operations include two average pooling operations and two maximum pooling operations, and obtaining four feature maps F s1 、F s2 、F s3 and F s4 ; Fs1 and F s2 Fusion is performed to obtain the feature map after average strip pooling; F s3 and F s4 The fusion is performed to obtain the feature map after maximum strip pooling; the feature map after average strip pooling and the feature map after maximum strip pooling are connected in series, and the concatenated feature map is convolved to obtain the feature weight map; the weight feature map and the four-dimensional tensor feature map are multiplied element by element to obtain the fused feature.

[0127] In this embodiment, monitoring assistance management includes: Based on the aforementioned visual domain, this solution can perform intelligent dynamic monitoring. It can be used for survey drone tracking or dynamic monitoring of other operational equipment. Specifically, it includes: obtaining the drone's 3D coordinate information, including 2D positioning information and altitude information; mapping the drone's 3D coordinate information to a 3D map in real time; and retrieval of the drone's monitoring information by locating the visual domain within which the drone's 3D coordinate information falls. If the drone enters the blind spot of the current monitoring device, its paired device is called upon to provide blind spot coverage. If a drone enters the system blind spot, the target monitoring devices corresponding to all adjacent visual areas in the system blind spot will be called to perform frame-by-frame analysis on the images in the video taken by the target monitoring devices, specifically including: uniformly sampling the images in the video taken by the target monitoring devices, sampling 3-5 frames of images per second, and converting them into grayscale images; taking the difference between adjacent grayscale images, that is, taking the difference between the grayscale values ​​of the pixels at the same position to obtain the difference of each pixel, and summing all the differences to obtain the total grayscale difference between the two frames of images; when the total grayscale difference is greater than a preset value (determined by pre-comparing the total grayscale difference of pixels in the same image with and without a drone), it is determined that the drone appears in the next frame, and the corresponding monitoring device is called to resume monitoring.

[0128] The above embodiments further illustrate the purpose, technical solutions and advantages of the present invention in detail. It should be understood that the above embodiments are only preferred implementation plans of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made to the present invention within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene, characterized in that: include: Obtain three-dimensional maps and location information of each monitoring device; Mapping the monitoring device onto a three-dimensional map according to the location information of the monitoring device; Collect relevant parameters of each monitoring device and draw the visualization area of ​​each monitoring device according to the relevant parameters; Screen out obstructions within the visual area of ​​each monitoring device and construct blind spots for each monitoring device; Mark the visual areas and blind spots of each monitoring area; Obtaining target point location information, determining the user's visual information based on the target point location information, and predicting the user's dragging trajectory on the three-dimensional map based on the user's visual information; According to the dragging trajectory of the user, a visualization area of ​​the monitoring device in the target point location area is loaded to the user; and the target is monitored in real time according to the visualization area.

2. The target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene according to claim 1 is characterized in that: The three-dimensional map includes high-precision terrain, high-precision tilt model and high-precision image data; the relevant parameters of the monitoring equipment include the installation height of the monitoring equipment, the orientation of the monitoring equipment, the focal length of the monitoring equipment, the viewing angle of the monitoring equipment, the opening and closing angle and the pitch angle of the monitoring equipment.

3. The target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene according to claim 1 is characterized in that: Drawing the visualization area of ​​each monitoring device according to relevant parameters includes: taking the position of the monitoring device as the origin, drawing the coverage area of ​​the monitoring device with a fixed viewing angle; adding the opening and closing angles and pitch angles brought by the monitoring device bracket, adjusting the coverage area based on the opening and closing angles and pitch angles, and obtaining the visual domain of the quadrangular pyramid; screening out obstructions in the visual domain, and using the projection surface of the obstructions to construct the blind spot of the monitoring device.

4. The target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene according to claim 3 is characterized in that: Constructing a blind spot in the field of view of a monitoring device includes: determining an obstruction and determining a projection surface of the obstruction, wherein a line connecting the centroid of the projection surface of the obstruction and the monitoring device is perpendicular to the projection surface of the obstruction; sampling is performed on the contour of the projection surface of the obstruction to obtain multiple sampling points; connecting the location of the monitoring device with the sampling points of the contour of the projection surface of the obstruction, and extending the line to the farthest imaging surface to form a blind spot in the field of view.

5. The target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene according to claim 1 is characterized in that: Marking the visualization area and blind spot of each monitoring device includes: marking the overlapping visualization area with the target color; marking the overlapping part of the blind spot of any monitoring device and the visual field of other monitoring devices as the blind spot of the monitoring device; marking the part of any monitoring device that does not overlap with the visual field of other monitoring devices as the system blind spot.

6. The target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene according to claim 1 is characterized in that: Predicting the dragging trajectory of a user on a three-dimensional map includes: obtaining the user's current visual range and the visual range of the previous n screen drags; extracting the center point of the visual range, connecting the first center point C1 and the last center point Cn to obtain a displacement reference line; calculating the shortest straight-line distance between other center points and the displacement reference line; calculating the average and variance of the shortest straight-line distances of each point; determining the trajectory of the user's screen dragging based on the average and variance; and loading a visualization area for the user based on the screen dragging trajectory.

7. The target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene according to claim 6, characterized in that: The formula for calculating the mean and variance of the shortest straight-line distance of each point is: Among them, A is the average value of the shortest straight line distance, n is the number of points, S i is the shortest straight-line distance of the i-th point, and σ is the variance of the shortest straight-line distance.

8. The target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene according to claim 1 is characterized in that: Real-time monitoring of the target based on the visualization area includes: obtaining the three-dimensional coordinate information of the target; mapping the three-dimensional coordinate information of the target to a three-dimensional map in real time; calling the visual field of the target's three-dimensional coordinates to obtain the target monitoring information; if the target drives into the blind spot of the current monitoring device, then calling its paired device to cover the blind spot; if the target drives into the system blind spot, then calling the monitoring devices corresponding to all adjacent visual fields within the system blind spot range, analyzing the corresponding monitoring device video frame by frame, obtaining the target information, and completing real-time monitoring.

9. The target monitoring method based on visual field analysis of massive monitoring equipment in a three-dimensional scene according to claim 8, characterized in that: Frame-by-frame analysis of the corresponding monitoring device video includes: uniformly sampling the video and converting the sampled image into a grayscale image; taking the difference between adjacent grayscale images to obtain the difference between each pixel; summing all the differences to obtain the total grayscale difference between the two frames of images; setting a threshold. When the total grayscale difference is greater than the threshold, it is determined that the target appears in the next frame and the corresponding monitoring device is called to monitor the target; otherwise, the adjacent images are compared again.

Images