A video monitoring field of view analysis method based on an improved PDERL algorithm

Abstract

This invention discloses a video surveillance view analysis method based on PDERL, relating to the field of GIS spatial analysis. The method includes: acquiring DSM terrain data and video surveillance parameter information; utilizing a partitioning strategy based on video surveillance field of view, viewing distance, and other information to ensure that distant grid points on the same line of sight can utilize nearby grid point information within any field of view; optimizing the initial reference line according to whether the observation point is located at a grid point using different strategies; for each analysis sub-region, determining the start and end coordinates of the horizontal / vertical grid lines of the analysis region based on the dividing line formed by the field of view, and simultaneously judging the relationship between the starting dividing line and the boundary region values ​​to optimize the grid point search; if multiple regions exist, merging the visual results of each region to generate a corresponding raster visibility file, and visualizing the results in GIS. This invention achieves an effective balance between speed and accuracy in the field of view analysis, and has significant reference value for view analysis scenarios requiring rapid response and high accuracy.

Description

Technical Field

[0001] This invention relates to the field of GIS spatial analysis, and more particularly to a video surveillance view analysis method based on an improved PDERL algorithm. It is primarily applicable to view analysis in video surveillance and enables visualization. Background Technology

[0002] With rapid social and economic development, public safety has received increasing attention. Video surveillance, as an extension of human vision, is a crucial technological means to ensure urban safety and is widely used in densely populated public places. However, due to complex and varied terrain, relying solely on parameters such as camera viewing distance and field of view is insufficient to comprehensively assess the visible area; a comprehensive analysis considering factors such as actual terrain, landforms, and building structures is necessary. Field of view analysis, by systematically examining the scope of the field of view to achieve a comprehensive assessment of a specific area, has wide applications in various fields, including visual quality assessment, territorial planning, site selection, spatial layout, coverage assessment, and archaeological site protection.

[0003] In recent years, with increasingly refined spatial resolution and elevation values ​​in DEMs, traditional view domain analysis algorithms have struggled to fully utilize DEM data, and their inherent errors have become more pronounced. In most scenarios, it is crucial to improve computational efficiency while maintaining algorithm accuracy. This is of great significance for the widespread application and practical performance improvement of view domain analysis algorithms. The PDERL algorithm is a fast approximate view domain algorithm that proposes a PDE spatial reference frame. By constructing and updating reference lines, it assesses the visibility of survey lines, achieving a good balance between accuracy and speed. The XPDERL algorithm, by altering the partitioning strategy, allows observation points to utilize grid line information in each region to the maximum extent. Experiments demonstrate that the XPDERL algorithm, while sacrificing some accuracy, can improve its efficiency to a level close to that of approximation algorithms.

[0004] The aforementioned inventions all treat the visible field of view as a circular field of view. However, video surveillance involves sector-shaped field of view analysis. Besides the latitude and longitude coordinates, viewing distance, and elevation of the video surveillance system, the field of view angle is also needed to determine the analysis range. Existing algorithms are not suitable for video surveillance field of view analysis. To address this issue, this invention proposes a fast approximate field of view analysis algorithm, SPDERL, based on the PDERL algorithm and the XPDERL partitioning strategy. This ensures its practicality and effectiveness in video surveillance field of view analysis, providing valuable reference for field of view analysis scenarios requiring rapid response and high accuracy. Summary of the Invention

[0005] The purpose of this invention is to provide a video surveillance view analysis method based on an improved PDERL algorithm, so as to achieve fast approximate view analysis of video surveillance.

[0006] To achieve the above objectives, the present invention provides the following steps:

[0007] (1) Obtain DSM data and video monitoring parameters under three terrain conditions, and establish a two-dimensional array of visible range based on the parameters;

[0008] (2) Determine the zoning strategy based on the field of view and the visible area;

[0009] (3) Optimize the initial reference line by selecting different strategies based on the location of the observation point;

[0010] (4) Determine the start and end boundary coordinates within the visible range based on the dividing line equation and the grid coordinates of the video monitoring points;

[0011] (5) Optimization of grid point search within the visible area;

[0012] (6) Merge the two-dimensional visual arrays of each sub-region and visualize the results.

[0013] Furthermore, step (1) specifically includes the following steps:

[0014] Read the elevation values ​​from the Digital Surface Model (DSM) and remove outliers, storing them in a two-dimensional array H. Then, based on the DSM data geographic transformation parameters (top-left x-coordinate, horizontal resolution, rotation parameter, top-left y-coordinate, rotation parameter, vertical resolution), uniformly store the DSM elevation array with the bottom-left corner as the origin. The starting latitude and longitude of the terrain file is (lon_s, lat_s), and the grid spacing is d. x d y ;

[0015] Read parameters from the monitoring attribute data table, including line-of-sight distance r and the latitude and longitude of the monitoring point (lon). c ,lat c ), scan start angle α, scan end angle β, and elevation value H a The video surveillance latitude and longitude coordinates are combined with the geographic transformation parameters of the DSM file to convert them into grid point coordinate values ​​P. c (x c ,y c );

[0016] Based on the monitoring attribute parameters, a fan-shaped region centered on the monitoring point and with a line-of-sight distance r as its radius was first determined. The angular range of this region was determined by the scanning start angle α and the scanning end angle β. Two boundary dividing lines, L1 and L2, correspond to the areas passing through the monitoring point P, respectively. c And the slopes are straight lines with tanα and tanβ, as shown in equations (1) and (2):

[0017] L1:y1=tanα*(xxc )+y c (1)

[0018] L2:y2=tanβ*(xx c )+y c (2)

[0019] Furthermore, step (2) specifically includes the following steps:

[0020] Specific zoning strategies are used to divide different regions, with a minimum angle of 0 degrees representing true north, increasing clockwise to a maximum of 360 degrees. Angle ranges are calculated based on the input start and end angles, categorized into four main types: [0, 90), [90, 180), [180, 270), [270, 360), and [0, 360]. These four types are further subdivided, as shown in Table 1.

[0021] Table 1 Specific Partitioning Strategies

[0022]

[0023] Furthermore, step (3) specifically includes the following steps:

[0024] If the observation point is located on a grid point, then equations (3) and (4) are satisfied:

[0025] (x c -lon_s)%d x =0(3)

[0026] (y c -lat_s)%d y =0(4)

[0027] Then construct the grid points (P0,…,P) of the observation point and its row / column. c-1 ,P c ,P c+1 ,…,P c+k ,…P n The line of sight is used to calculate the elevation H at the intersection of each line of sight and the grid using linear interpolation. ci , H ci Compared with the actual elevation value H of each grid point ai Compare and judge its visibility; if P c To P i A line of sight is constructed if and only if on the same line of sight (P0,…,P…) i-1 All can be seen and H ci ≥H ai Then P iVisible if not visible otherwise; since these grid points are in the same row / column as the observation point and will not obscure subsequent rows / columns, the next row / column is considered visible and an initial reference line RL is constructed; when the observation point is not on a grid point, the visibility of the nearest row / column grid point to the observation point is accurately calculated through linear interpolation, and an initial reference line is constructed.

[0028] The reference line construction first considers the first row / column of the grid points in the observation point analysis as visible, and converts the x-coordinate, y-coordinate, and elevation h of each grid point in the geocentric coordinate system into the proximity p, direction d, and elevation e in PDE space; the conversion formulas are as shown in equations (5), (6), and (7):

[0029]

[0030]

[0031]

[0032] Adjacent grid points are transformed into continuous polylines in PDE space. The PDERL algorithm retains the slope 'a' and the final 'd' value of each polyline to store information about nearby grid points. LR records this information, and LC represents the current row / column of the analysis. LC visibility is determined by finding the 'd' value interval [d0,…d] within LR for each grid point. c-1 ,d c ,d c+1 ,d n The corresponding broken line is used, and the value of e is calculated based on a and d. This e is then... c The value corresponding to the largest e in relation to d is max The values ​​are compared to determine visibility, and then updated. For example, symbol (8) represents visibility, and vice versa:

[0033] e c -e max ≥0(8)

[0034] Furthermore, step (4) specifically includes the following steps:

[0035] First, based on the grid coordinates of the first column / row analyzed from the video surveillance points, and combined with the equation of the dividing line, determine the starting and ending coordinates. Second, using the grid diagonal and the dividing line, calculate the adjustment value required for moving one row / column outward relative to the previous row / column; the initial adjustment value is d. s The termination adjustment value is d e If the coordinates of each column / row change by a fixed amount of 1 and L1 and L2 are both straight lines, then the change value of y is as shown in equation (9):

[0036] Δy=kΔx(9)

[0037] Where k represents the slope of the dividing line equation;

[0038] Therefore, d s With d e The calculation formulas are shown in equations (10) and (11):

[0039] d s =tanβ(10)

[0040] d e =tanα(11)

[0041] When the dividing line of the monitoring point is perfectly aligned with the diagonal of the grid, no adjustment value needs to be modified. If the dividing line shows an increasing trend, its adjustment value should be rounded up, as shown in equation (12).

[0042]

[0043] Among them ys i The y-coordinate of the starting grid for the current analysis column is represented by ys. i-1 Represents the y-coordinate of the starting grid in the previous column;

[0044] For dividing lines that show a decreasing trend, their adjustment values ​​should be rounded down, as shown in equation (13):

[0045]

[0046] Among them ye i The y-coordinate of the current analysis column's ending grid is represented by ye. i-1 This represents the y-coordinate of the terminating grid in the previous column.

[0047] Furthermore, step (5) specifically includes the following steps:

[0048] To ensure that the coordinates of rows or columns after adjustment do not exceed the boundaries of the visual analysis area, this invention introduces a callback strategy. When the coordinates of a row or column exceed the visual analysis area due to adjustment, the coordinates are adjusted to the boundary value of the visual analysis area. This strategy ensures that the analysis area represented by the final column or row is not lost due to the adjustment value.

[0049] The grid position of the dividing line in the starting direction is compared with the boundary region. If the starting grid position exceeds the boundary region, the analysis is terminated. This reduces invalid searches of non-target regions and improves analysis efficiency.

[0050] Furthermore, step (6) specifically includes the following steps:

[0051] If multiple regions exist, the visibility results of each region are merged, and the visibility of the region is ultimately maintained through a two-dimensional visibility array. Based on this visibility array, a raster visibility file is created, and video monitoring points are saved as shapefiles with attributes such as latitude and longitude, line of sight, start angle, end angle, and viewpoint height. The results are then visualized in GIS.

[0052] Compared with the prior art, the beneficial effects of the present invention are:

[0053] This invention proposes a partitioning strategy to enable visibility analysis of the visible area formed by any field of view angle using the SPDERL algorithm. It also addresses issues such as misjudgment of edge point visibility and search redundancy caused by partitioning strategies. Furthermore, it optimizes the initial reference line construction to avoid significant errors when the observation point is located at a grid point. Finally, it combines the physical characteristics of video surveillance field of view angles with the SPDERL algorithm through a partitioning strategy to achieve fast approximate view analysis for video surveillance. This algorithm achieves an effective balance between speed and accuracy in the field of view analysis for devices such as video surveillance whose analysis range dynamically changes according to physical characteristics. It also provides important reference value for view analysis scenarios requiring rapid response and high accuracy. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart illustrating the overall technical process of the method of the present invention.

[0056] Figure 2 This is a 3D topographical diagram of the DEM.

[0057] Figure 3 This is a schematic diagram of the video surveillance analysis area;

[0058] Figure 4 This is a diagram illustrating the bottom right partitioning strategy;

[0059] Figure 5 This is a diagram illustrating the right-side partitioning strategy and the left and right-side partitioning strategies.

[0060] Figure 6 A schematic diagram for optimizing the initial reference line;

[0061] Figure 7 This is a schematic diagram of transforming from the geocentric coordinate system to the PDE coordinate system;

[0062] Figure 8This is a diagram illustrating the start and end boundary indices and extended values;

[0063] Figure 9 This is a schematic diagram of the start and end coordinate callback strategy;

[0064] Figure 10 A diagram illustrating the optimization of search termination conditions;

[0065] Figure 11 Visualize the results.

[0066] Figure 12 The graph shows the comparison results of algorithm speeds;

[0067] Figure 13 This is a comparison chart of algorithm accuracy.

[0068] Figure 14 The average of the aggregated error points of each algorithm under different terrains and viewpoint heights. Detailed Implementation

[0069] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0070] The purpose of this invention is to provide a video surveillance view analysis method based on an improved PDERL algorithm, so as to achieve fast approximate view analysis of video surveillance.

[0071] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0072] This invention provides a video surveillance view analysis method based on an improved PDERL algorithm. Figure 1 The view analysis flowchart provided by this invention includes the following methods:

[0073] like Figure 1 The flowchart shown is a process for an improved video surveillance view analysis algorithm based on PDERL, including:

[0074] Step S1: Read the DEM data downloaded from the Copernicus website and create a video surveillance point feature within its coverage area, which includes attributes such as field of view and viewing distance.

[0075] Step S1 specifically includes:

[0076] Step S11: Acquire terrain file data and perform data processing.

[0077] In this step, the present invention uses three DEM datasets, corresponding to plains, hills, and mountains, respectively. Each DEM dataset consists of 3600*3600 grid points. Elevation values ​​are obtained through the coordinates of the corresponding grid points. Outliers (less than 0 or greater than the maximum height) are identified and processed. The final 3D terrain map is shown below. Figure 2 As shown.

[0078] Step S12: Input the video monitoring point parameter information and calculate the visible area.

[0079] In this step, the present invention acquires video surveillance point parameter information (line of view r, latitude and longitude, scanning start angle α, scanning end angle β, elevation value H). a The visible area is constructed based on its viewing distance and field of view. For example... Figure 3 As shown, the latitude and longitude of the video surveillance data are combined with the geographic transformation parameters of the DSM file to convert it into grid point coordinate values ​​(x). c ,y c The rectangular boundary region consisting of min_x, max_x, min_y, and max_y is determined by the viewing distance r, and the equations L1 and L2 of the visible region boundary dividing lines can be derived from (x... c ,y c ), α, β are determined, and the actual analysis region consists of the region where the included angle range formed by β-α intersects with the boundary region.

[0080] Step S2: Adopt a specific zoning strategy based on the angle range and analysis area, specifically including:

[0081] like Figure 4 As shown, when the starting angle α∈[0,90), the ending angle β∈[180,270), and the angle range γ∈[180,270), the analysis range is divided into Region I (located below the video surveillance) and Region II (located to the right). The analysis direction for Region I is from west to east and from north to south; the analysis direction for Region II is from south to north and from west to east. Taking Region I as an example, west to east represents the search direction for the current row, and north to south represents the overall search direction for Region I. This division strategy allows each partition to utilize vertical or horizontal grid line information. Other cases are as follows... Figure 5 As shown.

[0082] Step S3: Different strategies are employed to construct the initial reference line depending on whether the observation point is located within a grid point. Specifically, this includes:

[0083] Step S31: Determine if the observation point is located at a grid point, and select different initial reference line construction strategies.

[0084] like Figure 6 As shown, if the observation point is located on a grid point, then equations (1) and (2) are satisfied:

[0085] (x c -lon_s)%d x =0(1)

[0086] (y c -lat_s)%d y =0(2)

[0087] Then construct the grid points (P0,…,P) of the observation point and its row / column. c-1 ,P c ,P c+1 ,…,P c+k ,…P n The line of sight is used to calculate the elevation H at the intersection of each line of sight and the grid using linear interpolation. ci , H ci Compared with the actual elevation value H of each grid point ai Compare and judge its visibility; if P c To P i A line of sight is constructed if and only if on the same line of sight (P0,…,P…) i-1 All can be seen and H ci ≥H ai Then P i Visible, otherwise invisible; since these grid points are in the same row / column as the viewpoint and will not obscure subsequent rows / columns, the next row / column is considered visible and the initial reference line RL is constructed;

[0088] When the observation point is not on a grid point, the visibility of the nearest row / column grid point to the observation point is accurately calculated through linear interpolation, and an initial reference line is constructed.

[0089] Step S32: Initial reference line construction.

[0090] The initial reference line construction first considers the first row / column of the grid points in the observation point analysis as visible, and converts the x-coordinate, y-coordinate, and elevation h of each grid point in the geocentric coordinate system into the proximity p, direction d, and elevation e in PDE space; the conversion formulas are shown in equations (3), (4), and (5):

[0091]

[0092]

[0093]

[0094] like Figure 7As shown, adjacent grid points are transformed into continuous polylines in PDE space. The PDERL algorithm retains the slope 'a' and the final 'd' value of each polyline to store information about nearby grid points. LR records this information, and LC represents the current row / column of the analysis. LC visibility is determined by finding the 'd' value interval [d0,…d0] in the LR corresponding to the grid point's 'd' value. c-1 ,d c ,d c+1 ,d n The corresponding broken line is used, and the value of e is calculated based on a and d. This e is then... c The value corresponding to the largest e in relation to d is max The values ​​are compared to determine visibility, and then updated. If the condition is met (6), the object is visible; otherwise, it is not visible.

[0095] e c -e max ≥0(6)

[0096] Step S4: Determine the start and end boundary coordinates within the visible range based on the dividing line equation and the grid coordinates of the video monitoring points. This specifically includes:

[0097] Step S41: Determine the equations of the start and end dividing lines based on the coordinates of the video monitoring points and the slope of the start and end dividing lines.

[0098] like Figure 8 As shown, the coordinates of the video surveillance grid points are (x... c ,y c In the image below, the video monitoring point is not located on a grid point, so the elevation value cannot be directly obtained by accessing the array via coordinates. In this case, the nearest grid point is used as the center coordinate for analysis, denoted as (x1, y1). The starting column coordinate is start_y, the ending column coordinate is end_y, and the column start adjustment value is d. s The column termination adjustment value is d e The equation for the initial dividing line is shown in equation (7):

[0099] L2: y2=tanβ*(xx c )+y c (7)

[0100] The equation for the termination dividing line is shown in equation (8):

[0101] L1:y1=tanα*(xx c )+y c (8)

[0102] Step S42: The visible area determines the search direction according to the partitioning strategy table.

[0103] The visual analysis area is the right side of the video surveillance, oriented from south to north and from west to east. This side analysis involves incrementing the x-coordinate column by column from the right side of the video surveillance point. The x-coordinate of the starting column is x1+1. Substituting this x-coordinate into L1 and L2 yields the start and end coordinates. The start and end coordinates of subsequent columns do not require this calculation method. The x-coordinate of each column increases by 1, and L1 and L2 are both straight lines. The change in y-coordinate is shown in equation (9):

[0104] Δy=kΔx (9)

[0105] Where k represents the slope of the dividing line equation.

[0106] The corresponding adjustment value for L1 and L2 is d. e d s The calculation formulas are shown in equations (10) and (11):

[0107] d s =tanβ(10)

[0108] d e =tanα(11)

[0109] Step S43: Start and end grid coordinate adjustment strategy.

[0110] After adding ds and de, the start and end grid coordinates of each column are not within the grid points, so they need to be rounded. In order to make the information of near-layer grid points usable by the information of far-layer grid points, de along the positive y-axis is rounded up, and ds along the negative y-axis is rounded down. Therefore, the start and end grid coordinates of subsequent columns are calculated as shown in equations (12) and (13):

[0111]

[0112]

[0113] Among them ys i The y-coordinate of the starting grid for the current analysis column is represented by ys. i-1 This represents the y-coordinate of the starting grid in the previous column; where ye i The y-coordinate of the current analysis column's ending grid is represented by ye. i-1 This represents the y-coordinate of the terminating grid in the previous column.

[0114] Step S5: Optimize grid point search within the visible area, specifically including:

[0115] Step S51: To ensure that the coordinates of a row or column after the adjustment value is added do not exceed the boundaries of the visual analysis area, this invention introduces a callback strategy. When the coordinates of a row or column exceed the visual analysis area due to adjustment, the coordinates are adjusted to the boundary value of the visual analysis area. This strategy ensures that the analysis area represented by the final column or row is not lost due to the adjustment value.

[0116] As shown Figure 9 in the figure, the analysis area is on the south side, and the horizontal grid line information is utilized. The coordinates of the first grid point in the starting row are (s x , s y ), and the starting dividing line adjustment value is d s . It can be seen from the figure that the starting coordinates of the second row are s x - d s , and this value is less than the boundary min_x of the visible area. If the starting coordinates of the subsequent rows are assigned as s x , and the actual analysis area is RegionⅡ, the visibility of the grid points in RegionⅠ will be lost. To prevent the loss of the visibility information of the grid points in RegionⅠ, the coordinate callback strategy is used to adjust the coordinate values: the starting coordinates of the subsequent rows are all assigned as the left boundary value of the visible area, that is, min_x, so as to ensure that the visibility information of the grid points in RegionⅠ and RegionⅡ can be determined.

[0117] Step S52: To optimize the grid point search, by comparing the grid position where the dividing line is located in the starting direction with the boundary area, if the starting grid position exceeds the boundary area, the analysis is terminated, so as to reduce the ineffective search for non-target areas and improve the analysis efficiency.

[0118] As shown Figure 10 in the figure, the analysis area is located on the southwest side of the video monitoring point, the row search direction is from north to south, and the overall search direction is from east to west. The starting coordinates of the first column are (s x , s y ), and the PDERL algorithm will analyze from the column where s x is located to the column where min_x is located, but it will analyze the grid points in non-target areas. Therefore, a new restriction is added to the search condition. The starting coordinates corresponding to L1 are recorded as start_y. If it exceeds the boundary area min_y, that is, start_y < min_y, the search is terminated.

[0119] Step S6: Merge the two-dimensional visible arrays of each sub-region and visualize the results

[0120] Merge the visible results of each region (if there are multiple regions). Finally, the visibility of the region is maintained by a two-dimensional visible array. This array is centered on the video monitoring point and has no longitude and latitude information. Therefore, it is necessary to find out the longitude and latitude information of the starting point of this array, establish the corresponding grid visibility file, and save the video monitoring point as a shapefile file with attributes such as longitude and latitude, viewing distance, starting angle, ending angle, and viewing point height, and visualize the results in GIS, as Figure 11 shown in the figure.

[0121] To better implement the effects of the present invention, the following experiments are carried out:

[0122] The algorithm of this invention is implemented in Python. The experimental area is three DEMs (N34E114, N41E119, and N28E097) in the Copernicus DEM, with a DSM resolution of 30m. The experimental test computer has an Intel(R) Core(TM) i7-10875H CPU @ 2.30GHz, 16GB of memory, and a 64-bit Windows 10 operating system.

[0123] 1. Algorithm speed comparison

[0124] The speed of the SPDERL algorithm of this invention was compared with that of the Xdraw approximation algorithm. With a fixed radius of 10000m, observation points were randomly selected in three different terrains, with latitude, longitude, starting angle, and ending angle. The starting observation height was 0m, and the ending height was 5000m. Each experiment was repeated 50 times at 50m intervals to calculate the average time of the SPDERL algorithm / the Xdraw algorithm time. The experimental results are as follows: Figure 12 As shown.

[0125] As shown in the figure, the time taken by the algorithm of this invention is less than half that of the Xdraw algorithm in any terrain region. In mountainous terrain, the time consumed by the SPDERL algorithm increases with the increase of the viewpoint height. In plain and hilly terrain, when the viewpoint height is close to the ground, the average time consumption ratio between the algorithm of this invention and the Xdraw algorithm is less than 0.2. As the viewpoint height increases, the average time consumption ratio fluctuates around 0.35.

[0126] 2. Algorithm Accuracy Comparison

[0127] This experiment compared the accuracy of the algorithm of this invention with that of the R3 algorithm. Random locations, radii, and latitudes and longitudes were selected in three different terrains. The initial observation height was 0, and the experiment was repeated 50 times, with the height increasing by 50m until it reached 5000. The experimental results are as follows: Figure 13 As shown.

[0128] As shown in the figure, the overall average error rate of the algorithm of this invention is less than 5%, and the overall trend is that the error rate gradually decreases with the increase of viewpoint height. Specifically, in plains areas, the error rate is relatively high at low viewpoint heights, fluctuating around 0.005 with increasing viewpoint height, and stabilizing below 0.005 after 3000 meters. In mountainous terrain, the error rate peaks at approximately 1000 meters and 2000 meters, while the overall error rate is relatively low and stable at other heights. In hilly terrain, the error rate is highest in the first 1000 meters of viewpoint height, and relatively low and stable at other heights.

[0129] 3. Error point aggregation

[0130] Based on the PDERL algorithm, the proposed SPDERL algorithm is compared with the XDraw and XPDERL algorithms in terms of error point aggregation. This invention uses the x-NAE index based on x-neighborhood area error (x-NE) for verification. x-NAE refers to the region where 90% of the grid points within x grid distances of the center grid point have visible errors. The degree of error point aggregation is measured by the statistical occurrence rate of x-NAE (x-NAER). The x-neighborhood area error rate (X-NAER) is calculated by dividing the number of occurrences of X-neighborhood area error (X-NAE) by the total number of grid points in the analysis area. In this experiment, PDERL, XPDERL, XDraw, and SPDERL were used with fixed radii, random field of view angles, and random viewpoint positions on three DEMs, from a height of 3m to 75m above the ground, with 3m intervals between each experiment, for 1000 trials each. Figure 14 Table 2 presents the experimental results:

[0131] Table 2 shows the sum of the mean errors at each neighborhood aggregation point, and the ratio of the SPDERL algorithm to the XPDERL algorithm in this paper.

[0132]

[0133] Overall, among the three terrain types, the error aggregation in hilly areas is significantly higher than that in plains and mountains. This is likely due to the greater concentration of view analysis errors caused by the complexity of hilly terrain. Plains have the lowest error aggregation, possibly because the terrain is relatively flat with fewer obstructions to the view. Furthermore, as the neighborhood range increases, the error aggregation ratio in all terrain types shows a decreasing trend, indicating that the SPDERL algorithm is more sensitive to neighborhood changes than the XPDERL algorithm, and its error point aggregation is more gradual. Specifically, in mountainous terrain, the highest average aggregation error point value of the SPDERL algorithm is 0.004, far less than the XPDERL algorithm's 0.035 and the XDraw algorithm's 0.06. The overall error value of the SPDERL algorithm is also lower than that of the XPDERL algorithm and far lower than that of the XDraw algorithm. In plains terrain, the error point aggregation degree of the three algorithms is less than that of the other two terrain types. The highest average error point value of the SPDERL algorithm is only 0.000102, and there is no error point aggregation problem after the 2-neighborhood. In hilly terrain, the error point aggregation degree of the three algorithms is the highest among the three terrain types. Among them, the average error point of the algorithm proposed in this invention reaches 0.0041, but it is more moderate compared to the XPDERL and XDraw algorithms. In terms of error point aggregation performance, the algorithm SPDERL proposed in this invention significantly outperforms the XDraw algorithm and is superior to the XPDERL algorithm.

[0134] Through the above experiments, the algorithm of this invention has achieved an effective balance between speed and accuracy in the field of field of view analysis for devices such as video surveillance whose analysis range changes dynamically according to physical characteristics. At the same time, it has important reference value for field of view analysis scenarios that require fast response and high accuracy.

[0135] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. Specific examples have been used to illustrate the principles and implementation methods of the invention. The descriptions of the embodiments above are only for the purpose of helping to understand the method and core ideas of the invention; furthermore, those skilled in the art will recognize that, based on the ideas of the invention, there will be changes in specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of the invention.

Claims

1. A video surveillance view analysis method based on an improved PDERL algorithm, characterized in that, It includes the following steps: (1) Obtain DSM data and video monitoring parameters under three terrain conditions, and establish a two-dimensional array of visible range based on the parameters; (2) Determine the zoning strategy based on the field of view and the visible area; Step (2) specifically includes: dividing different areas into corresponding partitioning strategies, specifying that the minimum angle is 0 degrees to represent due north, and increasing clockwise with a maximum value of 360 degrees; calculating the field of view angle range based on the input start and end angles, and determining the analysis area location and the search direction of the horizontal or vertical grid lines according to the intervals [0,90), [90,180), [180,270), [270,360), and [0,360]; when the field of view angle range is [0,90) and the analysis area is located to the right of the video monitoring point, the vertical grid lines are used, and the search is performed in the order from south to north and from west to east; when the field of view angle range is [90,180), the start angle range is [0,90), and the end angle range is [180,270], the analysis area is divided into the lower side and the right side, where the part less than 180 degrees uses the horizontal grid lines located to the right of the video monitoring point, and the part greater than 180 degrees uses the vertical grid lines located to the lower side of the video monitoring point; (3) Optimize the initial reference line by selecting different strategies based on the location of the observation point; (4) Determine the start and end boundary coordinates within the visible range based on the dividing line equation and the grid coordinates of the video monitoring points; Step (4) specifically includes the following steps: First, based on the grid coordinates of the first column / row analyzed from the video surveillance points, and combined with the equation of the dividing line, determine the starting and ending coordinates. Second, considering the positional relationship between the grid diagonal and the dividing line, calculate the adjustment value required for moving one row / column outward relative to the previous row / column. The initial adjustment value is d. s The termination adjustment value is d e The coordinates of each column / row change by a fixed amount of 1, and L1 and L2 are both straight lines. Then the change value of y is as shown in equation (1): Where k represents the slope of the dividing line equation; Therefore, d s With d e The calculation formulas are shown in equations (2) and (3): When the dividing line of the monitoring point is perfectly aligned with the diagonal of the grid, no adjustment value needs to be modified. If the dividing line shows an increasing trend, its adjustment value should be rounded up, as shown in equation (4). Among them ys i The y-coordinate of the starting grid for the current analysis column is represented by ys. i-1 Represents the y-coordinate of the starting grid in the previous column; For dividing lines that show a decreasing trend, their adjustment values ​​should be rounded down, as shown in equation (5): Among them ye i The y-coordinate of the current analysis column's ending grid is represented by ye. i-1 This represents the y-coordinate of the terminating grid cell in the previous column; (5) Optimization of grid point search within the visible area; (6) Merge the two-dimensional visual arrays of each sub-region and visualize the results.

2. The video surveillance view analysis method based on the improved PDERL algorithm according to claim 1, characterized in that, Step (1) specifically includes the following steps: Read the DSM elevation values ​​and remove outliers, storing them in a two-dimensional array H. Then, based on the DSM data geographic transformation parameters, including the top-left x-coordinate, horizontal resolution, rotation parameters, top-left y-coordinate, rotation parameters, and vertical resolution, store the DSM elevation array with the bottom-left corner as the origin. The terrain file's starting latitude and longitude is (lon_s, lat_s), and the grid spacing is d. x d y ; Read parameters from the monitoring attribute data table, including line-of-sight r and monitoring point P. c latitude and longitude (lon) c ,lat c ), scan start angle α, scan end angle β, and elevation value H a The video surveillance latitude and longitude coordinates are combined with the geographic transformation parameters of the DSM file to convert them into grid point coordinates (x). c ,y c ); Based on the monitoring attribute parameters, a fan-shaped region centered on the monitoring point and with a line-of-sight distance r as its radius was first determined. The angular range of this region was determined by the scanning start angle α and the scanning end angle β. Two boundary dividing lines, L1 and L2, correspond to the areas passing through the monitoring point P, respectively. c And the slopes are straight lines with tanα and tanβ, as shown in equations (6) and (7): L1: L2: 。 3. The video surveillance view analysis method based on the improved PDERL algorithm according to claim 1, characterized in that, Step (3) specifically includes the following steps: If the observation point is located on a grid point, then equations (8) and (9) are satisfied: Then construct the grid points (P0,...,P) of the observation point and its row / column. c-1 ,P c ,P c+1 ,…,P c+k ,…P n The line of sight is used to calculate the elevation H at the intersection of each line of sight and the grid using linear interpolation. ci , will H ci Compared with the actual elevation value H of each grid point ai Compare and judge its visibility; if P c To P i A line of sight is constructed if and only if on the same line of sight (P0,…,P…) i-1 All can be seen and H ci ≥H ai Then P i Visible if not visible otherwise; since these grid points are in the same row / column as the observation point and will not obscure subsequent rows / columns, the next row / column is considered visible and an initial reference line RL is constructed; when the observation point is not on a grid point, the visibility of the nearest row / column grid point to the observation point is accurately calculated through linear interpolation, and an initial reference line is constructed. The reference line construction first considers the first row / column of the grid points in the observation point analysis as visible, and converts the x-coordinate, y-coordinate, and elevation h of each grid point in the geocentric coordinate system into the proximity p, direction d, and elevation e in the PDE space; the conversion formulas are as shown in equations (10), (11), and (12): Adjacent grid points are transformed into continuous polylines in PDE space. The PDERL algorithm retains the slope 'a' and the final 'd' value of each polyline to store information about nearby grid points. LR records this information, and LC represents the current row / column of the analysis. LC visibility is determined by finding the 'd' value interval [d0,…d] within LR for each grid point. c-1 ,d c ,d c+1 ,d n The corresponding polyline is given, and e is calculated based on a and d. c The value, and the e c The value corresponding to the largest e in relation to d is max The values ​​are compared to determine visibility, and then updated; for example, symbol (13) represents visibility, and vice versa. .

4. The video surveillance view analysis method based on the improved PDERL algorithm according to claim 1, characterized in that, Step (5) specifically includes the following steps: When the coordinates of a row or column exceed the visible analysis area due to adjustment, the coordinates are adjusted to the boundary value of the visible analysis area. This strategy ensures that the analysis area represented by the final column or row is not lost due to the adjustment value. The grid position of the dividing line in the starting direction is compared with the boundary region. If the starting grid position exceeds the boundary region, the analysis is terminated. This reduces invalid searches of non-target regions and improves analysis efficiency.

5. The video surveillance view analysis method based on the improved PDERL algorithm according to claim 1, characterized in that, Step (6) specifically includes the following steps: If there are multiple sub-regions, the visibility results of each region are merged, and the visibility of the region is maintained by a two-dimensional visibility array. Based on this visibility array, a raster visibility file is created, and the video monitoring points are saved as shapefiles with attributes such as latitude and longitude, viewing distance, start angle, end angle, and viewpoint height. The results are then visualized in GIS.

Images