Parallel optimization visualization method for large-range monitoring video fused with geographic scene

Abstract

The invention belongs to the technical field of video geographic information visualization, and discloses a parallel optimization visualization method for a large-range monitoring video fused with a geographic scene, which comprises the following steps: S1, initializing parameters, forming a three-dimensional video tile set according to existing monitoring camera parameters, wherein a view cone range formed by a single monitoring camera and a video fusion scene are defined as video three-dimensional tiles; s2, constructing an octree index model oriented to large-range multi-monitoring video scene fusion, inserting cameras with video overlapping areas into the same node during construction, and then inserting cameras in non-overlapping video areas into the remaining nodes; and S3, the scene is optimized in parallel based on WebGL, large-scale geographical fusion scene dynamic scheduling is realized based on a scene scheduling algorithm, and large-scale efficient visualization is realized. According to the invention, large-range efficient visualization of video data is realized by using a scene parallel optimization rendering strategy, and the large-range video monitoring effect and efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of video geographic information visualization, and particularly relates to a parallel optimization visualization method for fusing a large-range monitoring video with a geographic scene. Background Art

[0002] Video surveillance data is widespread in urban scenes, with characteristics such as low cost, real-time nature, and spatial features, and is an important data source for current GIS. The fusion of video data with a three-dimensional geographic scene to achieve immersive access to video and information can be widely applied to important fields such as disaster monitoring, traffic law enforcement, and military applications.

[0003] Existing research has made many contributions to the fusion of video data with a three-dimensional geographic scene, which can be divided into a method of enhancing a geographic scene with video and a method of enhancing a video with a geographic scene. Among them, the method of enhancing a video with a geographic scene uses the ShadowMap technology to map video textures onto the geographic scene and performs fusion processing, which is more in line with human cognitive habits.

[0004] Currently, the method of enhancing a video with a geographic scene mainly includes video registration and fusing a video with a three-dimensional scene. Regarding the video registration method, a large amount of work has been done in existing research, which can be divided into a direct estimation method and an indirect estimation method. Regarding fusing a video with a geographic scene, since the way that Cesium supports rectangular fitting of video textures does not conform to the actual spatial relationship, the ShadowMap technology is generally used to achieve the fusion of video and geographic scene, which is more in line with the actual scene. Existing technical methods mainly use a pre-downsampling method to form multiple video files to construct an LOD model, and use a quadtree or octree method to construct a hierarchical and block-based organization method. However, limited by the Web side and the real-time nature of the video stream, this technology cannot pre-downsample the real-time video stream to construct an LOD model; the current technical methods use an index tree (such as a quadtree or octree) organization method, and do not uniformly organize multiple cameras with video overlapping areas, and cannot uniformly schedule the camera group in the overlapping area, which may cause artifacts in the multi-video overlapping area. At the same time, the parallel processing ability of the rendering pipeline is not fully utilized, which limits the subsequent application of the large-range monitoring video fusion scene.

[0005] The present invention mainly aims at the following two deficiencies existing in the existing large-range monitoring video-enhanced GIS method:

[0006] (1) The current organization method does not consider the unified management of multiple overlapping video cameras, and moreover, the existing LOD model is difficult to apply to real-time video streams. That is, the organization method is not applicable.

[0007] (2) The current large-scale scenario scheduling algorithm does not consider the parallel optimization calculation of WebGL, and the overlapping areas of multiple videos are not effectively fused, resulting in low visualization efficiency. The current method still cannot meet the visualization requirements of large-scale monitoring scenarios. That is, the algorithm does not meet the requirements. Summary of the Invention

[0008] The object of the present invention is to provide a parallel optimization visualization method for fusing large-scale monitoring videos with geographical scenes, which is used to solve the technical problems that the existing video organization methods are difficult to be applicable to the video streams of multiple overlapping video cameras, and the overlapping areas of multiple videos are difficult to be effectively fused, resulting in low visualization efficiency.

[0009] The described parallel optimization visualization method for fusing large-scale monitoring videos with geographical scenes includes the following steps.

[0010] S1. Initialize parameters, and form a three-dimensional video tile set according to the existing monitoring camera parameters, where the viewing frustum range formed by a single monitoring camera and the video fusion scene are defined as video three-dimensional tiles.

[0011] S2. Build an octree index model for fusing large-scale multi-monitoring video scenes. When building, first insert the cameras with video overlapping areas into the same node, and then insert the cameras in non-overlapping video areas into the remaining nodes.

[0012] S3. Optimize the scene in parallel based on WebGL, and implement dynamic scheduling of large-scale geographical fusion scenes based on the scene scheduling algorithm to achieve high-efficiency visualization of large scales; the scheduling process of this step is based on the organization method of the octree index model in step S2, and uses the dynamic LOD strategy to uniformly process the video three-dimensional tiles in the current node OctreeNode with the same rendering pipeline.

[0013] Preferably, the specific steps of step S1 include the following.

[0014] S1.1. Define the viewing frustum range formed by a single monitoring camera and the video fusion scene as video three-dimensional tiles.

[0015] S1.2. Form clusters of video three-dimensional tiles from the three-dimensional video tile sets corresponding to the camera parts in the overlapping areas, and each cluster forms a cluster set.

[0016] Preferably, the expression of the video three-dimensional tile is defined as:

[0017]

[0018] Where S represents the acquired shadow map; V represents the video stream; g(V) represents the video frames extracted from the video stream; f(S, g(V), a) represents the formation of a video texture with geographical features from the shadow map S and the video frames g(V), and then superimposing the video texture onto the three-dimensional geographical scene according to the transparency value a (the value range of a is between 0 and 1).

[0019] Preferably, each cluster Cluster has a minimum bounding box OBB cluser and the number N of three-dimensional video tiles; the other three-dimensional video tiles have a minimum bounding box OBB, and the corresponding number of three-dimensional video tiles is 1; each cluster Cluster forms a cluster set ClusterSet.

[0020] Preferably, step S2 specifically includes.

[0021] S2.1: Set the maximum value of each node in the current octree index for storing three-dimensional video tiles as T, and calculate the minimum outer bounding of the current three-dimensional tile set to obtain the root node Root of the octree node.

[0022] S2.2 Traverse the clusters Cluster in the cluster set ClusterSet and insert the cluster Cluster into the nodes of the octree.

[0023] Repeat step S2.2 until all the clusters Cluster in the cluster set ClusterSet are inserted into the octree.

[0024] S2.3 Insert the remaining three-dimensional video tiles that do not belong to the cluster Cluster into the corresponding octree nodes.

[0025] Repeat step S2.3 until all the remaining three-dimensional video tiles are inserted into the octree.

[0026] Preferably, the step S2.2 specifically includes.

[0027] S2.2.1 Set the current node OctreeNode, and let OctreeNode = Root.

[0028] S2.2.2 Obtain each cluster Cluster in turn; among them, the minimum bounding box of the current cluster Cluster is OBB cluser and the number of three-dimensional video tiles therein is N.

[0029] S2.2.3 Make a judgment. If the minimum bounding box OBB of the current node OctreeNode node does not contain the minimum bounding box OBB of the current cluster cluser , then jump back to step S2.2.2 to obtain the next cluster Cluster; otherwise, execute the next step.

[0030] S2.2.4 If the number N of video three-dimensional tiles in the current cluster is greater than the remaining storage capacity of the current node OctreeNode, then the current node is formed into 8 octree child nodes OctreeChild according to the octree segmentation method; then traverse the child nodes OctreeChild, reset the current node OctreeNode, make OctreeNode = OctreeChild, and then jump back to step S2.2.2 to obtain the next cluster Cluster; otherwise, insert the video three-dimensional tiles in the current cluster Cluster into the current node OctreeNode.

[0031] Preferably, the step S2.3 specifically includes the following.

[0032] S2.3.1 Set the current node OctreeNode, make OctreeNode = Root.

[0033] S2.3.2 Obtain each video three-dimensional tile in turn; among them, the minimum bounding box of the video three-dimensional tile is OBB.

[0034] S2.3.3 If the minimum bounding box OBB of OctreeNode node does not contain OBB, then jump back to step S2.3.2 to obtain the next video three-dimensional tile and execute the subsequent steps; otherwise, execute the next step.

[0035] S2.3.4 If the current node OctreeNode has no remaining storage capacity, then obtain 8 octree child nodes OctreeChild; if there are no child nodes in the current node OctreeNode at this time, then form 8 octree child nodes OctreeChild for the current node OctreeNode according to the octree segmentation method; after obtaining the child nodes, reset the current node OctreeNode, make OctreeNode = OctreeChild, and then jump back to step S2.3.2 to perform subsequent operations on the next video three-dimensional tile; if the current node OctreeNode has remaining storage capacity, insert the video three-dimensional tile into the current node OctreeNode.

[0036] Preferably, in the step S3, the update of the three-dimensional geographical scene is divided into the following three stages.

[0037] S3.1 Before scene update: Use the breadth-first traversal algorithm to judge the visibility of each node of the root node Root between the current camera frustum and the octree node, obtain the visible octree node set NodeSet and the corresponding video three-dimensional tile set TileSet, and obtain the LOD model of the current video three-dimensional tile during the traversal.

[0038] S3.2 Scene update: Asynchronously request the video data service for the video 3D tile set TileSet, and render the field of view of the current monitoring camera and the video texture; wherein, in this step, the set of all video 3D tiles in the same node is transmitted to the GPU side at once, and the union operation is performed on the shadow map regions of multiple video 3D tiles at the GPU side to calculate the fusion.

[0039] After S3.3 scene update: Asynchronously call to delete the invisible video 3D tiles. An asynchronous request mechanism is adopted in this stage.

[0040] Preferably, the specific implementation steps of building the LOD model in step S3.1 are as follows.

[0041] S3.1.1 If the current video 3D tile is visible, set the interval extraction threshold threshold, and the default count = 0; in this step, different interval extraction thresholds threshold are set according to the distance distance between the camera and the video 3D tile.

[0042] S3.1.2 Refresh the scene page according to the Cesium scene update loop mechanism, and increment count = count + 1 for each update.

[0043] S3.1.3 If count < threshold, jump back to step S3.1.2; otherwise, execute the next step.

[0044] S3.1.4 When count = 0, extract the current video frame Frame.

[0045] Preferably, the specific implementation steps of calculating the fusion and rendering in step S3.2 are as follows.

[0046] S3.2.1 Obtain the NodeSet and the corresponding video 3D tile set TileSet.

[0047] S3.2.2 Transfer the video 3D tiles of the video 3D tile set TileSet in the same node to the GPU side at once.

[0048] S3.2.3 Obtain the video frame Frame and the shadow map ShadowMap in the video 3D tile set TileSet.

[0049] S3.2.4 Utilize the parallel processing ability of the WebGL shader to perform the union operation on the shadow map set and the video frame set respectively to obtain a fused video frame FrameU and a fused shadow map ShadowMapU.

[0050] S3.2.5 Render the current video-fused geographic scene based on the fused video frame FrameU and the fused shadow map ShadowMapU.

[0051] The present invention has the following advantages: In step S2 of the present invention, the octree organization method is improved. In the improved octree index model, cameras with video overlapping areas are first inserted into the same node, and then cameras in non-overlapping video areas are inserted into the remaining nodes, which is convenient for subsequent unified management and scheduling. By using the dynamic LOD strategy, the 3D video tiles in the OctreeNode are uniformly processed by the same rendering pipeline, improving the visualization effect and efficiency. Through WebGL parallel processing, the required tile nodes are quickly obtained by traversing the large-scale organization method of video fusion, and the scene parallel optimization rendering strategy is used to achieve efficient visualization of large-scale video data, improving the large-scale monitoring video effect and efficiency, and providing technical support for subsequent business applications such as navigation and automated monitoring. Existing LOD models are difficult to apply to real-time video streams. Description of the Drawings

[0052] Figure 1 It is a basic flowchart of a parallel optimization visualization method for fusing a large-scale monitoring video with a geographic scene according to the present invention.

[0053] Figure 2 It is a local visualization effect diagram in a large-scale camera video overlapping environment obtained by applying the present invention.

[0054] Figure 3 It is a comparison chart of the frame rate (FPS) obtained by comparing the present invention with the prior art. Detailed Embodiments

[0055] Next, with reference to the drawings, through the description of the embodiments, the specific embodiments of the present invention will be further described in detail to help those skilled in the art have a more complete, accurate and in-depth understanding of the inventive concept and technical solution of the present invention.

[0056] As Figure 1 shown, the present invention provides a parallel optimization visualization method for fusing a large-scale monitoring video with a geographic scene, including the following steps.

[0057] S1, Initialize parameters.

[0058] In this step, a set of 3D video tiles is formed according to the existing monitoring camera parameters, specifically including the following sub-steps.

[0059] S1.1 Define the frustum range formed by a single monitoring camera and the video fusion scene as a video 3D tile, which can be abstracted into the following expression.

[0060]

[0061] Where S represents the acquired shadow map; V represents the video stream; g(V) represents the video frames extracted from the video stream; f(S, g(V), a) represents the formation of a video texture with geographical features from the shadow map S and the video frames g(V), and then superimposing the video texture onto the 3D geographical scene according to the transparency value a (where the value range of a is between 0 and 1).

[0062] Among them, the shadow map S includes camera parameters and the shadow map formed based on the camera parameters. The camera parameters are divided into internal parameters and external parameters. The internal parameters include the field of view angle, near plane, far plane, and video aspect ratio; the external parameters mainly include the geographical location of the camera (i.e., longitude, latitude, and height), and the camera pose characteristics (i.e., yaw, pitch, and roll). The video stream V includes a uniform resource locator and the corresponding network link, and the video stream V forms a video after video address resolution.

[0063] S1.2 The clustering Cluster of video 3D tiles is formed by the set of 3D video tiles corresponding to the camera parts in the overlapping area, and each clustering Cluster forms a clustering set ClusterSet. Among them, each clustering Cluster has a minimum bounding box OBB cluser and the number N of video 3D tiles. The other video 3D tiles have a minimum bounding box OBB, and the corresponding number of video 3D tiles is 1.

[0064] S2. Construct an octree index model for large-scale multi-surveillance video scene fusion.

[0065] Different from the traditional octree organization, this step uses an improved octree index model. When constructing, the cameras with video overlapping areas are first inserted into the same node, and then the cameras in non-overlapping video areas are inserted into the remaining nodes, so as to facilitate subsequent unified management and scheduling. This step specifically includes the following sub-steps.

[0066] S2.1: Set the maximum value of the 3D video tiles stored in each node of the current octree index to T, and calculate the minimum external bounding of the current 3D tile set to obtain the root node Root of the octree node.

[0067] S2.2 Traverse the clustering Cluster in the clustering set ClusterSet and insert the clustering Cluster into the nodes of the octree. This step includes the following sub-steps.

[0068] S2.2.1 Set the current node OctreeNode, and let OctreeNode = Root.

[0069] S2.2.2 Obtain each clustering Cluster in turn. Among them, the minimum bounding box of the current clustering Cluster is OBB cluserAnd the number of video 3D tiles therein is N.

[0070] S2.2.3 makes a judgment. If the minimum bounding box OBB of the current node OctreeNode node does not contain the minimum bounding box OBB of the current cluster cluser , then jump back to step S2.2.2 to obtain the next cluster Cluster; otherwise, execute the next step.

[0071] S2.2.4 If the number of video 3D tiles N of the current cluster is greater than the remaining storage capacity of the current node OctreeNode, then the current node is divided into 8 octree child nodes OctreeChild according to the octree splitting method; then traverse the child nodes OctreeChild, reset the current node OctreeNode, make OctreeNode = OctreeChild, and then jump back to step S2.2.2 to obtain the next cluster Cluster; otherwise, insert the video 3D tiles in the current cluster Cluster into the current node OctreeNode.

[0072] Repeat step S2.2 until all clusters Cluster in the cluster set ClusterSet are inserted into the octree.

[0073] S2.3 Insert the remaining video 3D tiles that do not belong to the cluster Cluster into the corresponding octree nodes. This step mainly includes the following sub-steps.

[0074] S2.3.1 Set the current node OctreeNode, make OctreeNode = Root.

[0075] S2.3.2 Obtain each video 3D tile in turn. Among them, the minimum bounding box of the video 3D tile is OBB.

[0076] S2.3.3 If the minimum bounding box OBB of OctreeNode node does not contain OBB, then jump back to step S2.3.2 to obtain the next video 3D tile and execute the subsequent steps; otherwise, execute the next step.

[0077] S2.3.4 If the current node OctreeNode has no available storage space, obtain 8 octree child nodes OctreeChild; if there are no child nodes for the current node OctreeNode at this time, form 8 octree child nodes OctreeChild for the current node OctreeNode according to the octree segmentation method. After obtaining the child nodes, reset the current node OctreeNode, set OctreeNode = OctreeChild, and then jump back to step S2.3.2 to perform subsequent operations on the next video 3D tile. If the current node OctreeNode has available storage space, insert the video 3D tile into the current node OctreeNode.

[0078] Repeat step S2.3 until all the remaining video 3D tiles are inserted into the octree.

[0079] S3. Optimize the scene in parallel based on WebGL (Web Graphics Library).

[0080] This step realizes the dynamic scheduling of large-scale geographic fusion scenes based on the scene scheduling algorithm, achieving efficient visualization on a large scale. The scheduling process of this step is based on the organization method of the octree index model in step S2, and uses the dynamic LOD strategy to uniformly process the video 3D tiles in the current node OctreeNode with the same rendering pipeline, improving the visualization effect and efficiency. This process is based on the Cesium page rendering and update process, and divides the 3D geographic scene update into the following three stages.

[0081] Before scene update in S3.1: Use the breadth-first traversal algorithm to judge the visibility of each node of the root node Root with the current camera frustum, obtain the visible octree node set NodeSet and the corresponding video 3D tile set TileSet, and obtain the LOD model of the current video 3D tile during traversal.

[0082] Among them, the LOD model mainly realizes the construction process of the dynamic LOD model by obtaining the video frame sampling frequency according to the setting, and defines different levels of LOD models by setting different extraction frame frequency methods. The specific implementation steps for constructing the LOD model are as follows:

[0083] S3.1.1 If the current video 3D tile is visible, set the interval extraction threshold threshold, and the default count = 0. This step sets different interval extraction thresholds threshold according to the distance distance between the camera and the video 3D tile.

[0084] S3.1.2 Refresh the scene page according to the Cesium scene update loop mechanism, and increment count = count + 1 for each update.

[0085] S3.1.3 If count < threshold, jump back to step S3.1.2; otherwise, execute the next step.

[0086] S3.1.4 When count = 0, extract the current video frame Frame.

[0087] Step 3.1 Define different interval extraction thresholds threshold according to the actual application effect. Obviously, the farther the video 3D tiles are from the camera, the higher the set interval extraction threshold threshold, the lower the frequency of extracting video frames, and the lower the corresponding LOD model level.

[0088] S3.2 Scene update: Asynchronously request the video data service for the video 3D tile set TileSet, and render the field of view of the current monitoring camera and the video texture.

[0089] Among them, this step transfers the set of all video 3D tiles in the same node to the GPU side at once, and performs a union operation on the shadow map regions of multiple video 3D tiles at the GPU side to calculate the fusion, avoiding multiple inputs to the GPU and reducing the rendering efficiency, and also contributing to the realization of texture fusion in the video overlapping region. The specific implementation steps of calculation fusion and rendering are as follows.

[0090] S3.2.1 Obtain the NodeSet and the corresponding video 3D tile set TileSet.

[0091] S3.2.2 Transfer all the video 3D tiles in the video 3D tile set TileSet in the same node to the GPU side at once.

[0092] S3.2.3 Obtain the video frame Frame and the shadow map ShadowMap in the video 3D tile set TileSet.

[0093] S3.2.4 Use the parallel processing ability of WebGL shaders to perform a union operation on the shadow map set and the video frame set respectively to obtain a fused video frame FrameU and a fused shadow map ShadowMapU.

[0094] S3.2.5 Render the current video fusion geographical scene based on the fused video frame FrameU and the fused shadow map ShadowMapU.

[0095] S3.3 After scene update: Asynchronously call the invisible video 3D tiles to delete the tiles. This stage adopts an asynchronous request mechanism.

[0096] In the dynamic scheduling process, the same rendering pipeline is used to optimize the same node of the octree in parallel, and the video overlapping areas are processed uniformly, achieving a better fusion effect while improving the rendering efficiency.

[0097] This method traverses the video to quickly obtain the required tile nodes in a large-scale fusion organization manner, and uses the scene parallel optimization rendering strategy to achieve efficient visualization of large-scale video data, improving the effect and efficiency of large-scale surveillance videos, and providing technical support for subsequent business applications such as navigation and automated monitoring.

[0098] Such as Figure 2 、 Figure 3 As shown, the rendering effect of this method in a large-scale camera video overlapping environment is simulated through experiments, and it is compared with the existing technology using the traditional octree organization method and the combination of the traditional octree organization method and the dynamic LOD model in this method. The experiment records the rendering efficiency at the same scene at 584m for different simulation scenarios with the number of videos from 1 to 70000. Obviously, the rendering efficiency of this method is higher than that of the existing technology.

[0099] The present invention has been described exemplarily above in conjunction with the accompanying drawings. Obviously, the specific implementation of the present invention is not limited by the above-mentioned manner. As long as various non-substantive improvements are made by adopting the inventive concept and technical solution of the present invention, or the inventive concept and technical solution of the present invention are directly applied to other occasions without improvement, they are all within the protection scope of the present invention.

Claims

1. A parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes, characterized by: The following steps are involved: S1. Initialize parameters and form a 3D video tile set according to existing surveillance camera parameters, wherein a viewing cone range and a video fusion scene formed by a single surveillance camera are defined as a video 3D tile; S2. Construct an octree index model for the fusion of large-scale multi-surveillance video scenes. When constructing, first insert the cameras with overlapping video areas into the same node, and then insert the cameras with non-overlapping video areas into the remaining nodes; S3, based on WebGL parallel optimization scenes, realizes dynamic scheduling of large-scale geographic fusion scenes based on scene scheduling algorithms, and realizes large-scale efficient visualization; The scheduling process of this step is based on the organization of the octree index model in step S2, and uses a dynamic LOD strategy to uniformly process the video 3D tiles in the current node OctreeNode using the same rendering pipeline.

2. The parallel optimization visualization method for integrating large-scale surveillance video with geographical scenes according to claim 1 is characterized by: The step S1 specifically includes: S1.

1. The viewing cone range and video fusion scene formed by a single surveillance camera are defined as a video 3D tile; S1.

2. A set of 3D video tiles corresponding to the camera part of the overlapping area is used to form a cluster of video 3D tiles, and each cluster forms a cluster set.

3. The parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes according to claim 2 is characterized in that: The expression for defining a 3D video tile is: Where S represents the acquired shadow map; V represents the video stream; g(V) represents the video frame extracted from the video stream; f(S, g(V), a) represents the video texture with geographical features formed by the shadow map S and the video frame g(V), and then the video texture is superimposed on the three-dimensional geographical scene according to the transparency value a (a ranges from 0 to 1).

4. A parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes according to any one of claims 1 to 3, characterized in that: Each cluster has a minimum bounding box OBB cluser And the number of video 3D tiles N; other video 3D tiles have a minimum bounding box OBB, and the corresponding number of video 3D tiles is 1; each cluster Cluster forms a cluster set ClusterSet.

5. The parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes according to claim 4 is characterized in that: Step S2 specifically includes: S2.1: Set the maximum value of the three-dimensional video tile stored in each node of the current octree index to T, calculate the minimum outer bounding box of the current three-dimensional tile set to obtain the root node Root of the octree node; S2.2 traverses the cluster Cluster in the cluster set ClusterSet and inserts the cluster Cluster into the node of the octree; Repeat step S2.2 until all clusters in the cluster set ClusterSet are inserted into the octree; S2.3 insert the remaining three-dimensional video tiles that do not belong to the cluster Cluster into the corresponding octree nodes; Repeat step S2.3 until all remaining video 3D tiles are inserted into the octree.

6. The parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes according to claim 5 is characterized in that: The step S2.2 specifically includes: S2.2.1 Set the current node OctreeNode, let OctreeNode = Root; S2.2.2 Obtain each cluster Cluster in turn; the minimum bounding box of the current cluster Cluster is OBB cluser The number of video 3D tiles is N; S2.2.3 makes a judgment, if the minimum bounding box OBB of the current node OctreeNode node The minimum bounding box OBB that does not contain the current cluster cluser , then jump back to step S2.2.2 to obtain the next cluster Cluster; otherwise, execute the next step; S2.2.4 If the number of video 3D tiles N in the current cluster is greater than the number of free storage in the current node OctreeNode, eight octree child nodes OctreeChild are formed for the current node according to the octree partitioning method; then the child nodes OctreeChild are traversed, and the current node OctreeNode is reset, so that OctreeNode = OctreeChild, and then jump back to step S2.2.2 to obtain the next cluster Cluster; otherwise, the video 3D tiles in the current cluster Cluster are inserted into the current node OctreeNode.

7. The parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes according to claim 5 is characterized in that: The specific steps of step S2.3 are as follows: S2.3.1 Set the current node OctreeNode, and let OctreeNode = Root; S2.3.2 obtain each video 3D tile in turn; wherein the minimum bounding box of the video 3D tile is OBB; S2.3.3 if the minimum bounding box of OctreeNode OBB node If OBB is not included, jump back to step S2.3.2, obtain the next video 3D tile and execute the following steps; otherwise, execute the next step; S2.3.4 If the current node OctreeNode has no remaining storage capacity, obtain 8 octree child nodes OctreeChild; if there are no child nodes for the current node OctreeNode at this time, form 8 octree child nodes OctreeChild for the current node OctreeNode according to the octree division method; after obtaining the child nodes, reset the current node OctreeNode, let OctreeNode = OctreeChild, and then jump back to step S2.3.2 to perform subsequent operations on the next video 3D tile; if the current node OctreeNode has remaining storage capacity, insert the video 3D tile into the current node OctreeNode.

8. The parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes according to claim 4 is characterized in that: In step S3, the 3D geographical scene update is divided into the following three stages: S3.1 Before scene update: Use the breadth-first traversal algorithm to judge the visibility of each node between the current camera frustum and the octree node for the root node Root, obtain the visible octree node set NodeSet and the corresponding video 3D tile set TileSet, and obtain the LOD model of the current video 3D tile during the traversal; S3.2 Scene update: Asynchronously request the video data service for the video 3D tile set TileSet, and render the field of view of the current monitoring camera and the video texture; in this step, the set of all video 3D tiles in the same node is transmitted to the GPU side at one time, and the union operation is performed on the shadow map areas of multiple video 3D tiles at the GPU side to calculate the fusion; S3.3 After scene update: Asynchronously call the method to delete the invisible video 3D tiles. An asynchronous request mechanism is adopted in this stage.

9. The parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes according to claim 8 is characterized in that: The specific implementation steps for constructing the LOD model in step S3.1 are as follows: S3.1.1 If the current video 3D tile is visible, set the interval extraction threshold threshold, and the default count = 0; in this step, different interval extraction thresholds threshold are set according to the distance distance between the camera and the video 3D tile; S3.1.2 Refresh the scene page according to the Cesium scene update loop mechanism, and increment count = count + 1 for each update; S3.1.3 If count < threshold, jump back to step S3.1.2; otherwise, perform the next step; S3.1.4 When count = 0, extract the current video frame Frame.

10. The parallel optimization visualization method for integrating large-scale surveillance video with geographic scenes according to claim 8, characterized in that: The specific implementation steps for calculating the fusion and rendering in step S3.2 are as follows: S3.2.1 Obtain NodeSet and the corresponding video 3D tile set TileSet; S3.2.2 Transmit the video 3D tiles in the video 3D tile set TileSet in the same node to the GPU side at one time; S3.2.3 obtains the video frame Frame and the shadow map ShadowMap in the video three-dimensional tile set TileSet; S3.2.4 utilizes the parallel processing capability of the WebGL shader to respectively calculate and merge the shadow map set and the video frame set to obtain a fused video frame FrameU and a fused shadow map ShadowMapU; S3.2.5 renders the current video fused geographic scene based on the fused video frame FrameU and the fused shadow map ShadowMapU.

Images