A GIS visualization system for multi-source monitoring data fusion

Abstract

This invention discloses a GIS visualization system for multi-source monitoring data fusion, specifically in the field of geographic information visualization. It addresses the problem of object misidentification when linking two-dimensional maps with three-dimensional scenes. By constructing object anchoring images and object picking images, which respectively carry monitoring object identifiers and display entity identifiers, and establishing cross-view recognition mapping results, the system can read back the monitoring object identifier from the selected pixel and locate the highlighted area of ​​the target view when switching views or selecting links. Simultaneously, it outputs a details call identifier and performs verification and correction decisions on the highlighted area of ​​the target view, achieving consistent association between highlighted display and detailed information. This improves the accuracy and stability of GIS visualization linking display in multi-source monitoring data fusion scenarios.

Description

Technical Field

[0001] This invention relates to the field of geographic information visualization, and more specifically, to a GIS visualization system for the fusion of multi-source monitoring data. Background Technology

[0002] An existing geographic information visualization system for multi-source monitoring data fusion is designed for application scenarios that aggregate monitoring information from different sources onto a single platform for unified display and interactive viewing. The system typically integrates monitoring points, status information, alarm information, trend results, and related business information, then displays them in a geographic information interface using both 2D map views and 3D scene views. It also includes features such as layer switching, object selection, linked positioning, detailed viewing, and process replay, allowing users to continuously observe, compare, and make judgments within a single visualization interface. Existing solutions for this type of system already achieve data access, fusion display, and interactive linkage, emphasizing unified management and visualization. Therefore, in practical use, users often directly rely on highlighted or selected monitoring objects on the screen for subsequent analysis and action.

[0003] The problem encountered during the above usage is that when the same monitored object is switched or linked between a 2D map view and a 3D scene view, its position and highlight status remain continuous. However, the objects used for detail display, alarm association, or trend display may change, leading to semantic shifts in the fused visualization results. This issue is more likely to occur when there are many layers, dense objects, and continuous interactive operations, manifesting as inconsistencies between the highlighted objects and the objects called for details, affecting the understanding of the monitoring status and the use of the fused results.

[0004] To address the aforementioned problems, a technical solution is provided. Summary of the Invention

[0005] To overcome the aforementioned deficiencies of the prior art, embodiments of the present invention provide a GIS visualization system for multi-source monitoring data fusion. By constructing object anchoring images and object picking images, which respectively carry monitoring object identifiers and display entity identifiers, and establishing cross-view recognition mapping results, the system can read back the monitoring object identifier from the selected pixel and locate the highlighted area of ​​the target view when switching views or making linked selections. At the same time, it outputs a details call identifier and performs a re-examination verification and correction decision on the highlighted area of ​​the target view, thereby achieving a consistent association between the highlighted display and the details information. This improves the accuracy and stability of GIS visualization linkage display in multi-source monitoring data fusion scenarios, thus solving the problems mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution: Anchoring and mapping module: Reads the monitoring object identification and spatial location data from the multi-source monitoring data fusion results, writes the displayable objects in the planar map view into the object anchoring image, so that the pixel position carries the monitoring object identification; Picking and Mapping Module: Generates object picking images based on the rendering results of the stereo scene view, writes display entity identifiers, and maps the display entity identifiers to the monitoring object identifiers according to the spatial location attribution relationship, forming a cross-view recognition mapping result; Linked positioning module: Responds to view switching or linked selection operations, reads the selected pixel in the corresponding image of the view and reads back the monitoring object identifier, locates the display area in the target view based on the cross-view recognition mapping results, and generates a highlighted area and a details call identifier; Verification and correction module: Performs re-examination verification on the highlighted area. When the re-examination is inconsistent, it extracts the boundary offset representation and kernel same object representation, inputs them into the pre-trained discrimination model to obtain the view same object confidence coefficient, and corrects the detail call identifier or reconstructs the highlighted area and updates the mapping cache accordingly.

[0007] Furthermore, based on the monitoring object identifier and spatial location data in the multi-source monitoring data fusion results, and combined with the projection relationship and viewport range of the current planar map view, a screen outline pixel set is generated. The expansion scale is determined according to the symbol display size and minimum selection constraint, and a selectable pixel area corresponding to the monitoring object is formed around the screen outline pixel set.

[0008] Furthermore, during the object anchoring image writing stage, the overlapping pixel positions are uniquely assigned according to the layer display order, screen anchoring center distance, and monitoring object identification order. Then, the monitoring object identification is written into the object anchoring image according to a fixed channel encoding rule, and a readback check is performed on non-empty pixels to maintain the consistency of the monitoring object identification.

[0009] Furthermore, based on the rendering results of the current stereoscopic scene view, an object picking image and depth buffer are generated. Off-screen rasterization rendering is used to write the display entity identifier code and determine the pixel ownership according to the depth first rule. The depth parallel positions are determined to have unique display entity identifiers according to the display entity identifier order. Background pixels are written to the background reserved value and the display entity identifier readback verification is performed.

[0010] Furthermore, based on spatial location data, a set of pixels covering the projection of the monitored object and a depth range of the monitored object are formed in the current stereoscopic scene view. A set of supporting pixels is constructed by combining the object picking image rereading results. Cross-view recognition mapping results are generated according to the display entity support ratio, monitored object coverage ratio, screen distance and monitored object identification order, and a mapping index table is established.

[0011] Furthermore, based on the view switching or linked selection operation, the image corresponding to the current view is determined and the selected pixels are read. When the image corresponding to the current view is the object anchoring image, the monitoring object identifier is read back according to the object anchoring image encoding. When the image corresponding to the current view is the object picking image, the entity identifier is read back first and then the monitoring object identifier is read back through the cross-view recognition mapping result, and the unique index positioning is completed in the monitoring object identifier set.

[0012] Furthermore, the set of display entity identifiers corresponding to the monitored object identifier is queried according to the mapping index table, and a set of candidate pixels for the target view is constructed. The set of candidate pixels for the projection coverage of the monitored object and the depth range of the monitored object are combined to form a set of constraint candidate pixels. The highlighted area of ​​the target view is determined by eight-neighbor connected component labeling and hierarchical comparison. At the same time, the detailed call identifier consistent with the monitored object identifier is output.

[0013] Furthermore, a set of repeated pixels is established based on the highlighted area of ​​the target view. The entity identifier is displayed by reading back the image of the object and the monitoring object identifier is read back through the cross-view recognition mapping result. The effective set of repeated pixels is screened out. The main monitoring object identifier is determined by the frequency of the monitoring object identifier and the distance between the center of the highlighted area of ​​the target view. Empty areas or empty identifiers enter the state of insufficient repeated pixels.

[0014] Furthermore, a set of boundary pixels is extracted around the highlighted area of ​​the target view and a boundary ring is constructed. The set of boundary inverse mapping breaks where the pixel re-extraction monitoring object identifier and the detail call identifier are inconsistent is statistically analyzed and a boundary inverse mapping break ratio is formed. At the same time, the boundary ring is removed from the highlighted area of ​​the target view to form a kernel region. The kernel same object readback depth ratio is formed by continuously reading back the set in a fixed direction.

[0015] Furthermore, a preliminary consistency judgment is made between the main monitoring object identifier and the details call identifier. If the evidence of re-retrieval is valid and there is an inconsistency, the ratio of the boundary inverse mapping fracture zone and the ratio of the kernel same object readback depth are input into the pre-trained discrimination model to generate the view same object credibility coefficient. Based on the view same object credibility coefficient, the details call identifier is corrected or the target view highlighted area is reconstructed based on the mapping cache and the re-retrieval is confirmed.

[0016] The technical effects and advantages of the GIS visualization system for multi-source monitoring data fusion proposed in this invention are as follows: This invention focuses on the linked display of two-dimensional planar map views and three-dimensional scene views in multi-source monitoring data fusion scenarios. It establishes a consistent identification link from pixel position to monitoring object identification, and then to cross-view positioning and verification correction. By forming a readable object anchoring image in the planar map view and a readable object picking image in the three-dimensional scene view, and by using cross-view recognition mapping results to achieve unified object identity, the invention eliminates reliance on interface style, color appearance, or manual visual judgment to determine object ownership when switching views or selecting linked views. It can stably correspond to the same monitoring object across different view representations, thus ensuring that the linked highlighted area and the detail call identifier have a consistent object semantic basis, reducing misselection and incorrect linkage phenomena in scenarios with dense monitoring objects, occlusion, adjacent objects, and partial overlap.

[0017] Building upon this foundation, this invention further introduces a decision path for re-verification and correction of highlighted areas in the target view. Instead of simply reconstructing the display area uniformly when inconsistencies arise, it combines boundary and kernel verification information to determine the source of inconsistency. Then, it performs correction processing on the detail call identifier or the highlighted area of ​​the target view separately, and maintains consistency in subsequent linkages through mapping caching. The resulting overall processing logic can improve the reliability of cross-view recognition results while maintaining the continuity of linkage operations. This makes the highlighted objects seen by monitoring personnel during map browsing, scene viewing, and detail calls more consistent with the retrieved information, resulting in lower comprehension costs, more stable usage, and better suited to GIS visualization application scenarios after multi-source monitoring data fusion. Attached Figure Description

[0018] Figure 1 This is a schematic diagram of the structure of a GIS visualization system for multi-source monitoring data fusion according to the present invention; Figure 2 This is a schematic diagram of the overall processing flow of the present invention; Figure 3 This is a schematic diagram of the process for generating the anchored image of the object of the present invention; Figure 4 A schematic diagram of the image picking and cross-view recognition mapping of the object of the present invention; Figure 5 This is a schematic diagram illustrating the linkage positioning of a two-dimensional map view and a three-dimensional scene view according to the present invention; Figure 6 This is a schematic diagram illustrating the review and correction decision-making process of this invention. Detailed Implementation

[0019] 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.

[0020] Please see Figures 1-6 This invention provides a GIS visualization system for multi-source monitoring data fusion, comprising: Anchoring and mapping module: Reads the monitoring object identifier and spatial location data from the fusion results of multi-source monitoring data, and writes the displayable objects in the planar map view into the object anchoring image. The displayable object refers to the monitoring object graphic expression unit that meets the display conditions within the current viewport range and is presented in the form of points, lines, surfaces or symbols, so that the pixel position carries the monitoring object identifier and forms a readable area.

[0021] Picking and Mapping Module: Generates object picking images based on the rendering results of the stereo scene view, writes them into the display entity identifiers, and maps the display entity identifiers to the monitoring object identifiers according to their spatial location, forming a cross-view recognition mapping result.

[0022] Linked positioning module: In response to view switching or linked selection operations, it reads the selected pixel in the corresponding image of the view and reads back the monitoring object identifier. Based on the cross-view recognition mapping results, it locates the display area in the target view and generates a highlighted area and a details call identifier. The details call identifier is the object identifier used to trigger the query of the monitoring object details panel, alarm association information, or trend information.

[0023] The verification and correction module performs a re-examination of the highlighted area. When the re-examination is inconsistent, it extracts the boundary offset representation and the kernel same object representation, and inputs them into the pre-trained discrimination model to obtain the view same object confidence coefficient. The view same object confidence coefficient is used to characterize the degree of confidence of whether the highlighted area of ​​the target view and the detail call identifier correspond to the same monitored object. Based on this, the detail call identifier is corrected or the highlighted area is reconstructed and the mapping cache is updated.

[0024] The solution proposed in this invention is to establish the linkage between the 2D map view and the 3D scene view not on the interface appearance or rendering results, but on a retrievable object identity link. Specifically, an object anchoring image carrying the monitoring object identifier is first generated in the planar map view, and then an object pickup image carrying the display entity identifier is generated in the 3D scene view. A cross-view recognition mapping result from the display entity identifier to the monitoring object identifier is established through spatial location attribution, ensuring that selected pixels in any view can be uniformly read back to the monitoring object identifier. This allows the display area of ​​the same monitoring object to be located in the target view, and a details call identifier to be output. Furthermore, to address inconsistencies between highlighted areas and details caused by occlusion, proximity, or partial overlap, the pixel re-retrieval results within the highlighted area of ​​the target view are used for consistency verification. This is combined with verification information from the boundary layer and kernel layer for discrimination and correction, thereby achieving consistent processing of cross-view recognition, linked highlighting, and information retrieval within the same technical link.

[0025] The multi-source monitoring data fusion results have provided the identification and spatial location data of the monitored objects. However, the content displayed in the current planar map view still primarily serves human visual observation, and the pixel position itself does not inherently carry object identification information that can be directly retrieved. Here, "displayable objects" refers to the graphic representation of monitored objects that are currently within the viewport and have not been filtered out by layer switches, scale thresholds, or display rules. If view switching or linked selection actions rely solely on display color or symbol appearance, it is easy for the selected position to become disconnected from the object's identity in areas with dense objects. Therefore, the anchoring and mapping module needs to first convert the displayable objects in the current planar map view into object anchoring images, establishing a stable correspondence between pixel positions and monitored object identifications.

[0026] The following steps S101 to S104 are the specific processing steps of the anchoring mapping module during processor execution, which are used to convert the visual display results in the planar map view into object anchoring images that can be read back pixel by pixel to monitor object identification. Through this stage of processing, the anchoring mapping function at the system module level is implemented as an executable input sorting, pixel mapping, attribution resolution and encoding verification process.

[0027] S101: Organizing input objects and view parameters.

[0028] Before proceeding with the pixel writing operation, the monitored objects included in this round of anchoring and mapping processing are first organized into a writing object set. The writing object set refers to the set of monitored objects within the current planar map viewport that can participate in object anchoring image generation, and whose monitoring object identifiers and spatial location data are complete and available. Using the monitoring object identifiers and spatial location data already read from the multi-source monitoring data fusion results, the filtering and constraint organization are completed in this sub-step.

[0029] In the specific processing, the monitored objects that can be displayed in the viewport at the current moment are first filtered from the display results of the current planar map view. Then, a one-to-one correspondence is established between the corresponding monitored object identifiers and spatial location data. The monitored object identifiers use positive integer identifier values, and the background reserved values ​​are reserved separately and do not participate in any monitored object identifier value range. The spatial location data retains the original geometric expression form, allowing it to come from point expression, line expression, and area expression, but it must be unified into a spatial location point list that can participate in the view projection processing before entering the next sub-step. The viewport range of the current planar map view is represented by pixel coordinates. The horizontal coordinate is limited to integer pixel positions within the left and right boundaries, and the vertical coordinate is limited to integer pixel positions within the top and bottom boundaries. The layer display order follows the display stack order in the current planar map view, and the sorting direction remains fixed within the anchoring and mapping module for subsequent pixel assignment conflict resolution.

[0030] S102: Two-stage pixel mapping of spatial location data.

[0031] Before writing the monitored object identifier into the object anchoring image, the spatial location data needs to be converted into the screen pixel position in the current planar map view. If the pixels are written directly from the original spatial location data, changes in view scaling and translation will cause the object anchoring image to be inconsistent with the display position of the current planar map view. Therefore, a two-stage mapping is adopted, first completing the view projection, and then completing the viewport pixelation.

[0032] In the specific processing, the projection transformation used by the current planar map view is first performed on the spatial location point list of each monitored object to uniformly convert the original spatial coordinates into the planar coordinates of the current planar map view. After the projection transformation is completed, scaling and translation processing is performed according to the viewport transformation relationship of the current planar map view to obtain the screen pixel location point list. Each pixel position in the screen pixel location point list needs to be clipped at the boundary. Pixel positions that exceed the viewport range of the current planar map view are directly discarded and do not participate in subsequent writing. If a monitored object has no remaining pixel positions after clipping, it will not be included in the object anchoring image writing set at the current moment.

[0033] To ensure stable selection of point, line, and area objects in the linked positioning module, a unified rasterization process is required after the screen pixel position point list is formed. Point objects generate a center pixel set based on the symbol display center, line objects generate a continuous line pixel set based on the screen line segment direction, and area objects generate a region pixel set based on the screen outline and filled area. After unified rasterization, each monitored object receives a set of screen outline pixels and a set of basic pixels for writing; subsequent construction of selectable pixel regions begins with this result.

[0034] The above processing ensures that the current planar map view display position and the object anchor image writing position use the same set of view semantics. When the linkage positioning module reads the selected pixel in the corresponding image of the current view, the readback result is consistent with the visual position of the current planar map view.

[0035] S103: Selectable pixel region construction and pixel attribution conflict resolution.

[0036] While objects in the current planar map view are visually identifiable, this does not equate to having a sufficiently stable, readable region at the pixel level. A readable region here refers to a pixel area that can still be reliably read back to the same monitored object identifier under conditions of user click errors, view zooming, and panning, according to fixed encoding rules. Especially in densely populated object areas, if only outline pixels or center pixels are written, even a slight shift in the user's focus during view switching or linked selection actions may result in landing on a blank pixel position, making it impossible for the linked positioning module to reliably read back the monitored object identifier. Therefore, before writing the object anchoring image, it is necessary to first construct a selectable pixel region for each monitored object, and then handle pixel assignment conflicts caused by overlapping regions of multiple monitored objects.

[0037] In practice, the selectable pixel area is first expanded outward from the screen outline pixel set of each monitored object. The expansion scale uses a positive integer radius in pixel units, and the expansion radius is determined by choosing the larger of two sources: one is the screen radius converted from the display size of the corresponding symbol of the monitored object in the current planar map view, and the other is a pre-set minimum selectable radius. The former ensures that the selectable area is consistent with the visual display range, while the latter ensures that there is still a readable pixel area when the symbol is small or the line is thin. During expansion, screen pixel distance is used to determine the position of all pixels whose minimum horizontal and vertical step distance from the screen outline pixel set of the monitored object does not exceed the expansion radius and are all included in the selectable pixel area of ​​the corresponding monitored object.

[0038] When multiple monitored objects have overlapping selectable pixel regions, the same pixel location must belong to a unique monitored object; otherwise, the object anchoring image readback result will be uncertain. To avoid mixed writing, a layered comparison lexicographical assignment rule is adopted, that is, the comparison is performed layer by layer according to a pre-fixed order. Once the comparison result of a certain layer can distinguish the object, subsequent comparisons are stopped and a unique assignment is determined. First, the screen anchoring center is calculated for each monitored object. The screen anchoring center is calculated by the average horizontal and vertical pixel positions of the monitored object's screen outline pixel set. The obtained center position is used to represent the screen focus position of the monitored object in the current planar map view. Then, for each overlapping pixel location, the display order of the layers of the monitored objects involved in the overlap is compared first, and the monitored objects with higher display levels in the current planar map view, i.e., higher display priority, are preferentially retained. If the layer display order is the same, the Manhattan distance from the overlapping pixel position to the screen anchor center of each monitored object is compared. The Manhattan distance is defined as the sum of the absolute values ​​of the horizontal pixel difference and the absolute values ​​of the vertical pixel difference. The smaller the distance value, the closer the pixel is to the screen focus position of the monitored object. Therefore, the monitored object with a closer distance is retained first. If the distance is still the same, the order of the monitored object identification is compared, and the monitored object with the earlier order, that is, the smaller sort value, is selected as the unique writing object.

[0039] In one embodiment, the symbol display areas of two monitored objects partially overlap in the current planar map view. When a user clicks on the overlapping area, a highlighted marker appears. The processing action in the anchoring and mapping module does not treat the overlapping area as multiple objects. Instead, it first considers the layer display order, then the distance from the pixel to the screen anchoring center, and finally the order of the monitored object identifiers, thus ensuring that the overlapping area corresponds to only one monitored object identifier. This way, when the selected pixel is read in the linked positioning module, the reading action obtains a unique monitored object identifier, preventing multiple readouts for the same point.

[0040] Once completed, each selectable pixel location in the current planar map view has a defined affiliation, and pixel affiliation will not oscillate when writing to the object anchor image.

[0041] S104: Object anchoring image write encoding and readback verification.

[0042] The previous sub-step has determined which monitored object each selectable pixel location belongs to. The next step is to write the monitored object identifier into the object anchoring image. The writing method needs to meet two requirements: first, the object anchoring image can be read back pixel by pixel within the current planar map view; second, after the linkage positioning module reads the selected pixel, it can restore the monitored object identifier according to fixed rules, without the same identifier having multiple codes or the same code having multiple interpretations.

[0043] In specific processing, firstly, an object anchoring image is created that is consistent with the viewport range of the current planar map view. The image is segmented by color channel to carry the monitoring object identifier. If each pixel position does not belong to any selectable pixel area of ​​any monitoring object, a background reserved value is written. The background reserved value is a pre-reserved encoded value that does not correspond to any monitoring object identifier, such as a value of 0. If each pixel position belongs to a selectable pixel area of ​​a certain monitoring object, the corresponding monitoring object identifier is split and written to the color channel according to a fixed carry relationship. The splitting order remains unchanged. First, the lowest bit segment is written, then the second lowest bit segment is written, and then the higher bit segment is written. Each segment is limited to the range of integer values ​​allowed by a single color channel. When reading, the monitoring object identifier is reassembled in the completely reverse order. The reconstructed monitoring object identifier must be consistent with the monitoring object identifier before writing.

[0044] To prevent inconsistencies between encoding rules and writing operations, a read-back check is performed after writing is complete. The read-back check only checks the non-empty pixel positions in the object anchoring image, recovering the monitored object identifier pixel by pixel according to the read-back rules, and then verifying whether the recovered result belongs to the monitored object identifier set already compiled by the anchoring mapping module. Pixel positions that pass the check are retained, while pixel positions that fail are written back to the background retention value. Simultaneously, the corresponding monitored object identifier and pixel position are recorded for internal review within the anchoring mapping module.

[0045] In one embodiment, a monitored object in the current planar map view occupies a continuous pixel area in the object anchoring image, and the color channel stores the segmented monitoring object identification information. The linkage positioning module randomly selects a selected pixel position from this area for backreading, and after reassembling it according to a fixed carry relationship, obtains the same monitoring object identification. The reading action does not depend on the layer style color or the business panel status.

[0046] Once the anchoring and mapping module is completed, the displayable objects in the current planar map view have formed a readable pixel area in the object anchoring image. The monitoring object identifier is carried on the pixel position in a fixed encoding method. After the overlapping pixel positions are processed by a unified attribution rule, they correspond to only one monitoring object identifier. When the selection action falls on the object anchoring image, the monitoring object identifier can be obtained directly, and the identification entry remains clear.

[0047] The object anchoring image has solved the pixel readback problem in the current planar map view, but the displayed entities in the current stereo scene view are still at the rendering semantic layer. Material, lighting and occlusion relationships will change the visual presentation but will not change the object identity. The relationship between the displayed entity and the monitored object identifier cannot be stably established based solely on the screen color. Therefore, the picking and mapping module needs to form an object picking image outside the rendering result of the current stereo scene view and map the displayed entity identifier to the monitored object identifier to form a cross-view recognition mapping result.

[0048] The following steps S201 to S204 are the specific processing steps of the picking and mapping module during processor execution. They are used to convert the display entities in the stereo scene view into readable display entity identifiers, and further establish cross-view recognition mapping results from display entity identifiers to monitoring object identifiers. Through this stage of processing, the picking and mapping function at the system module level is implemented as a process of input unification, picking imaging, projection filtering, and mapping organization.

[0049] S201: Current stereoscopic scene view input is organized and coordinates are unified.

[0050] First, the fixed monitoring object identifiers and spatial location data from the anchoring and mapping module are integrated into the same processing chain as the displayed entities in the current 3D scene view. This "same processing chain" means that in the same processing round, a unified object index, unified identifier naming, and a unified coordinate reference are used to sequentially complete object picking image writing, pixel set construction, and cross-view recognition mapping result generation. If different naming or indexing methods are used for the same monitoring object in previous steps—for example, using one set of displayed entity numbers in the object picking image but another set of misaligned monitoring object indexes in the spatial location attribution determination—it will cause the object picking image readback results and cross-view recognition mapping results to not correspond one-to-one, resulting in naming inconsistencies and affecting subsequent lookups.

[0051] The monitoring object identifiers and spatial location data already read by the anchoring mapping module are reused without repeating the reading path. The monitoring object identifiers retain the positive integer identifier value representation. Background values ​​are not included in the monitoring object identifier value range. The spatial location data retains the original point column format and uses the same monitoring object index naming from the anchoring mapping module. The display entity set and display entity identifiers are read from the current stereo scene view rendering result. The display entity identifiers retain the positive integer identifier value representation. The background values ​​are separated from the display entity identifier value range. The viewport width and viewport height of the current stereo scene view are represented in pixels, and the pixel coordinate range is limited to within the viewport boundary.

[0052] Before spatial location data is included in subsequent projection, a unified transformation from geographic coordinates to stereoscopic scene coordinates is performed. This transformation uses the coordinate transformation rules corresponding to the current stereoscopic scene view. The coordinate transformation rules must include at least the following: using coordinates consistent with the current stereoscopic scene view... Figure 1 Establish consistent geographic and projection references, converting latitude and longitude or planar coordinates into scene world coordinates; adopt a viewpoint consistent with the current stereoscopic scene. Figure 1The established elevation datum converts elevations into scene height values. Translation, rotation, and scaling parameters of the current scene are applied to the conversion result, ensuring that the converted spatial location data and the geometric location of the displayed entities are in the same scene coordinate system. The conversion result is uniformly represented as a series of 3D scene coordinate points, using length units. After conversion, the geometric location and spatial location data of the displayed entities share the same scene coordinate semantics in the current 3D scene view, and subsequent projection coverage and depth range determinations use the same coordinate base.

[0053] S202: Object picking image generation and display entity identifier writing.

[0054] The material color and lighting changes in the current stereoscopic scene view rendering results are not suitable as a basis for recognition. Therefore, it is necessary to output the object picking image and depth buffer separately to ensure that the pixel re-picking action of the verification and correction module in the highlighted area of ​​the target view can stably read back and display the entity identification.

[0055] The off-screen rasterization rendering algorithm is used during processing. The off-screen rendering targets include the object pick image and the depth buffer. During off-screen rasterization rendering, material colors are not written, only the display entity identifier code colors are written. The pixel coordinate range is kept consistent with the current stereo scene viewport. Any fragments that exceed the viewport boundary do not enter the object pick image writing range.

[0056] For each pixel location, first collect the display entity candidates covering the current pixel location, then read the normalized depth value corresponding to each display entity candidate at the current pixel location. The normalized depth value is represented using a closed interval from zero to one. Then, determine the foreground display entity according to the depth-first rule. In the case of the same depth, use the display entity identifier order to eliminate parallel entities, and the display entity with the earlier order is taken as the foreground display entity. This order remains unchanged to avoid the edge position readback results from fluctuating between different frames.

[0057] Once the foreground entity is determined, the entity identifier is written into the object pickup image according to a fixed channel segmentation encoding rule. The encoding order is fixed: the lower-order segments are written first, followed by the higher-order segments. Each segment is limited to the integer range allowed by a single color channel. The depth buffer synchronously writes the normalized depth value corresponding to the foreground entity at the same pixel position. Background pixels are written with background retention values ​​and background depth retention values. When the encoding capacity of the entity identifier is insufficient, the current entity does not participate in the writing of the object pickup image and is recorded in the internal abnormal record set of the pickup mapping module. The abnormal record set is not included in the cross-view recognition mapping result.

[0058] In one embodiment, two device models in the current stereoscopic scene view partially overlap on the screen, and their material colors are similar, making them easy to confuse when observed manually. During off-screen rasterization rendering, the object picks up the image and writes the display entity identifier color, and the depth buffer writes the foreground depth. The pixels in the overlapping area only retain the foreground display entity identifier, and the readback action obtains a unique display entity identifier. The similar colors will not affect the recognition.

[0059] After the object image is written, a readback check is performed. The readback check restores the display entity identifier pixel by pixel according to the fixed segmented encoding and reassembly rules. The restoration result is only retained within the display entity identifier set, and the background retention value remains unchanged. This check action enables the review and correction module to read back the display entity identifier from the object image without relying on rendering style information.

[0060] S203: Construction of projection coverage of monitored objects and generation of candidates for spatial location attribution.

[0061] Relying solely on pixel proximity in an image is insufficient to distinguish adjacent objects. Monitoring the projection coverage and depth range of the object's spatial location data within the current stereoscopic scene view requires simultaneous filtering to ensure the mapping results are interpretable.

[0062] During processing, the coordinate points of the 3D scene are first projected point by point onto the current 3D scene view. The projection result includes the screen pixel position and the normalized depth value. The screen pixel position is represented in pixel units, and the normalized depth value is represented in a closed interval from zero to one. Projection points that exceed the viewport boundary are directly discarded and do not enter the subsequent rasterization set. After projection, the screen pixel position of the same monitored object is rasterized. The rasterization algorithm uses a combination of scanline rasterization and line segment discretization. Point objects form a central covering pixel set, line objects form a continuous line covering pixel set, and area objects form a filled covering pixel set. The merged result is uniformly recorded as the monitored object projection covering pixel set.

[0063] The normalized depth value point series of the same monitoring object is used to form the depth interval of the monitoring object. The lower bound of the depth interval is the minimum normalized depth value in the projection point series, and the upper bound of the depth interval is the maximum normalized depth value in the projection point series. Then, the boundary of the depth interval is expanded according to the depth buffer quantization step size. The expansion direction is one quantization step size towards the foreground from the lower bound and one quantization step size towards the background from the upper bound. After the boundary is expanded, it is still limited to a closed interval between zero and one. This process can cover the depth discrepancy difference caused by projection sampling and rasterization boundary, and reduce the situation where support pixels are mistakenly excluded.

[0064] The image readback results of the object picking are used to form the display entity pixel set. The processing method is to traverse all non-background pixels of the object picking image, and group pixels with the same readback display entity identifier into the same display entity pixel set. Then, a support pixel set is constructed for each display entity and each monitored object. The support pixel set must satisfy three conditions at the same time. The first condition is that the pixel is located in the display entity pixel set. The second condition is that the pixel is located in the monitoring object projection coverage pixel set. The third condition is that the depth value corresponding to the depth buffer falls within the depth range of the monitoring object. The pixel position that satisfies all three conditions is entered into the support pixel set.

[0065] Once completed, the display entity pixel set, the monitored object projection coverage pixel set, the monitored object depth range, and the supporting pixel set all have clear sources, and the cross-view recognition mapping result generation action has stable input.

[0066] S204: Generation and organization of cross-view recognition mapping results.

[0067] The linkage positioning module needs to locate the display area in the target view according to the monitoring object identifier. The verification and correction module needs to pick up the image of the object, read back the displayed entity identifier, and then search for the monitoring object identifier. Therefore, the cross-view recognition and mapping results must ensure that the displayed entity identifier is unique and that the search path is clear.

[0068] During processing, the display entity support ratio is first calculated for each display entity and each monitored object. The calculation order of the display entity support ratio is written as follows: first count the number of pixels in the supporting pixel set, then count the number of pixels in the display entity pixel set, and then divide the number of supporting pixels by the number of display entity pixels to obtain the display entity support ratio. When the display entity pixel set is empty, the display entity support ratio is not calculated and the current display entity is skipped directly. The display entity support ratio takes the value of a closed interval from zero to one.

[0069] The monitoring object coverage ratio is then calculated. The calculation order for the monitoring object coverage ratio is as follows: First, count the number of pixels in the supporting pixel set. Then, count the number of pixels in the monitoring object projection coverage pixel set. Finally, divide the number of supporting pixels by the number of monitoring object projection coverage pixels to obtain the monitoring object coverage ratio. If the monitoring object projection coverage pixel set is empty, the monitoring object coverage ratio is not calculated and the current monitoring object is skipped. The monitoring object coverage ratio takes the value of a closed interval from zero to one.

[0070] Next, the screen distance is calculated. The calculation order for the screen distance is as follows: First, calculate the average horizontal and vertical pixel positions of the set of pixels for the displayed entity to obtain the center of the displayed entity pixels. Then, calculate the average horizontal and vertical pixel positions of the set of pixels covered by the projection of the monitored object to obtain the center of the pixels covered by the projection of the monitored object. Finally, calculate the sum of the absolute values ​​of the horizontal and vertical pixel position differences to obtain the screen distance. The screen distance is calculated in pixels and takes a non-negative value.

[0071] The unique identification method uses a hierarchical comparison logic with a fixed comparison order. First, the display entity support ratio is compared, and the monitoring object with the higher display entity support ratio is given priority. Then, the monitoring object coverage ratio is compared, and the monitoring object with the higher coverage ratio is given priority. Next, the screen distance is compared, and the monitoring object with the shorter screen distance is given priority. If there are still ties, the order of the monitoring object identifiers is compared, and the monitoring object with the earlier order is given priority. In this way, a unique monitoring object identifier is determined for each display entity, forming a mapping record from the display entity identifier to the monitoring object identifier.

[0072] After the mapping records are generated, mapping sorting is performed. Mapping sorting includes three parts: background values ​​are not included in the mapping records, empty pixel display entities are not included in the mapping records, and only one mapping record is retained for the same display entity identifier. At the same time, a mapping index table is established according to the monitoring object identifier. The mapping index table stores the set of display entity identifiers belonging to the same monitoring object identifier. When the linkage positioning module looks up the set of display entity identifiers according to the monitoring object identifier, it directly uses the mapping index table. When the review and correction module reads back the monitoring object identifier according to the display entity identifier, it directly uses the mapping record.

[0073] After the above processing, the cross-view recognition mapping results simultaneously satisfy both the display entity identifier lookup and the monitoring object identifier retrieval usage paths.

[0074] After the picking and mapping module is completed, the displayed entities in the current stereo scene view have the ability to read back pixel by pixel in the object picking image. The displayed entity identifier is mapped to the monitored object identifier through the spatial location belonging relationship. The cross-view recognition mapping results and mapping index table are sorted out. The correspondence between the displayed entity side and the monitored object identifier side has been fixed. The image recognition and object identity no longer depend on visual texture judgment.

[0075] The object anchoring image and cross-view recognition mapping results have respectively established the object identity back-reading link between the planar map view and the stereo scene view. However, when the view switching or linkage selection action enters the execution stage, it is still necessary to converge the selected pixels to the monitored object identifier and then convert the monitored object identifier into a specific display area in the target view. Otherwise, there will be a situation where the object identity is correct but the highlight area of ​​the target view is drifted. Therefore, the linkage positioning module needs to complete the target view positioning around the monitored object identifier and generate a detail call identifier at the same time.

[0076] The following steps S301 to S304 are the specific processing steps of the linkage positioning module during processor execution. They are used to uniformly read back the selection result in any view as the monitoring object identifier, and locate the highlighted area and output the details call identifier in the target view. Through this stage of processing, the linkage positioning function at the system module level is implemented as a process of selection parsing, identifier reading back, area positioning and synchronous output.

[0077] S301: Select operation parsing and determine the image corresponding to the current view.

[0078] The view switching or linked selection operations are organized into an input structure that the linked positioning module can directly process. Linked selection operations refer to the user clicking, selecting, hovering, or using program commands to select a pixel location or object display area in the current 2D map view or the current 3D scene view, and requesting another view to simultaneously locate the same monitored object. If the processing action does not first fix the corresponding image in the current view, the selected pixel readback will result in a mix of the object's anchor image and the object's picked image, causing inconsistencies in the source of the monitored object's identifier.

[0079] The current view parameter, target view parameter, and selected pixel position are carried over in the view switching or linked selection operation. The current view parameter is used to indicate the view that initiates the selection operation, and the target view parameter is used to indicate the view that needs to perform linked positioning. The current view parameter can be either the current planar map view or the current stereo scene view, and the target view parameter can be either the current planar map view or the current stereo scene view. The selected pixel position is limited to the pixel range of the current view's viewport. The horizontal pixel position uses the width range of the current viewport, and the vertical pixel position uses the height range of the current viewport.

[0080] When the current view parameter indicates the current planar map view, the corresponding image is fixed as the object anchoring image; when the current view parameter indicates the current stereoscopic scene view, the corresponding image is fixed as the object picking image. In this implementation, the target view parameter indicating the current stereoscopic scene view is used as the processing path. The object picking image, depth buffer, cross-view recognition mapping result, mapping index table, monitored object projection coverage pixel set, and monitored object depth range output by the picking and mapping module are all included in this step's positioning process. When the target view parameter indicates the current planar map view, the selected pixel readback action remains unchanged, and the target view positioning action is changed to the pixel region retrieval path in the object anchoring image.

[0081] When the selected pixel location exceeds the current viewport range, the highlighted area of ​​the target view is written with an empty area reserved value, and the details call identifier is written with an empty identifier reserved value. The details call identifier is the monitoring object identifier used to call the monitoring object details panel, alarm association information, or trend information interface. Therefore, writing it with an empty identifier reserved value can prevent invalid pixel locations from triggering erroneous readbacks.

[0082] S302: Selected pixel readback and monitoring object identification confirmation.

[0083] All positioning judgments and output actions in the linkage positioning module revolve around the monitored object identifier. Selected pixel readback must first convert the selected pixel position into the monitored object identifier before entering the mapping index table lookup path. The mapping index table is a retrieval structure that uses the monitored object identifier as the key and the set of display entity identifiers belonging to that monitored object identifier as the value.

[0084] When the current view corresponds to an object anchoring image, the monitoring object identifier is restored according to the object anchoring image segmented encoding back-reading rules fixed by the anchoring image module. The restoration process adopts a fixed order: first, the low-order channel encoding value of the object anchoring image at the selected pixel position is read; then, the next lowest-order channel encoding value is read; then, the high-order channel encoding value is read; and finally, the monitoring object identifier is restored according to the fixed carry-back stitching order of the anchoring image module. The range of channel values ​​of the object anchoring image remains within the single-channel encoding integer range defined by the anchoring image module, and the range of values ​​of the restored monitoring object identifier remains within the set of monitoring object identifiers.

[0085] When the image corresponding to the current view is an object picking image, the entity identifier is restored and displayed according to the object picking image segmentation encoding back read rules that have been fixed in the picking mapping module. The restoration process adopts the channel reading order and reassembly order that are completely consistent with the picking mapping module. After the entity identifier is restored, the corresponding monitoring object identifier is found by using the cross-view recognition mapping results. The search action only accepts one mapping record. If the mapping record does not exist, it is considered an invalid selection.

[0086] When the object anchoring image readback result or the object picking image readback result equals the background retention value, the highlighted area of ​​the target view is written with an empty area retention value, and the details call identifier is written with an empty identifier retention value, ending the current processing. After obtaining the monitored object identifier through readback, a unique index positioning is performed in the monitored object identifier set. The index positioning action searches for a unique monitored object index according to the monitored object identifier equality rule. If the unique monitored object index exists, proceed to the next sub-step; if the unique monitored object index does not exist, write an empty area retention value and an empty identifier retention value.

[0087] In one embodiment, a pump station monitoring point symbol is already displayed in the effective encoding area of ​​the object anchoring image in the current planar map view. After selecting the operation landing point to enter the object anchoring image encoding pixel, the readback action directly restores the monitoring object identifier. The monitoring object index is then uniquely located by the monitoring object identifier set. The target view positioning process does not need to retrieve the equipment name or business category field again.

[0088] This step unifies the selection action to the monitoring object identifier, and the aforementioned mapping index table can then directly retrieve the corresponding set of display entity identifiers by the monitoring object identifier.

[0089] S303: Construction of candidate pixel set for target view and positioning of main display area.

[0090] This sub-step is responsible for the target view positioning action. The highlighted area of ​​the target view cannot be directly retrieved by simply taking the pixels from the set of display entity identifiers returned by the mapping index table. Otherwise, the attached display entities and occluded areas of the same monitored object will enter the highlighted area at the same time. When the verification and correction module retrieves the pixels, the boundary error is likely to increase. Constraints need to be added to the set of pixels covered by the projection of the monitored object and the depth range of the monitored object.

[0091] First, the mapping index table is used to query the set of display entity identifiers based on the monitored object identifier. If the query result is empty, the empty area is reserved in the highlighted area of ​​the target view. The detailed call identifier is still written according to the monitored object identifier, and the current processing ends. If the query result is not empty, all pixel positions of the object-picked image are traversed. For each pixel position, the object-picked image is read back. If the readback display entity identifier belongs to the queryed set of display entity identifiers, the current pixel position is written to the candidate pixel set of the target view.

[0092] After the target view candidate pixel set is formed, joint constraint filtering is performed using the monitored object projection coverage pixel set output by the picking and mapping module and the monitored object depth range. The filtering action uses three conditions that must be met simultaneously: the pixel position must be within the target view candidate pixel set, the pixel position must be within the monitored object projection coverage pixel set, and the normalized depth value of the current pixel position in the depth buffer must fall within the monitored object depth range. Pixel positions that meet all three conditions are added to the constraint candidate pixel set. When the constraint candidate pixel set is empty, the positioning input set is reverted to the target view candidate pixel set. When the target view candidate pixel set is empty, the highlighted area of ​​the target view is written with the empty area retention value, the details call identifier is written with the monitored object identifier, and the linkage positioning module ends the current processing.

[0093] The main display area positioning uses an eight-neighbor connected component labeling algorithm. The algorithm operates in a fixed order: first, it performs eight-neighbor connected component decomposition on the positioning input set; then, it counts the number of pixels in each connected component and the number of pixels overlapping the projection coverage of the monitored object; finally, it calculates the screen distance between the center position of the connected component and the center of the projected pixel coverage of the monitored object. The center position of the connected component is formed by the average of the horizontal and vertical positions of all pixels in the connected component. The center position of the projected pixel coverage of the monitored object is the same as the center position already generated by the picking and mapping module. The screen distance is calculated by summing the absolute values ​​of the horizontal and vertical position differences, using pixels and taking non-negative values.

[0094] The main display area selection adopts a hierarchical comparison order. First, the number of overlapping pixels in the projection coverage of the monitored object is compared, and connected components with a higher number of overlapping pixels are given priority. Then, the screen distance is compared, and connected components with a shorter screen distance are given priority. If there are still ties, the number of pixels in the connected components is compared, and connected components with more pixels are given priority. When all connected components are empty, the highlighted area of ​​the target view is written with the reserved value of the empty area.

[0095] In one embodiment, a monitored object in the current stereo scene view corresponds to a main model and a subordinate model. After querying the mapping index table, multiple display entity identifiers are obtained. The target view candidate pixel set formed by traversing the object picking image contains the main region and the subordinate region. After adding the monitored object projection coverage pixel set and the monitored object depth interval to the filter, the main position region is retained as the main connected component of the constrained candidate pixel set. The main display area selection result falls in the main region, and the subordinate region will not occupy the highlight area.

[0096] After this step is completed, the highlighted area of ​​the target view has formed a stable pixel set representation, and the boundary ring and kernel region in the verification and correction module can be constructed around the same highlighted area of ​​the target view.

[0097] S304: Target view highlight area generation and details call identifier output.

[0098] The highlighted area of ​​the target view and the details call identifier need to be generated and confirmed together in the same process; otherwise, inconsistencies will be introduced at the input when the review and correction module retrieves and compares them.

[0099] When the main display area exists, the highlighted area of ​​the target view is directly represented by the pixel set of the main display area. Simultaneously, the highlighted boundary pixel set and the highlighted fill pixel set of the target view are extracted from the main display area. The highlighted boundary pixel set of the target view is extracted using an eight-neighbor boundary tracking algorithm. The boundary tracking starting point is the first pixel position in the main display area where the boundary condition is met. Boundary tracking scans in a fixed direction sequence until returning to the starting point. The highlighted fill pixel set of the target view is directly equal to the main display area. When the main display area is empty, both the highlighted boundary pixel set and the highlighted fill pixel set of the target view are written to the empty area's reserved value.

[0100] The detailed call identifier is directly written to the monitoring object identifier determined by the selected pixel readback. The value range of the detailed call identifier is consistent with the monitoring object identifier set. Consistency verification is performed before output. This verification includes checking that the highlighted area of ​​the target view is not empty and checking that the detailed call identifier belongs to the monitoring object identifier set. If both checks pass, the highlighted area of ​​the target view and the detailed call identifier are output. If the highlighted area of ​​the target view is empty and the detailed call identifier is a monitoring object identifier, the detailed call identifier is still retained for output, used to verify whether the missing highlighted area or the missing identifier is due to a problem with the correction module. If the detailed call identifier does not belong to the monitoring object identifier set, the detailed call identifier is written with an empty identifier retention value.

[0101] In one embodiment, the main area in the current stereoscopic scene view has been selected as the main display area, the boundary tracking action generates a closed boundary, the filling pixel set maintains all pixels inside the main area, the details call identifier directly uses the monitoring object identifier determined in the previous sub-step, and the display area and details call identifier are consistent in name and source.

[0102] The target view highlight area and details call identifier output by the linkage positioning module are both generated by the same monitoring object identifier link. When the verification and correction module performs pixel re-extraction verification, it can directly compare the re-extraction result with the details call identifier.

[0103] After the linkage positioning module is completed, the selected pixel has been read back as the monitoring object identifier through the object anchoring image or object picking image, and is located to the main display area in the target view through the cross-view recognition mapping result. The highlighted area of ​​the target view and the detail call identifier are output synchronously on the same monitoring object identifier link, and the pixel area expression and the identifier expression form a one-to-one correspondence.

[0104] The highlighted area of ​​the target view and the detailed call identifier have been generated by the same monitoring object identifier link. However, occlusion, boundary adjacency, and partial overlap of displayed entities in the current stereo scene view may still introduce inconsistencies between the pixel layer and the identifier layer. Directly attributing all inconsistencies to highlighted area errors will amplify the reconstruction process, while directly attributing all inconsistencies to detailed call identifier errors will retain the offset area. Therefore, the review and correction module needs to perform pixel re-examination verification within the highlighted area of ​​the target view and calculate the boundary layer evidence and kernel layer evidence separately before making a correction decision.

[0105] The following steps S401 to S405 are the specific processing steps of the verification and correction module during processor execution. They are used to perform pixel re-examination verification, calculate boundary layer and kernel layer evidence and make correction decisions within the highlighted area of ​​the target view. Through this stage of processing, the verification and correction function at the system module level is implemented as the process of fixed entry point, pixel re-examination, evidence calculation, model discrimination and decision execution.

[0106] S401: Establishment of input sorting and verification entry point.

[0107] Before entering the processing stage, the highlighted area of ​​the target view and the detailed call identifier have been synchronously output by the linkage positioning module. The pixel area and the identifier area are within the same monitoring object identifier link. The review and correction module needs to fix the re-extraction judgment entry first to avoid continuing to perform boundary ring and kernel area calculations under empty area or invalid identifier conditions.

[0108] The system retains the object-picked image, depth buffer, and cross-view recognition mapping results output by the picking and mapping module, as well as the target view highlighted area, details call identifier, monitored object index, and monitored object identifier output by the linkage positioning module, without repeating the preceding acquisition process. The mapping cache uses a fixed name within the review and correction module, and stores the most recently verified consistent mapping. The initial state of the mapping cache is written to an empty cache state when the current session first enters the review and correction module.

[0109] When the highlighted area of ​​the target view is an empty area and the value is retained, pixel-by-pixel re-extraction is not performed, boundary ring construction is not performed, kernel region construction is not performed, the verification status is written as the highlighted area missing status, the details call identifier retains the output value of the linkage positioning module, the mapping cache retains the original state, and the current processing ends.

[0110] When the highlighted area of ​​the target view is a valid set of pixels, the repeated pixel set is directly taken from the highlighted area of ​​the target view. The pixel positions in the repeated pixel set are limited to the pixel coordinate domain of the current stereoscopic scene view, and all subsequent repeated statistics are performed around the repeated pixel set. This process ensures that the output area of ​​the linkage positioning module is completely consistent with the repeated area of ​​the verification and correction module, and there will be no deviation caused by resampling.

[0111] S402: Pixel-by-pixel repeat and generation of the main monitoring object identifier.

[0112] The actual comparison object of the review and correction module is not the original pixel color, but the distribution of the monitoring object identifier obtained after re-extraction. Only by reading back the pixels in the highlighted area of ​​the target view to the monitoring object identifier one by one can the judgment of the boundary layer and the kernel layer fall on the same semantic level.

[0113] The processing method combines pixel-by-pixel readback with identifier statistics. First, it iterates through each pixel position in the retrieved pixel set, recovers the displayed entity identifier according to the object picking image encoding readback rules fixed by the picking and mapping module, and then uses the cross-view recognition mapping results to find the monitoring object identifier corresponding to the displayed entity identifier. If a mapping record exists in the cross-view recognition mapping results, the corresponding monitoring object identifier is written to the current pixel position; if no mapping record exists in the cross-view recognition mapping results, an empty identifier reserve value is written to the current pixel position.

[0114] Subsequently, the valid set of repeated pixels is selected from the set of repeated pixels. The valid set of repeated pixels only retains the pixel positions where the monitored object identifier is not equal to the null identifier retention value. When the valid set of repeated pixels is empty, the main monitored object identifier is repeated to the null identifier retention value, the verification status is written as the repeated no valid identifier status, the verification and correction module jumps to the subsequent correction decision sub-step and follows the null identifier path, and no longer performs the normal calculation of the boundary inverse mapping fracture zone ratio and the kernel same object readback depth ratio.

[0115] When the valid set of repeated pixels is not empty, frequency statistics are performed on each monitoring object identifier within the valid set of repeated pixels. Frequency statistics use a group counting method, grouping pixels with the same monitoring object identifier into the same group, and then counting the number of pixels in each group. The monitoring object identifier with the most pixels is selected as the first candidate primary monitoring object identifier for repeated detection. When frequencies are tied, tie resolution is initiated. Tie resolution uses the center distance rule of the highlighted area of ​​the target view. First, the center of the pixel in the highlighted area of ​​the target view is calculated, and then the average Manhattan distance from the repeated pixel group corresponding to each tied monitoring object identifier to the center of the pixel in the highlighted area of ​​the target view is calculated. The monitoring object identifier with the shorter average Manhattan distance is selected as the primary monitoring object identifier for repeated detection. If there are still ties, a unique result is fixed according to the order of the monitoring object identifiers.

[0116] In one embodiment, the highlighted area of ​​the target view covers the main pipeline and the adjacent valve. After pixel-by-pixel re-extraction, the monitoring object identifier corresponding to the main pipeline occupies the main body in the effective re-extracted pixel set, while the monitoring object identifier corresponding to the valve only appears at the edge. Frequency statistics directly determine the monitoring object identifier corresponding to the main pipeline as the main monitoring object identifier for re-extraction, and subsequent comparison operations revolve around the main object.

[0117] After the main monitoring object identifier is generated, the verification and correction module already has the pixel-side main identifier result. Subsequent sub-steps can compare the detailed call identifier with the pixel-side result in layers.

[0118] S403: Calculation of the ratio between boundary ring construction and boundary inverse mapping fracture zone.

[0119] Boundary positions are most susceptible to occlusion by adjacent components and expansion errors in highlighted areas. Simply comparing the main monitoring object identifier with the details call identifier as a whole will obscure the details of edge offset. Calculating the boundary ring separately can separate the semantic offset of the edge from the main area.

[0120] The processing involves a combination of four steps: boundary extraction, boundary ring construction, broken pixel statistics, and broken segment decomposition. First, a set of boundary pixels is extracted from the highlighted area of ​​the target view according to an eight-neighbor boundary determination rule. This rule is fixed: if a pixel position within the highlighted area of ​​the target view has at least one pixel position outside its eight-neighbor range, the current pixel position is written into the boundary pixel set. Then, a boundary ring is constructed around the boundary pixel set within the highlighted area of ​​the target view. This boundary ring only retains pixel positions within the highlighted area of ​​the target view that have a one-pixel eight-neighbor contact relationship with the boundary pixel set.

[0121] When the boundary ring is empty, the ratio of the boundary inverse mapping fracture zone is written as a zero-value retention value, and the fracture segment set is written as an empty set. When the boundary ring is not empty, each pixel position in the boundary ring is traversed, and the pixel re-extraction monitoring object identifier generated in the previous sub-step is read. If the pixel re-extraction monitoring object identifier is inconsistent with the detail call identifier, the current pixel position is written into the boundary inverse mapping fracture pixel set. The boundary inverse mapping fracture pixel set is then decomposed into multiple fracture segment sets according to the eight-neighbor connectivity rule to avoid mixing scattered discrete pixels and continuous fracture regions during statistics.

[0122] The boundary inverse mapping fracture zone ratio is calculated using a range ratio statistical method. First, the total number of pixels in the entire fracture segment set is counted, then the total number of pixels in the boundary ring zone is counted. Finally, the boundary inverse mapping fracture zone ratio is obtained by dividing the total number of fracture pixels by the total number of pixels in the boundary ring zone. The boundary inverse mapping fracture zone ratio is fixed between zero and one. The closer the value is to zero, the stronger the semantic consistency between the boundary position and the monitored object corresponding to the details call identifier. The closer the value is to one, the more obvious the semantic fracture at the boundary position.

[0123] This step provides boundary layer evidence, which, when combined with kernel layer evidence, enables the discrimination model to distinguish between slight edge shifts and overall highlight area shifts.

[0124] S404: Calculation of the ratio of kernel region construction to kernel same-object readback depth.

[0125] The boundary inverse mapping fracture zone ratio can only describe the edge situation. The appearance of a fracture at the edge does not mean that the main body of the highlighted area in the target view is incorrect. Multi-directional continuous back-reading of the kernel region is used to confirm whether the main body area still falls stably on the monitoring object corresponding to the details call identifier. This step can prevent edge noise from directly triggering the reconstruction of the highlighted area.

[0126] The processing involves a combination of four steps: kernel region construction, kernel center determination, multi-directional continuous readback, and weakest direction ratio statistics. First, the kernel region is obtained by removing the boundary ring from the highlighted area of ​​the target view. If the kernel region is not empty, the center of the kernel region is determined by the average horizontal and vertical positions of all pixels in the kernel region. If the kernel region is empty, a degenerate kernel region path is used. The pixel position with the shortest Manhattan distance to the center of the highlighted pixel in the target view is selected as the degenerate kernel seed. The degenerate kernel region contains only this pixel position, and the center of the kernel region uses this pixel position.

[0127] Multi-directional continuous readback employs a fixed eight-directional path. The direction set includes horizontal forward, horizontal reverse, vertical forward, vertical reverse, and four diagonal directions. The step counting rule for each direction is unified: the starting pixel is counted as the first step. This rule is used to eliminate the denominator invalidation problem in single-pixel regions. For each direction, starting from the pixel position at the center of the kernel region, the process advances pixel by pixel along the fixed direction. While still within the highlighted area of ​​the target view, the geometrically reachable steps continue to be counted; the process stops when leaving the highlighted area of ​​the target view, thus obtaining the geometrically reachable depth of the current direction. Subsequently, continuous statistics of the monitored object identifier are performed along the same path. Starting from the starting pixel, the monitored object identifier is read pixel by pixel to retrieve the monitored object identifier. When the retrieved monitored object identifier matches the detailed call identifier, the continuous steps of the same object are continuously accumulated. Accumulation stops when the first discrepancy occurs, thus obtaining the continuous depth of the same object in the current direction. The continuous depth of the same object in each direction will not exceed the geometrically reachable depth of the corresponding direction.

[0128] After completing the statistics in eight directions, the minimum geometrically accessible depth and the minimum continuous depth of the same object are taken in each of the eight directions. Then, the minimum continuous depth of the same object is divided by the minimum geometrically accessible depth to obtain the kernel same object readback depth ratio. Since the step counting rule includes the starting pixel in the first step, the minimum geometrically accessible depth is always positive under the condition of the highlighted area of ​​the effective target view, and the kernel same object readback depth ratio remains between zero and one.

[0129] In one embodiment, the target view's highlighted area is close to adjacent components, a small number of broken pixels appear in the boundary ring, and when the kernel region center advances in eight directions, the main body region still continuously reads back to the monitoring object identifier that is consistent with the details call identifier. The kernel same object readback depth ratio remains at a high level, and the discrimination result will interpret the anomaly as an edge offset rather than an overall error in the main body's highlighted area.

[0130] This step provides evidence at the subject level, and the evidence at the boundary level and the evidence at the subject level then enter the discriminant model's inference path together.

[0131] S405: Discriminant model reasoning and corrective decision execution.

[0132] After the main monitoring object identifier, boundary inverse mapping fracture zone ratio, and kernel same object readback depth ratio have been obtained, the review and correction module needs to organize the three types of information into an executable correction path. The processing goal is not simply to judge consistency or inconsistency, but to place the correction action on the correct side of the detail call identifier or the highlighted area of ​​the target view.

[0133] First, a consistency check is performed. If the highlighted area of ​​the target view is valid and the main monitoring object identifier and the details call identifier are consistent, the verification status is set to consistent. After verification, the highlighted area of ​​the target view retains the output result of the linkage positioning module, and the details call identifier retains the output result of the linkage positioning module. At the same time, the current cross-view recognition mapping result is written to the most recent consistent mapping record in the mapping cache, and the review and correction module ends the current processing.

[0134] When the highlighted area of ​​the target view is empty or the main monitored object identifier is empty and its value is retained, the verification status is set to insufficient retrieval, and the judgment model inference is not entered. At this time, the details call identifier output by the linkage positioning module is maintained, the highlighted area of ​​the target view retains its original value, and the mapping cache is not updated to avoid making erroneous reconstruction actions under insufficient evidence.

[0135] If the initial consistency judgment is not met and the retrieved evidence is valid, the process proceeds to the discriminant model inference. The discriminant model uses a pre-trained model; in one implementation, this pre-trained model is a monotonic constrained discriminant tree model. The model input order is fixed: the first input is the boundary inverse mapping fracture zone ratio, and the second input is the kernel same-object readback depth ratio. This input order remains unchanged after deployment. The pre-trained discriminant model outputs a view same-object credibility coefficient, which represents the numerical credibility of whether the highlighted area of ​​the current target view and the details call identifier correspond to the same monitored object. The coefficient's value is fixed between zero and one. The credibility boundary coefficient is determined on validation samples during the discriminant model training phase and written into the read-only configuration during deployment; it is not dynamically modified during runtime.

[0136] The correction decision adopts a boundary comparison path. When the confidence coefficient of the same object in the view is not lower than the confidence boundary coefficient, the highlighted area of ​​the target view after verification remains the highlighted area of ​​the target view output by the linkage positioning module. After verification, the detailed call identifier is corrected to the identifier of the main monitored object, the verification status is written as the identifier correction status, and the most recent verification consistent mapping record in the mapping cache is updated to the current cross-view recognition mapping result.

[0137] When the confidence coefficient of the same object in the view is lower than the confidence boundary coefficient, the details call identifier after verification remains the details call identifier output by the linkage positioning module, and the highlighted area of ​​the target view enters the reconstruction path. The reconstruction path follows the target view candidate pixel set construction and main display area positioning logic already disclosed by the linkage positioning module, without repeating the process. The only change is that the cross-view recognition mapping result is replaced with the most recently verified consistent mapping record in the mapping cache. When the mapping cache is empty, the reconstruction path is not executed, the verification status is written as no cache reconstruction blocked status, and the highlighted area of ​​the target view remains the output result of the linkage positioning module. When the mapping cache contains the most recently verified consistent mapping record, after generating the highlighted area of ​​the reconstructed target view, a re-retrieval confirmation is performed. The re-retrieval confirmation action follows the pixel-by-pixel re-retrieval and the main monitoring object identifier generation path of the previous sub-steps of this step. When the re-retrieval confirmation result is consistent with the details call identifier, the highlighted area of ​​the target view after verification is replaced with the highlighted area of ​​the reconstructed target view, and the mapping cache retains the current most recently verified consistent mapping record. If the verification results are still inconsistent, the highlighted area of ​​the target view after verification will retain the output result of the linkage positioning module, the details call identifier after verification will retain the output result of the linkage positioning module, the verification status will be written as the verification conflict status, and the mapping cache will not be overwritten.

[0138] In one embodiment, the highlighted area of ​​the target view displays normally around the main pipeline, but the details call panel displays information of adjacent valves. The identifier of the main monitored object is inconsistent with the identifier of the details call. The ratio of the boundary inverse mapping fracture zone remains low, the ratio of the kernel same object readback depth remains high, the view same object confidence coefficient output by the discrimination model falls on the identifier correction path, the highlighted area of ​​the target view remains unchanged, the identifier of the details call is corrected to the identifier of the main monitored object, and the screen area and details information are consistent again.

[0139] Once completed, the verification and correction module provides the highlighted area of ​​the target view after verification and the detailed call identifier after verification, and saves the most recent verification consistent mapping record that can be reused to the mapping cache. The entire correction process has traceability and a stable execution path.

[0140] For example, the construction process of the pre-trained discriminant model can be as follows: A monotonic constrained discriminant tree model is used. The inputs are fixed as the boundary inverse mapping fracture zone ratio and the kernel same-object readback depth ratio obtained from the verification and correction module. The input order remains fixed: the first input is the boundary inverse mapping fracture zone ratio, and the second input is the kernel same-object readback depth ratio. The output is the view same-object confidence coefficient, with the output range limited to 0 to 1. Training samples come from historical linkage selection records and manual verification results. Sample labels are divided into two categories: one category indicates that the target view highlight area is retained and the detail call identifier is corrected; the other category indicates that the detail call identifier is retained and the target view highlight area is reconstructed. Before the samples are stored, they are cleaned by deleting records with failed object picking image readback, missing cross-view recognition mapping results, and empty target view highlight areas. Both input values ​​are uniformly truncated to the range of 0 to 1. Missing values ​​are written with fixed placeholder values ​​and a missing marker is recorded separately for training set screening.

[0141] For example, the training configuration can be as follows: a total of 12,000 samples, including 8,400 in the training set, 2,400 in the validation set, and 1,200 in the test set. Stratified sampling is used based on the monitoring object identifier to avoid the same monitoring object appearing in both the training and test sets simultaneously. Model parameters are set as follows: number of trees 200, maximum depth 4, learning rate 0.05, minimum number of samples per leaf node 30, row sampling ratio 0.8, column sampling ratio 1.0, and random seed 2026. The monotonic constraint direction is fixed as follows: the ratio of the boundary inverse mapping fracture zone decreases monotonically, and the ratio of the kernel same object readback depth increases monotonically, thereby ensuring that the model output aligns with the processing intentions of the verification and correction module. Figure 1 The training process employs an early stopping strategy on the validation set, stopping if there is no improvement in the validation results after 20 consecutive rounds. When the classes are imbalanced, minority class samples are copied to 0.8 times the number of majority class samples.

[0142] For example, model optimization and deployment can be as follows: First, use a grid-based approach to search for candidate combinations within the range of maximum depth 3 to 6, number of trees 100 to 400, and learning rate 0.03 to 0.10. The model is selected based on the combined constraints of the misclassification / reconstruction rate and misclassification / correction rate on the validation set. Then, perform a closed-loop validation on the test set. After confirming stable output, export the model as a fixed version file and embed the confidence threshold coefficient along with the model version into a read-only configuration file; for example, a confidence threshold coefficient of 0.62. During the runtime phase, only forward inference is performed, and model parameters are not updated. When the cumulative number of newly added manually reviewed samples reaches 3000, offline retraining generates a new version model, while the old version is retained for backtracking comparison.

[0143] After the verification and correction module is completed, the pixels in the highlighted area of ​​the target view have been retrieved and read back through the object picking image and cross-view recognition mapping results. The relationship between the main monitoring object identifier and the details call identifier is determined by the boundary inverse mapping fracture zone ratio and the kernel same object readback depth ratio. The view same object confidence coefficient drives the details call identifier correction or the target view highlighted area reconstruction. The output results are consistent on the pixel side and the identifier side.

[0144] Specifically, the above are merely preferred embodiments of this application and are not intended to limit this application.

[0145] The preset parameters can be pre-calibrated through offline simulation testing, or set to fixed values ​​according to the on-site operating procedures.

[0146] In the description of this specification, references to terms such as "an embodiment," "example," and "specific example" indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0147] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to any specific implementation. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A GIS visualization system for multi-source monitoring data fusion, characterized in that, include: Anchoring and mapping module: Reads the monitoring object identification and spatial location data from the multi-source monitoring data fusion results, writes the displayable objects in the planar map view into the object anchoring image, so that the pixel position carries the monitoring object identification; Picking and Mapping Module: Generates object picking images based on the rendering results of the stereo scene view, writes display entity identifiers, and maps the display entity identifiers to the monitoring object identifiers according to the spatial location attribution relationship, forming a cross-view recognition mapping result; Linked positioning module: Responds to view switching or linked selection operations, determines the image corresponding to the current view and reads the selected pixels based on the view switching or linked selection operation. When the image corresponding to the current view is an object anchoring image, it reads back the monitoring object identifier according to the object anchoring image code. When the image corresponding to the current view is an object picking image, it first reads back the displayed entity identifier and then reads back the monitoring object identifier through the cross-view recognition mapping result. It completes unique index positioning in the monitoring object identifier set. Based on the unique index positioning, the monitoring object identifier is positioned and displayed in the target view, generating a highlighted area and a details call identifier. The verification and correction module establishes a set of repeated pixels based on the highlighted area of ​​the target view. It reads back the entity identifier of the picked image and reads back the monitoring object identifier through cross-view recognition mapping results. It filters out the effective set of repeated pixels and determines the main monitoring object identifier to be repeated based on the frequency of the monitoring object identifier and the distance from the center of the highlighted area of ​​the target view. Empty areas or empty identifiers enter the state of insufficient repeated pixels. It extracts the set of boundary pixels around the highlighted area of ​​the target view and constructs a boundary ring. It counts the set of boundary inverse mapping breaks where the pixel repeated monitoring object identifier and the detail call identifier are inconsistent and forms the boundary inverse mapping break ratio as a representation of boundary offset. At the same time, the boundary ring is removed from the highlighted area of ​​the target view to form the kernel region. It continuously reads back the set in a fixed direction to form the kernel same object readback depth ratio as a representation of kernel same object. A preliminary consistency judgment is performed on the retrieved main monitoring object identifier and the details call identifier. When the target view highlight area is valid and the retrieved main monitoring object identifier and the details call identifier are consistent, the target view highlight area and the details call identifier after verification both maintain the output result of the linkage positioning module, and the current cross-view recognition mapping result is written to the most recent consistent mapping record in the mapping cache. When the target view highlight area is empty or the retrieved main monitoring object identifier is an empty identifier retention value, the retrieved insufficient state is entered and the discrimination model inference is not entered. When the target view highlight area is valid, the retrieved main monitoring object identifier is not an empty identifier retention value and the retrieved main monitoring object identifier and the details call identifier are inconsistent, the boundary inverse mapping fracture zone ratio and the kernel same object readback depth ratio are input into the pre-trained discrimination model to generate the view same object credibility coefficient. The view same object credibility coefficient is a numerical credibility degree characterizing whether the current target view highlight area and the details call identifier correspond to the same monitoring object. The value range is fixed between zero and one. The credibility boundary coefficient is determined on the verification samples during the discrimination model training phase and written into the read-only configuration during deployment. It is not dynamically modified during the runtime phase. The correction decision adopts a boundary comparison path. When the confidence coefficient of the same object in the view is not lower than the confidence boundary coefficient, the highlighted area of ​​the target view after verification remains the highlighted area of ​​the target view output by the linkage positioning module. The detailed call identifier after verification is corrected to the main monitoring object identifier. At the same time, the most recent verification consistent mapping record in the mapping cache is updated to the current cross-view recognition mapping result. When the confidence coefficient of the same object in the view is lower than the confidence boundary coefficient, the detailed call identifier after verification remains the detailed call identifier output by the linkage positioning module. The highlighted area of ​​the target view enters the reconstruction path. The reconstruction path replaces the cross-view recognition mapping result with the most recent verification consistent mapping record in the mapping cache. When the mapping cache is empty, path reconstruction is not performed, and the highlighted area of ​​the target view maintains the output of the linkage positioning module. If the mapping cache contains the most recently verified consistent mapping record, a re-fetch confirmation is performed after generating the highlighted area of ​​the reconstructed target view. If the re-fetch confirmation result matches the details call identifier, the highlighted area of ​​the verified target view is replaced with the highlighted area of ​​the reconstructed target view, and the mapping cache retains the most recently verified consistent mapping record. If the re-fetch confirmation result is still inconsistent, the highlighted area of ​​the verified target view retains the output result of the linkage positioning module, and the details call identifier retains the output result of the linkage positioning module. The mapping cache is not overwritten.

2. The GIS visualization system for multi-source monitoring data fusion according to claim 1, characterized in that, The anchoring mapping module includes: Based on the monitoring object identifiers and spatial location data in the multi-source monitoring data fusion results, and combined with the projection relationship and viewport range of the current planar map view, a set of screen outline pixels is generated. The expansion scale is determined according to the symbol display size and minimum selection constraint, and a selectable pixel area corresponding to the monitoring object is formed around the set of screen outline pixels.

3. A GIS visualization system for multi-source monitoring data fusion according to claim 2, characterized in that, The anchoring mapping module also includes: During the object anchoring image writing stage, the overlapping pixel positions are uniquely assigned according to the layer display order, screen anchoring center distance, and monitoring object identification order. Then, the monitoring object identification is written into the object anchoring image according to a fixed channel encoding rule, and a readback check is performed on non-empty pixels to maintain the consistency of the monitoring object identification.

4. A GIS visualization system for multi-source monitoring data fusion according to claim 3, characterized in that, The pick-and-map module includes: Based on the rendering results of the current stereo scene view, an object picking image and depth buffer are generated. Off-screen rasterization rendering is used to write the display entity identifier code and determine the pixel ownership according to the depth first rule. For depth positions with parallel positions, a unique display entity identifier is determined according to the display entity identifier order. Background pixels are written to the background reserved value and the display entity identifier readback verification is performed.

5. A GIS visualization system for multi-source monitoring data fusion according to claim 4, characterized in that, The pick-and-map module also includes: Based on spatial location data, a set of pixels covering the projection of the monitored object and a depth range of the monitored object are formed in the current stereo scene view. A set of supporting pixels is constructed by combining the object picking image readback results. Cross-view recognition mapping results are generated and a mapping index table is established according to the order of display entity support ratio, monitored object coverage ratio, screen distance and monitored object identification.

6. A GIS visualization system for multi-source monitoring data fusion according to claim 5, characterized in that, The linkage positioning module also includes: The set of display entity identifiers corresponding to the monitored object identifier is queried from the mapping index table, and a set of candidate pixels for the target view is constructed. The set of candidate pixels for the projection coverage of the monitored object and the depth range of the monitored object are combined to form a set of constraint candidate pixels. The highlighted area of ​​the target view is determined by eight-neighbor connected component labeling and hierarchical comparison. At the same time, the detailed call identifier consistent with the monitored object identifier is output.

Images