A map tile and range data preloading method based on operation prediction

Abstract

The application discloses a map tile and range data preloading method based on operation prediction. The method acquires a previous operation action, a current operation action, a current field of view range, a current zoom level and a current map application scenario of a user in a map client; forms a continuous operation combination according to the previous operation action and the current operation action, inputs the continuous operation combination and the current map application scenario into a state transition prediction model including a second-order Markov state transition matrix, obtains a candidate next operation action and a transition probability thereof to determine a pre-judged operation action; and then determines a target field of view range according to the pre-judged operation action, the current field of view range and the current zoom level, and generates a set of map tiles to be preloaded and a set of range data to be preloaded. The method can match preloaded objects with a next field of view area that the user is likely to enter and a current application scenario, and reduce the preloading of irrelevant map tiles and range data.

Description

Technical Field

[0001] This invention belongs to the field of geographic information data processing technology, specifically relating to a method for preloading map tiles and extent data based on operational prediction. Background Technology

[0002] In Geographic Information Systems (GIS), online map services, and navigation applications, the loading speed of map tiles and extent data (such as POIs, terrain, road details, etc.) directly determines the user experience. Current map loading technologies suffer from two core deficiencies: Firstly, there is the lag in the passive loading mode. Existing solutions often trigger data loading only after the user completes operations such as zooming and panning, resulting in screen stuttering, frequent blank areas, and a poor user experience in weak network environments or scenarios with large amounts of data.

[0003] Secondly, the preloading technology lacks specificity. Existing related patents have not effectively addressed the issue of behavior prediction in 2D map scenarios: CN105677771A's preloading method based on spatial computational domain similarity matching relies primarily on spatial region similarity analysis and does not consider the temporal patterns of user operations; CN106600523A focuses on tile-based rapid loading of 3D model data, optimizing the segmentation and loading efficiency of 3D data, which is unrelated to 2D map tiles and user behavior prediction; CN113271338A's intelligent preloading for mobile augmented reality scenarios adapts to AR virtual-real fusion requirements but does not address behavior prediction for basic map operations.

[0004] Furthermore, the few preloading solutions that attempt to incorporate user behavior only remain at the level of simple operation sequence statistics, failing to build accurate temporal behavior models and capture the probabilistic patterns of user operations. Therefore, how to more rationally determine the preloading area, preloading data type, and loading priority during map interaction remains a technical direction that urgently needs improvement. Summary of the Invention

[0005] The purpose of this invention is to provide a method for preloading map tiles and range data based on operation prediction, so as to solve the problem that the preloaded area and preloaded data type are not well matched with the user's next operation during continuous interactions such as zooming and panning in existing map clients.

[0006] To achieve the above-mentioned technical objectives, the technical solution adopted by the present invention is as follows: A method for preloading map tile and extent data based on operation prediction includes: The system obtains the user's previous action, current action, current field of view, current zoom level, and current map application scenario in the map client. The previous action and the current action include zooming or translation operations divided by direction. The previous operation and the current operation form a continuous operation combination. The continuous operation combination and the current map application scenario are input into the state transition prediction model to obtain multiple candidate next operation actions and their transition probabilities. Then, the predicted operation action is determined based on each candidate next operation action and its transition probability. The state transition prediction model includes a second-order Markov state transition matrix. The target field of view is determined based on the predicted operation action, the current field of view range, and the current zoom level, and a corresponding set of map tiles to be preloaded and a set of range data to be preloaded are generated.

[0007] As one preferred embodiment, the continuous operation combination is formed when the time interval between the previous operation and the current operation is less than a preset time threshold, or when the two are adjacent operation events in the same continuous interaction process.

[0008] As one of the preferred solutions, the current map application scenarios include route navigation scenarios, point of interest query scenarios, continuous area browsing scenarios, terrain query scenarios, and thematic layer query scenarios. Different current map application scenarios correspond to different state transition probability data and range data types.

[0009] As one preferred approach, when the current operation does not exist or fails, the candidate next operation and its transition probability are determined using the first-order state transition probability data based on the current operation.

[0010] As one preferred option, when at least two candidate next actions have the same transition probability, or the difference in transition probability is less than a preset threshold, the predicted action is determined based on the user's historical operation preferences, route direction, or the location of the target object in the current field of view.

[0011] As one preferred embodiment, the target field of view is determined by translating or expanding the target field of view relative to the current field of view when the predicted operation is a translation operation; and by determining the target field of view based on the current field of view center and the target zoom level when the predicted operation is a zoom operation.

[0012] As one preferred embodiment, the set of range data to be preloaded is determined based on the target field of view and the current map application scenario. The range data includes one or more of the following: road data, point of interest data, terrain data, regional boundary data, route direction data, and thematic layer data.

[0013] As one preferred approach, after generating the set of map tiles to be preloaded and the set of range data to be preloaded, they are compared with local cached data, and the map tiles or range data that are not matched are identified as the data to be requested.

[0014] As one preferred approach, the loading order of the requested data is set based on the data type, its positional relationship with the current field of view, the local cache status, and the current network status.

[0015] As one preferred approach, the state transition probability data in the state transition prediction model is updated based on the matching result between the user's actual next action and the predicted action.

[0016] The advantages and beneficial effects of this invention are as follows: This invention predicts the next action by combining consecutive operations and considering the current map application scenario. The predicted action determines the target field of view, and then generates a corresponding set of map tiles and a set of data for the area to be preloaded. This establishes a correspondence between the preloaded objects and the next field of view the user is likely to enter. Compared to methods that preload only the area surrounding the current field of view or a fixed direction, this method reduces the inclusion of irrelevant areas and data types in the preloaded objects and provides a clearer data foundation for subsequent cache comparisons, network requests, and loading scheduling. Detailed Implementation

[0017] To enable those skilled in the art to better understand the present invention, the technical solution of the present invention will be further described below with reference to specific embodiments.

[0018] This invention provides a method for preloading map tiles and extent data based on operation prediction, applied to a map client. The map client can be a mobile navigation map, an online map application, or a professional geographic information system client. The map content displayed on the map client includes map tiles and extent data related to the current map view. The extent data may include one or more of the following: road data, point of interest data, terrain data, region boundary data, route direction data, and thematic layer data.

[0019] When a user interacts with the map, the map client acquires the user's previous action, current action, current field of view, current zoom level, and current map application scenario. The current field of view can be the geographic coordinate range corresponding to the area currently displayed by the map client, and the current zoom level can be the tile level in the map tile system. The current map application scenario can be determined based on the map client's business mode, currently enabled functions, or the currently displayed layer type. Current map application scenarios include route navigation, point of interest (POI) queries, continuous area browsing, terrain queries, and thematic layer queries, etc. The previous and current actions include zooming operations or translation operations categorized by direction. Zooming operations can include zoom in and zoom out; translation operations can include left translation, right translation, up translation, and down translation. Specifically, scroll wheel zooming, two-finger zooming, and button zooming can be recorded as zooming operations; mouse dragging, touch swiping, and arrow key movement can be recorded as translation operations in the corresponding direction.

[0020] The map client forms continuous operation combinations based on the previous and current actions. A continuous operation combination is defined as follows: if the time interval between the previous and current actions is less than a preset time threshold, or if they are adjacent events within the same continuous interaction process. For example, if a user first performs a zoom-in action and then performs a right pan action within a preset time, a zoom-right pan action combination is formed; if the user performs two consecutive right pan actions, a right pan action-right pan action combination is formed; and if the user performs two consecutive zoom-in actions, a zoom-zoom-zoom action combination is formed. Through these continuous operation combinations, the client can determine the user's field of view changes over a short period by utilizing the correlation between adjacent actions, rather than relying solely on the current action for isolated judgment. The map client inputs the continuous operation combinations and the current map application scenario into a state transition prediction model to obtain each candidate next action and its transition probability. The state transition prediction model includes a second-order Markov state transition matrix, which represents the probability of each candidate next action occurring given that the previous and current actions and the current map application scenario are known. Since the matrix makes predictions based on continuous operation combinations and the current map application scenario, it can distinguish different interaction trends such as right translation after zooming in, continuous right translation, and left translation after zooming out. This makes the prediction results closer to the user's actual field of view direction or zoom level, improves the consistency between the predicted operation and the actual next operation, and makes it easier to understand the operation intention.

[0021] In one implementation, the state transition prediction model can be configured with different second-order Markov state transition matrices or state transition probability data according to different current map application scenarios. The system can directly match the corresponding prediction data according to the current map application scenario, without mechanically interpreting zooming, panning, and other operations as fixed intentions, thereby reducing prediction bias caused by scenario differences.

[0022] For example, in a navigation scenario, the user is currently near a road intersection. Their previous action was zooming in, and their current action is a right pan. The map client generates a continuous action combination of "zoom in - right pan" and compares this combination with the navigation scenario's input state transition prediction model. If, among the model's candidate next actions, the right pan action has the highest transition probability, it is selected as the predicted action. This process allows subsequent preloading to be more focused on the route the user is likely to continue viewing or the area adjacent to the intersection, rather than simply loading the entire area around the current viewpoint.

[0023] If at least two candidate next actions have the same transition probability, or the difference between their transition probabilities is less than a preset threshold, the map client can determine the predicted action based on the user's historical operation preferences, route direction, or the location of the target object within the current field of view. For example, if the transition probabilities of left panning and right panning are the same, and the current navigation route extends to the right, or if the user has recently viewed intersection areas to the right multiple times after zooming in in the same scenario, then right panning can be determined as the predicted action. This method reduces random selection when probabilities are equal, making the predicted action more consistent with the current interaction context.

[0024] After determining the anticipated operation, the map client determines the target field of view based on the anticipated operation, the current field of view, and the current zoom level. When the anticipated operation is a translation, the map client translates or expands the field of view relative to the current field of view according to the translation direction. Since the target field of view is limited by the anticipated operation, the general preloading of map areas in irrelevant directions can be reduced. The translation or expansion range can be determined based on the width or height of the current field of view. For example, an area 1.5 to 2 times the width or height of the current field of view in the translation direction can be used as the target field of view. In a weak network environment, a smaller expansion factor can be used to reduce the amount of data requested at once; in a high-speed network environment, a larger expansion factor can be used to improve the continuity of the image during continuous translation.

[0025] When the predicted operation is a zoom-in operation, the map client determines the zoomed-in target field of view based on the current field of view center and the target zoom level. Specifically, the map client converts the target field of view into a range of tile row and column numbers at the corresponding zoom level, and determines the map tiles within that range as the set of map tiles to be preloaded. Thus, the set of map tiles to be preloaded directly corresponds to the predicted target field of view, avoiding indiscriminate expansion of the preloaded area.

[0026] The map client also generates a set of data to be preloaded based on the target field of view and the current map application scenario. The current map application scenario can include route navigation, point-of-interest (POI) query, continuous area browsing, terrain query, and thematic layer query scenarios. The route navigation scenario can be further subdivided into urban road navigation, highway or intercity navigation, pedestrian navigation, cycling navigation, public transportation navigation, truck navigation, and intersection detail navigation. Accordingly, in the route navigation scenario, the set of data to be preloaded can include road data, route direction data, intersection data, lane data, or traffic restriction data; in the POI query scenario, it can include POI data, address tag data, or surrounding road data; in the continuous area browsing scenario, it can include road data, POI overview data, or background geographic feature data; in the terrain query scenario, it can include terrain data, contour data, slope data, or elevation data; and in the thematic layer query scenario, it can include area boundary data, thematic feature data, layer attribute data, or image data. By determining the range data type according to the application scenario of the subdivided map, the range data to be preloaded can be matched with the actual display requirements within the target field of view, reducing requests for irrelevant range data and data overlay delays.

[0027] In a further implementation, when a previous operation is absent or invalid, the map client can use first-order state transition probability data based on the current operation to determine the candidate next operation and its transition probability. For example, if a user only performs a zoom-in operation once after opening the map, or if the time interval between the previous operation and the current operation exceeds a preset time threshold, a valid continuous operation combination cannot be formed. In this case, initial prediction can be made based on the first-order state transition probability data corresponding to the current operation. This method can supplement second-order prediction, enabling the system to obtain preloaded objects even when a valid preceding operation is lacking.

[0028] In a further implementation, after generating the set of map tiles to be preloaded and the set of range data to be preloaded, the map client can compare them with locally cached data. If the corresponding map tiles or range data already exist in the local cache, they are used as directly accessible data; if the corresponding map tiles or range data do not exist in the local cache, the missing map tiles or range data are identified as data to be requested. This avoids repeatedly requesting cached data and reduces network resource consumption.

[0029] For requested data, the map client can set the loading order based on data type, its positional relationship with the current field of view, local cache status, and current network status. For example, in a panning scenario, map tiles and road data of the area at the forefront of the panning direction are loaded first, followed by background texture data of the remaining area; in a zoom-in scenario, high-resolution tiles, road detail data, and points of interest (POIs) are loaded first, followed by terrain data or background texture data; in a zoom-out scenario, area boundary data and main road data are loaded first, followed by overview POIs or background texture data. By setting this loading order, when network resources are limited, the map content that is most easily perceived by the user in the next step can be prioritized.

[0030] In a further implementation, the map client can update the state transition probability data in the state transition prediction model based on the matching result between the user's actual next action and the predicted action. When the user's actual next action matches the predicted action, the prediction is recorded as a valid prediction, and the transition weight of the predicted action under the corresponding continuous action combination is increased. When the user's actual next action does not match the predicted action, the prediction is recorded as an invalid prediction, and the transition weight corresponding to the predicted action is decreased, or the transition weight corresponding to the actual next action is increased. After the update, the corresponding probability distribution can be normalized, allowing the model to gradually adjust according to changes in user operating habits and application scenarios. The update and optimization cycle can be adaptively adjusted according to the prediction accuracy. For example, when the accuracy is ≥80%, the cycle is 3 days; when the accuracy is 70% ≤ accuracy < 80%, the cycle is 1 day; and when the accuracy is < 70%, the cycle is 12 hours, ensuring the accuracy and timeliness of the probability, adapting to changes in user operating habits, and differing from the existing fixed-cycle update method.

[0031] This invention predicts the next action by combining consecutive operations and considering the current map application scenario. The predicted action determines the target field of view, and then generates a corresponding set of map tiles and a set of data for the area to be preloaded. This establishes a correspondence between the preloaded objects and the next field of view the user is likely to enter. Compared to methods that preload only the area surrounding the current field of view or a fixed direction, this method reduces the inclusion of irrelevant areas and data types in the preloaded objects and provides a clearer data foundation for subsequent cache comparisons, network requests, and loading scheduling.

[0032] Furthermore, different current map application scenarios can be further subdivided into sub-application scenarios and corresponding to different state transition probability data, or to different data types of the range to be preloaded.

[0033] The route navigation scenario can be further subdivided into one or more sub-application scenarios, including urban road navigation, highway or intercity navigation, pedestrian navigation, cycling navigation, electric vehicle navigation, public transportation navigation, truck navigation, and intersection detail navigation. For example, in the urban road navigation scenario, the map client can identify one or more of the following as the pre-loaded range data within the target field of view: road data, intersection data, lane data, traffic signal data, road connectivity data, and point-of-interest data. In this scenario, when a user zooms in and continues to pan, they are more likely to point to adjacent roads, intersections, or turning areas; therefore, the state transition prediction model can use a second-order Markov state transition matrix corresponding to the urban road navigation scenario.

[0034] The continuous area browsing scenario can be further subdivided into sub-application scenarios such as urban area browsing, park browsing, scenic area browsing, and administrative division browsing. In this type of scenario, the data to be preloaded may include road data, point-of-interest overview data, background geographic feature data, administrative division data, and region name data. This scenario is primarily suitable for users who continuously drag, pan, or zoom to view the map.

[0035] The terrain query scenario can be further subdivided into sub-application scenarios such as elevation viewing, contour line viewing, slope viewing, mountain landform viewing, and DEM-related layer viewing. In these sub-application scenarios, the data to be preloaded can include terrain data, elevation data, contour line data, slope data, shaded landform data, and landform annotation data. The Google Map Tiles API provides terrain tiles, and ArcGIS also has elevation layer types, which can support this type of scenario as a classification of map layer usage states.

[0036] The thematic layer query scenarios can be further subdivided into sub-application scenarios such as administrative boundary query scenarios, planned land use query scenarios, pipeline network query scenarios, water system query scenarios, disaster monitoring query scenarios, inspection business query scenarios, engineering layer query scenarios, and remote sensing image query scenarios. In these sub-application scenarios, the data to be preloaded can include regional boundary data, thematic feature data, layer attribute data, image data, vector tile data, and business annotation data. ArcGIS Web layer types include feature, tile, vector tile, map image, imagery, scene, 3D tiles, and elevation, and this layer classification can serve as a practical basis for further subdivision of thematic layer query scenarios.

[0037] By further subdividing the current map application scenario into different sub-application scenarios, the state transition prediction model can match corresponding state transition probability data for different travel modes, query objects, or layer types, making the predicted operation actions more consistent with the current map business status. Correspondingly, the data set to be preloaded can also switch to different data types such as roads, intersections, lanes, points of interest, terrain, elevation, or thematic layers depending on the sub-application scenario, thereby reducing the preloading of irrelevant range data. This processing subjectes the preloaded object to constraints from consecutive operation combinations, sub-application scenarios, and the target field of view, which helps reduce invalid requests and improves the data connection speed during map zooming, panning, or querying.

[0038] The present invention has been described above by way of example. It should be noted that any simple modifications, alterations or other equivalent substitutions that can be made by those skilled in the art without creative effort without departing from the core of the present invention fall within the protection scope of the present invention.

Claims

1. A method for preloading map tile and extent data based on operation prediction, characterized in that, include: The system obtains the user's previous action, current action, current field of view, current zoom level, and current map application scenario in the map client. The previous action and the current action include zooming or translation operations divided by direction. When the time interval between the previous operation and the current operation is less than a preset time threshold, or when the two are adjacent operation events in the same continuous interaction process, a continuous operation combination is formed based on the previous operation and the current operation. The continuous operation combination and the current map application scenario are input into the state transition prediction model to obtain multiple candidate next operation actions and their transition probabilities. The predicted operation action is determined based on each candidate next operation action and its transition probability. The state transition prediction model includes a second-order Markov state transition matrix. Different current map application scenarios correspond to different state transition probability data and range data types. The range data includes one or more of the following: road data, point of interest data, terrain data, regional boundary data, route direction data, and thematic layer data. The target field of view is determined based on the predicted operation action, the current field of view range, and the current zoom level. When the predicted operation action is a translation operation, the target field of view range is determined by translating or expanding relative to the current field of view range according to the translation direction. When the predicted operation is a zoom operation, the target field of view is determined based on the current field of view center and the target zoom level; a set of map tiles to be preloaded and a set of range data to be preloaded are generated based on the target field of view and the current map application scenario, and the set of map tiles to be preloaded and the set of range data to be preloaded are preloaded.

2. The method according to claim 1, characterized in that, When the current operation does not exist or is invalid, the candidate next operation and its transition probability are determined by using the first-order state transition probability data based on the current operation.

3. The method according to claim 1, characterized in that, When at least two candidate next actions have the same transition probability, or the difference in transition probability is less than a preset threshold, the predicted action is determined based on the user's historical operation preferences, route direction, or the location of the target object in the current field of view.

4. The method according to claim 1, characterized in that, The set of map tiles to be preloaded and the set of range data to be preloaded are compared with the local cache data. The map tiles or range data that are not matched are identified as data to be requested. The loading order of the data to be requested is set according to the data type, the positional relationship with the current field of view, the local cache status, and the current network status.

5. The method according to claim 1, characterized in that, The state transition probability data in the state transition prediction model is updated based on the matching result between the user's actual next action and the predicted action.

6. The method according to claim 5, characterized in that, When the user's actual next action is consistent with the predicted action, the prediction is recorded as a valid prediction, and the transfer weight of the predicted action under the corresponding continuous operation combination is increased. When the user's actual next action is inconsistent with the predicted action, the prediction is recorded as invalid, and the transfer weight corresponding to the predicted action is reduced, or the transfer weight corresponding to the actual next action is increased. After the update, the corresponding probability distribution is normalized.