Super-large-scale indoor and outdoor high-precision scene rapid construction method and system

Abstract

The invention belongs to the technical field of scene construction, and provides a super-large-scale indoor and outdoor high-precision scene rapid construction method and system. The method comprises the following steps: dividing a target area into an outdoor area and an indoor area; the method comprises the following steps: acquiring a plurality of groups of remote sensing image data sets of an outdoor area, identifying from the remote sensing image data sets to obtain a first infrastructure data set and suspected infrastructure data sets, performing secondary confirmation on each suspected infrastructure, and updating the first infrastructure data set based on a secondary confirmation result to obtain a second infrastructure data set; a dynamic request mechanism is adopted to request to obtain video data corresponding to the indoor area, and a third infrastructure data set is identified from the video data; and performing coordinate mapping and position alignment on the second infrastructure data set and the third infrastructure data set to obtain a super-large-scale indoor and outdoor high-precision scene. The method can achieve the accurate recognition of indoor and outdoor infrastructures, and achieves the quick creation of a super-large-scale high-precision scene.

Description

Technical Field

[0001] The present invention relates to the technical field of scene construction, and in particular to a method and system for quickly constructing ultra-large-scale indoor and outdoor high-precision scenes. Background Art

[0002] With the rapid development of science and technology, the demand for large-scale high-precision indoor and outdoor scene construction is growing, mainly used in the field of traffic navigation. High-precision scene models are the key to achieving efficient decision-making and precise control of navigation driving. This kind of navigation application includes both conventional vehicle navigation and navigation of various robots.

[0003] In the process of high-precision scene construction, it is necessary to identify and extract various types of infrastructure and their contours, location distribution information, etc. from the images of these indoor and outdoor scenes, so as to construct outdoor high-precision scenes and indoor high-precision scenes, and then align and integrate them to obtain the required ultra-large-scale indoor and outdoor high-precision scenes. At present, outdoor scene construction often relies on images taken by drones or satellite remote sensing, while indoor scene construction mostly uses surveillance cameras to obtain video images. However, there are many problems with the existing construction methods, mainly reflected in: 1) When identifying infrastructure, it is difficult to distinguish temporary equipment that does not belong to infrastructure in the image. For example, in a road scene, a car is carrying a statue and parked on the side of the road (for example, it is parked at the entrance and exit of the main road and the parallel auxiliary road along the road). At this time, it is easy to identify it as part of the road based on remote sensing images, that is, it is determined that there is no entrance and exit between the main road and the auxiliary road, but the area corresponding to the entrance and exit is a statue. These temporary objects similar to infrastructure will seriously interfere with the identification of infrastructure, resulting in many errors in the constructed high-precision scenes.

[0004] 2) In some indoor areas where scene construction is required, the actual boundary distribution of the passable area cannot be accurately identified because the temporary stacking objects in the scene are too messy.

[0005] Therefore, there is an urgent need for a new method that can quickly and accurately construct ultra-large-scale indoor and outdoor high-precision scenes. Summary of the invention

[0006] In response to the above technical problems, the present invention provides a method, system, electronic device, computer storage medium and computer program product for quickly constructing ultra-large-scale indoor and outdoor high-precision scenes.

[0007] The present invention discloses a method for quickly constructing ultra-large-scale indoor and outdoor high-precision scenes. The method comprises the following steps: receiving a target area corresponding to a high-precision scene input by a user, and dividing the target area into an outdoor area and an indoor area based on map information corresponding to the target area; acquiring multiple groups of remote sensing image data sets corresponding to the outdoor area in different acquisition time periods, identifying and obtaining a first infrastructure data set and a suspected infrastructure data set from the multiple groups of remote sensing image data sets, performing secondary confirmation on each suspected infrastructure in the suspected infrastructure data set, and updating the first infrastructure data set based on the secondary confirmation result to obtain a second infrastructure data set; adopting a dynamic request mechanism to request to obtain video data corresponding to the indoor area, and identifying and obtaining a third infrastructure data set from the video data; performing coordinate mapping and position alignment on the second infrastructure data set and the third infrastructure data set to obtain an ultra-large-scale indoor and outdoor high-precision scene.

[0008] Optionally, the identifying of a first infrastructure dataset and a suspected infrastructure dataset from the multiple groups of remote sensing image datasets includes: using a composite recognition model to identify objects in any group of the remote sensing image datasets, obtaining a number of objects and their recognition confidences, identifying the objects with recognition confidences higher than a confidence threshold as infrastructure, and classifying them into the first infrastructure dataset; wherein the composite recognition model is used to identify a number of specified types of infrastructure; and classifying objects with recognition confidences not higher than a confidence threshold into the suspected infrastructure dataset.

[0009] Optionally, the secondary confirmation of each suspected infrastructure in the suspected infrastructure data set includes: determining the recognition confidence of each suspected infrastructure, and determining the first target quantity and target span time according to the recognition confidence; taking the shooting time of the remote sensing image data set corresponding to the recognition confidence as a reference, screening multiple groups of other remote sensing image data sets according to the target span time to obtain the first target quantity, and secondary confirmation of the suspected infrastructure based on the corresponding multiple remote sensing images in each other remote sensing image data set to obtain a secondary confirmation result, and the secondary confirmation result includes confirmation as infrastructure and confirmation as non-infrastructure.

[0010] Optionally, the suspected infrastructure is reconfirmed based on the corresponding multiple remote sensing images in each other remote sensing image data set to obtain a secondary confirmation result, including: determining a number of associated sub-identification models corresponding to the sub-identification model of the suspected infrastructure; each of the associated sub-identification models refers to an infrastructure type whose similarity with the sub-identification model used to identify is higher than a similarity threshold; using each of the associated sub-identification models to separately identify multiple remote sensing images, and fusing the identification results to obtain the secondary confirmation result.

[0011] Optionally, the dynamic request mechanism is used to request the video data corresponding to the indoor area, including: requesting a group of video data corresponding to the indoor area, analyzing the object position change amplitude from the first group of video data, and determining a second target quantity according to the change amplitude level to which the object position change amplitude belongs; and requesting multiple groups of video data corresponding to the second target quantity of the indoor area.

[0012] The present invention also discloses a system for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes, the system comprising a receiving and preprocessing module, an outdoor analysis and recognition module, an indoor analysis and recognition module, and a scene integration generation module; the receiving and preprocessing module receives a target area corresponding to a high-precision scene input by a user, and divides the target area into an outdoor area and an indoor area based on map information corresponding to the target area; the outdoor analysis and recognition module obtains multiple groups of remote sensing image data sets corresponding to the outdoor area in different acquisition time periods, identifies a first infrastructure data set and a suspected infrastructure data set from the multiple groups of remote sensing image data sets, performs secondary confirmation on each suspected infrastructure in the suspected infrastructure data set, updates the first infrastructure data set based on the secondary confirmation result, and obtains a second infrastructure data set; the indoor analysis and recognition module adopts a dynamic request mechanism to request to obtain video data corresponding to the indoor area, and identifies a third infrastructure data set from the video data; the scene integration generation module performs coordinate mapping and position alignment on the second infrastructure data set and the third infrastructure data set to obtain an ultra-large-scale indoor and outdoor high-precision scene.

[0013] Optionally, the outdoor analysis and identification module is used to: use a composite recognition model to identify objects in any group of the remote sensing image data sets, obtain a number of objects and their recognition confidences, identify the objects whose recognition confidences are higher than a confidence threshold as infrastructure, and classify them into the first infrastructure data set; wherein the composite recognition model includes a plurality of sub-recognition models for respectively identifying a number of specified types of infrastructure; and classify the objects whose recognition confidences are not higher than the confidence threshold into a suspected infrastructure data set.

[0014] Optionally, the outdoor analysis and identification module is further used to: determine the identification confidence of each of the suspected infrastructures, and determine the first target quantity and target crossing time according to the level of the identification confidence; based on the shooting time of the remote sensing image data set corresponding to the identification confidence, screen multiple groups of other remote sensing image data sets that obtain the first target quantity according to the target crossing time, and perform secondary confirmation on the suspected infrastructure based on the corresponding multiple remote sensing images in each of the other remote sensing image data sets to obtain a secondary confirmation result, which includes confirmation as infrastructure and confirmation as non-infrastructure.

[0015] Optionally, the outdoor analysis and identification module is also used to: determine several associated sub-identification models corresponding to the sub-identification model of the suspected infrastructure; wherein the similarity between each associated sub-identification model and the type of infrastructure used to identify the sub-identification model is higher than a similarity threshold; use each associated sub-identification model to separately identify multiple remote sensing images, fuse each identification result, and obtain the secondary confirmation result.

[0016] Optionally, the indoor analysis and identification module is used to: request to obtain a group of video data corresponding to the indoor area, analyze the object position change amplitude from the group of video data, and determine the second target quantity according to the change amplitude level to which the object position change amplitude belongs; request to obtain multiple groups of video data corresponding to the second target quantity of the indoor area.

[0017] The present invention also discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement any of the above methods.

[0018] The present invention also discloses a computer storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program is executed by a processor to implement any of the above methods.

[0019] The present invention also discloses a computer program product, which includes computer codes. When the computer codes are executed by a processor of an electronic device, any of the above methods is implemented.

[0020] The above scheme of the present invention performs secondary identification of outdoor targets suspected to be infrastructure, and adopts a dynamic request mechanism to obtain video data of indoor areas, thereby achieving accurate identification of indoor and outdoor infrastructure, and further achieving rapid creation of ultra-large-scale high-precision scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings required for use in the embodiments will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present invention and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 It is a flow chart of a method for quickly constructing ultra-large-scale indoor and outdoor high-precision scenes disclosed in an embodiment of the present invention.

[0023] Figure 2 It is a schematic diagram of the structure of the composite recognition model disclosed in the embodiment of the present invention.

[0024] Figure 3 It is a structural schematic diagram of a system for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes disclosed in an embodiment of the present invention. DETAILED DESCRIPTION

[0025] The following is a description of the implementation of the present application by specific specific embodiments. People familiar with the technology can easily understand other advantages and effects of the present application from the contents disclosed in this specification. Obviously, the described embodiments are part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in the field without creative work are within the scope of protection of the present application.

[0026] In addition, the technical features involved in the different embodiments of the present application described below can be combined with each other as long as they do not conflict with each other.

[0027] In response to the above technical issues, such as Figure 1 As shown, an embodiment of the present invention discloses a method for quickly constructing ultra-large-scale indoor and outdoor high-precision scenes, the method comprising the following steps: S101, receiving a target area corresponding to a high-precision scene input by a user, and dividing the target area into an outdoor area and an indoor area based on map information corresponding to the target area.

[0028] For example, if a user wants to build a high-precision scene of a large commercial district and its surrounding office buildings in a city, the user enters the name or coordinate range of the area on the cloud platform, and the cloud obtains the electronic map information of the area. According to the building outlines, roads and other signs in the electronic map, the open-air parts of the commercial district such as streets and squares are divided into outdoor areas, and the interior of shopping malls and office buildings in the commercial district are divided into indoor areas.

[0029] S102, obtaining a plurality of sets of remote sensing image data sets corresponding to the outdoor area in different acquisition time periods, identifying a first infrastructure data set and a suspected infrastructure data set from the plurality of sets of remote sensing image data sets, performing secondary confirmation on each suspected infrastructure in the suspected infrastructure data set, updating the first infrastructure data set based on the secondary confirmation result, and obtaining a second infrastructure data set.

[0030] For the above-mentioned urban blocks, the cloud obtains remote sensing images of the area taken at different times (taken by satellite or drone), such as images taken during the day on weekdays, during the day on weekends, and at night. Through the image recognition algorithm, roads, buildings, etc. are preliminarily identified from these images to form the first infrastructure data set. However, during the recognition process, some temporary objects (such as a temporary stage with a color close to the road surface, and a large truck temporarily parked on the roadside to load and unload sculptures) may be mistakenly identified as infrastructure. These misjudged objects constitute the suspected infrastructure data set.

[0031] Then, the cloud performs a second confirmation of the suspected infrastructure by analyzing the changes in the position and state of objects in the images at different time periods (for example, the position of a truck changes at different time periods, while the position of the real road infrastructure remains unchanged). Finally, the objects confirmed to belong to the infrastructure in the suspected infrastructure dataset are added to the first infrastructure dataset to obtain a more accurate second infrastructure dataset.

[0032] S103: Requesting to obtain video data corresponding to the indoor area by using a dynamic request mechanism, and identifying a third infrastructure data set from the video data.

[0033] For shopping malls and office buildings in commercial areas, the cloud uses a dynamic request mechanism to communicate with the relevant terminal devices in these indoor areas to request them to provide corresponding video data. From the acquired video data, image recognition technology is used to identify infrastructure such as counters, passages, elevators in shopping malls, and office area divisions, corridors, stairs, etc. in office buildings, forming the third infrastructure data set.

[0034] Due to the adoption of a dynamic request mechanism, different video image acquisition strategies can be decided according to the actual conditions of different indoor areas, thereby achieving high-precision scene construction of indoor scenes.

[0035] S104: coordinate mapping and position alignment are performed on the second infrastructure dataset and the third infrastructure dataset to obtain an ultra-large-scale high-precision indoor and outdoor scene.

[0036] The cloud maps the coordinates of the second infrastructure dataset containing information such as outdoor roads and three-dimensional contours and locations of buildings, and the third infrastructure dataset containing information such as the layout of indoor shopping mall counters and the internal structure of office buildings, based on a unified geographic coordinate system, and then performs alignment processing. For example, the geographic coordinates of the outdoor entrance of the shopping mall are associated with the coordinates of the internal entrance of the shopping mall, and the location of the office building outdoors is aligned with the corresponding locations of each floor inside the office building, thereby constructing a super-large-scale indoor and outdoor high-precision scene of the entire city block. Users can obtain a complete high-precision scene of the target area through the cloud for navigation of cars, robots (such as food delivery robots), etc.

[0037] The above scheme of the present invention performs secondary identification of outdoor targets suspected to be infrastructure, and adopts a dynamic request mechanism to obtain video data of indoor areas, thereby achieving accurate identification of indoor and outdoor infrastructure, and further achieving rapid creation of ultra-large-scale high-precision scenes.

[0038] Optionally, the identifying of a first infrastructure dataset and a suspected infrastructure dataset from the multiple groups of remote sensing image datasets includes: using a composite recognition model to identify objects in any group of the remote sensing image datasets to obtain a number of objects and their recognition confidences, identifying the objects whose recognition confidences are higher than a confidence threshold as infrastructure and classifying them into the first infrastructure dataset; wherein the composite recognition model includes a plurality of sub-recognition models for respectively identifying a number of specified types of infrastructure; and classifying the objects whose recognition confidences are not higher than a confidence threshold into the suspected infrastructure dataset.

[0039] In this embodiment, if Figure 2 As shown, the composite recognition model is designed to recognize multiple specified types of infrastructure, which may include several sub-recognition models, each of which is used to recognize a type of infrastructure, for example, sub-recognition model A is used to recognize roads, and sub-recognition model B is used to recognize buildings. When processing remote sensing images of urban blocks, the sub-recognition models only recognize these pre-set specified types of infrastructure.

[0040] When processing the remote sensing images of the above-mentioned urban blocks, any set of remote sensing image data sets (such as the set of images taken during the day on weekdays) is input into the composite recognition model. The composite recognition model analyzes and recognizes each object in the image (such as roads, buildings, vehicles, temporary structures, etc.), and assigns a recognition confidence to each recognized object. The recognition confidence reflects the degree of certainty of the model that the object is correctly recognized as a certain object, for example, expressed as a numerical value, the higher the value, the greater the possibility that the model believes that the object is correctly recognized. For example, for an office building in the image, the composite recognition model gives a corresponding recognition confidence by analyzing its shape, texture, relationship with the surrounding environment and other features.

[0041] When the recognition confidence of an object given by the composite recognition model is higher than the set confidence threshold, it can be more reliably considered that the object is infrastructure. For example, in a remote sensing image of a city block, after a road is analyzed by the composite recognition model, its recognition confidence is high and exceeds the confidence threshold. In this way, the road is identified as infrastructure and its related information (such as location, shape, etc.) is included in the first infrastructure data set. In this way, the first infrastructure data set initially contains information on objects that the model believes are likely to be infrastructure.

[0042] If the composite recognition model gives a recognition confidence of an object that is not higher than the set confidence threshold, it means that the model is not sure whether the object is infrastructure. For example, in an image of a city block, a temporary stage with a color close to the road surface, or a large truck temporarily parked on the roadside to load and unload sculptures, these objects may have a low recognition confidence after being recognized by the model and do not reach the confidence threshold. At this time, these objects are classified into the suspected infrastructure dataset. The objects in the suspected infrastructure dataset will be reconfirmed later to further determine whether they are really infrastructure.

[0043] Optionally, the secondary confirmation of each suspected infrastructure in the suspected infrastructure data set includes: determining the recognition confidence of each suspected infrastructure, and determining the first target quantity and target span time according to the recognition confidence; taking the shooting time of the remote sensing image data set corresponding to the recognition confidence as a reference, screening multiple groups of other remote sensing image data sets according to the target span time to obtain the first target quantity, and secondary confirmation of the suspected infrastructure based on the corresponding multiple remote sensing images in each other remote sensing image data set to obtain a secondary confirmation result, and the secondary confirmation result includes confirmation as infrastructure and confirmation as non-infrastructure.

[0044] In this embodiment, for each object that has been classified into the suspected infrastructure data set (such as a temporary stage in an urban block image, a truck for loading and unloading sculptures, etc.), two key parameters are first determined based on the recognition confidence obtained above: the first target number and the target span time.

[0045] Generally speaking, the lower the recognition confidence, the more uncertain the model's judgment of the object. In this case, a larger number of first targets (that is, more groups of other remote sensing image data sets need to be referenced, such as 10 or 20) and a longer target span time (that is, the shooting time span between the referenced remote sensing images is larger, such as 1 hour or 2 hours) are set to more accurately judge the true attributes of the object. On the contrary, for suspected infrastructure with relatively high recognition confidence (but still below the confidence threshold), a smaller number of first targets are set and the target span time is also shorter.

[0046] After determining the first target quantity and target span duration corresponding to each suspected infrastructure, the shooting time of the corresponding remote sensing image when the suspected infrastructure was first identified is used as the reference time point. Then, based on the determined target span duration, other remote sensing image datasets with the first target quantity are screened out from the existing remote sensing image datasets taken at different time periods.

[0047] After screening out the other remote sensing image data sets of the first target number, analyze the multiple remote sensing images related to the suspected infrastructure in each data set. For example, for the truck loading and unloading the sculpture, check the remote sensing images taken 1 hour, 2 hours, etc. before the initial remote sensing image, and 1 hour, 2 hours, etc. after the initial remote sensing image (the target span is 1 hour), and observe the location, status, and relationship with the surrounding environment of the truck at these different times. If it is found that the location of the truck has changed significantly at different time points and does not conform to the static characteristics of the infrastructure, it can be confirmed that it is non-infrastructure; if it is found in multiple sets of images that the location, shape, and other features of the object remain unchanged and conform to the characteristics of a certain infrastructure, it can be confirmed that it is infrastructure. Through such a comprehensive analysis based on multiple sets of remote sensing images at different times, a secondary confirmation result for each suspected infrastructure is obtained, which is clearly distinguished as infrastructure or non-infrastructure, thereby further improving the accurate identification of outdoor infrastructure.

[0048] Among them, the remote sensing images corresponding to the same suspected infrastructure in different remote sensing image datasets have slight differences in shooting angles in addition to the different shooting times, which can be used to analyze whether the suspected infrastructure is an infrastructure or a temporary object.

[0049] Optionally, the secondary confirmation of the suspected infrastructure is performed based on the corresponding multiple remote sensing images in each other remote sensing image data set to obtain the secondary confirmation result, including: determining several associated sub-identification models corresponding to the sub-identification model of the suspected infrastructure; wherein the similarity between each of the associated sub-identification models and the type of infrastructure used to identify the sub-identification model is higher than a similarity threshold; using each of the associated sub-identification models to separately identify multiple remote sensing images, and fusing each identification result to obtain the secondary confirmation result.

[0050] In this embodiment, although the multiple sub-recognition models included in the composite recognition model are for different types of infrastructure, some infrastructures have certain visual similarities in remote sensing images. For example, water towers and signal towers installed on roads both usually have tall columnar or tower-like shapes and have high visual similarities in remote sensing images.

[0051] In view of this, the present invention determines these sub-recognition models as the associated sub-recognition models corresponding to the sub-recognition model, and then dispatches these associated sub-recognition models to respectively identify the multiple remote sensing images obtained above together with the sub-recognition model, and fuses the recognition results of the associated sub-recognition model and the sub-recognition model to obtain the final secondary confirmation result. For example, the recognition result of the sub-recognition model is: infrastructure-sculpture, recognition confidence 90%; the recognition result of the associated sub-recognition model 1 is: infrastructure-A, recognition confidence 65%; the recognition result of the associated sub-recognition model 2 is: infrastructure-B, recognition confidence 73%; the recognition result of the associated sub-recognition model 3 is: infrastructure-C, recognition confidence 86%. The recognition confidence of each associated sub-recognition model is integrated, for example, the mean, median or maximum value is obtained, and the integrated comprehensive recognition confidence is, for example, 86% (maximum value). At this time, according to the preset control relationship, the auxiliary coefficient corresponding to 86% is determined to be 0.9 (the auxiliary coefficient is positively correlated with the integration value). Correspondingly, the recognition confidence of the suspected infrastructure is 90%*0.9=81%, and 81% is higher than the confidence threshold of 80%. At this time, the secondary confirmation result of the suspected infrastructure is determined to be confirmed as infrastructure.

[0052] Therefore, while using multiple remote sensing images to perform secondary confirmation of suspected infrastructure, the present invention further refers to an associated sub-recognition model for identifying similar infrastructure to assist in analyzing whether the suspected infrastructure belongs to the corresponding type of infrastructure, thereby improving the accuracy of the analysis.

[0053] Optionally, the dynamic request mechanism is used to request the video data corresponding to the indoor area, including: requesting a group of video data corresponding to the indoor area, analyzing the object position change amplitude from the group of video data, and determining a second target quantity according to the change amplitude level to which the object position change amplitude belongs; and requesting multiple groups of video data corresponding to the second target quantity of the indoor area.

[0054] In this embodiment, when it is necessary to obtain video data of a certain indoor area (such as inside a shopping mall) to construct a high-precision scene, the cloud first requests a set of video data from the relevant video acquisition equipment (such as a surveillance camera) of the indoor area. Then, computer vision and image analysis technology are used to identify various objects in the video (such as goods, shopping carts, pedestrians, etc. in the mall), and the position changes of these objects in the video frame are tracked to calculate the overall position change amplitude of the objects in the video screen. For example, most of the goods in the video screen are located in area A in the first period, and mainly in area B in the second period, which means that there are obvious goods entering, stacking, and transporting in the video screen, such as a storage area. In this case, since the goods are handled more frequently, it is easier to quickly determine the passable area in the indoor area. At this time, fewer groups of video data are requested.

[0055] On the contrary, the goods in the video are always concentrated in area A, which means that the goods are not moved frequently. At this time, it is difficult to determine the passable area in the indoor area from the set of video data, because the goods always block the passable area. At this time, more sets of video data are requested to obtain video frames where goods are moved in a passable area, and then the boundary of the passable area is extracted.

[0056] The boundary of the traversable area extracted in the above manner can be used for navigation of various robots, for example.

[0057] like Figure 3As shown, an embodiment of the present invention further discloses a system for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes, the system comprising a receiving and preprocessing module, an outdoor analysis and recognition module, an indoor analysis and recognition module, and a scene integration generation module; the receiving and preprocessing module receives a target area corresponding to a high-precision scene input by a user, and divides the target area into an outdoor area and an indoor area based on map information corresponding to the target area; the outdoor analysis and recognition module obtains multiple groups of remote sensing image data sets corresponding to the outdoor area in different acquisition time periods, identifies a first infrastructure data set and a suspected infrastructure data set from the multiple groups of remote sensing image data sets, performs secondary confirmation on each suspected infrastructure in the suspected infrastructure data set, updates the first infrastructure data set based on the secondary confirmation result, and obtains a second infrastructure data set; the indoor analysis and recognition module uses a dynamic request mechanism to request to obtain video data corresponding to the indoor area, and identifies a third infrastructure data set from the video data; the scene integration generation module performs coordinate mapping and position alignment on the second infrastructure data set and the third infrastructure data set to obtain an ultra-large-scale indoor and outdoor high-precision scene.

[0058] An embodiment of the present invention further discloses an electronic device, comprising: at least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method described in the above embodiment.

[0059] An embodiment of the present invention further discloses a computer storage medium, wherein the computer storage medium stores a computer program, and the computer program is executed by a processor to implement the method described in the above embodiment.

[0060] The computer-readable storage medium described above may include, but is not limited to, electronic, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices or equipment, or any suitable combination of the above. Alternatively, the computer-readable storage medium may be a machine-readable signal medium. More specific examples of machine-readable storage media may include electrical connections based on one or more lines, portable computer disks, hard disks, random access memories (RAM), read-only memories (ROM), erasable programmable read-only memories (EPROM or flash memory), optical fibers, portable compact disk read-only memories (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination of the above.

[0061] An embodiment of the present invention further discloses a computer program product, which includes computer code. When the computer code is executed by a processor of an electronic device, the method described in the above embodiment is implemented.

[0062] It should be understood that the various forms of processes shown above can be used to reorder, add or delete steps. For example, the steps described in the present invention can be executed in parallel, sequentially or in different orders, as long as the desired results of the technical solution of the present invention can be achieved, and this document does not limit this.

[0063] The above specific implementations do not constitute a limitation on the protection scope of the present invention. It should be understood by those skilled in the art that various modifications, combinations, sub-combinations and substitutions can be made according to design requirements and other factors. Any modification, equivalent substitution and improvement made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes, characterized in that: The method comprises the following steps: receiving a target area corresponding to a high-precision scene input by a user, and dividing the target area into an outdoor area and an indoor area based on map information corresponding to the target area; Acquire multiple sets of remote sensing image data sets corresponding to the outdoor area at different acquisition time periods, identify a first infrastructure data set and a suspected infrastructure data set from the multiple sets of remote sensing image data sets, perform secondary confirmation on each suspected infrastructure in the suspected infrastructure data set, and update the first infrastructure data set based on the secondary confirmation result to obtain a second infrastructure data set; Using a dynamic request mechanism to request to obtain video data corresponding to the indoor area, and identifying a third infrastructure data set from the video data; The second infrastructure dataset is coordinate mapped and position aligned with the third infrastructure dataset to obtain an ultra-large-scale high-precision indoor and outdoor scene.

2. The method for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes according to claim 1, characterized in that: A first infrastructure dataset and a suspected infrastructure dataset are identified from the plurality of groups of remote sensing image datasets, comprising: using a composite recognition model to identify objects in any group of the remote sensing image datasets to obtain a plurality of objects and their recognition confidences, identifying the objects whose recognition confidences are higher than a confidence threshold as infrastructures, and classifying them into the first infrastructure dataset; wherein the composite recognition model comprises a plurality of sub-recognition models for respectively identifying a plurality of specified types of infrastructures; and classifying the objects whose recognition confidences are not higher than the confidence threshold into the suspected infrastructure dataset.

3. The method for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes according to claim 2, characterized in that: A secondary confirmation is performed on each suspected infrastructure in the suspected infrastructure data set, including: determining the recognition confidence of each suspected infrastructure, and determining the first target quantity and target span time according to the recognition confidence; taking the shooting time of the remote sensing image data set corresponding to the recognition confidence as a reference, screening multiple groups of other remote sensing image data sets for the first target quantity according to the target span time, and performing a secondary confirmation on the suspected infrastructure based on the corresponding multiple remote sensing images in each other remote sensing image data set to obtain a secondary confirmation result, and the secondary confirmation result includes confirmation as infrastructure and confirmation as non-infrastructure.

4. The method for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes according to claim 2, characterized in that: The suspected infrastructure is reconfirmed based on multiple corresponding remote sensing images in other remote sensing image data sets to obtain a secondary confirmation result, including: determining a number of associated sub-identification models corresponding to the sub-identification model of the suspected infrastructure; wherein the similarity between each associated sub-identification model and the type of infrastructure used to identify the sub-identification model is higher than a similarity threshold; using each associated sub-identification model to separately identify multiple remote sensing images, fusing each identification result, and obtaining the secondary confirmation result.

5. The method for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes according to claim 1, characterized in that: A dynamic request mechanism is adopted to request to obtain the video data corresponding to the indoor area, including: requesting to obtain a group of video data corresponding to the indoor area, analyzing the object position change amplitude from the group of video data, and determining the second target quantity according to the change amplitude level to which the object position change amplitude belongs; requesting to obtain multiple groups of video data corresponding to the second target quantity of the indoor area.

6. A system for rapidly constructing ultra-large-scale indoor and outdoor high-precision scenes, characterized by: The system includes a receiving and preprocessing module, an outdoor analysis and recognition module, an indoor analysis and recognition module, and a scene integration generation module; The receiving and preprocessing module receives a target area corresponding to a high-precision scene input by a user, and divides the target area into an outdoor area and an indoor area based on map information corresponding to the target area; The outdoor analysis and identification module obtains multiple groups of remote sensing image data sets corresponding to the outdoor area in different acquisition time periods, identifies a first infrastructure data set and a suspected infrastructure data set from the multiple groups of remote sensing image data sets, performs secondary confirmation on each suspected infrastructure in the suspected infrastructure data set, updates the first infrastructure data set based on the secondary confirmation result, and obtains a second infrastructure data set; the indoor analysis and identification module adopts a dynamic request mechanism to request to obtain video data corresponding to the indoor area, and identifies a third infrastructure data set from the video data; the scene integration generation module performs coordinate mapping and position alignment on the second infrastructure data set and the third infrastructure data set to obtain an ultra-large-scale indoor and outdoor high-precision scene.

7. The ultra-large-scale indoor and outdoor high-precision scene rapid construction system according to claim 6, characterized in that: The outdoor analysis and identification module is used to: use a composite recognition model to identify objects in any group of the remote sensing image data sets, obtain a number of objects and their recognition confidences, identify the objects whose recognition confidences are higher than a confidence threshold as infrastructure, and classify them into the first infrastructure data set; wherein the composite recognition model includes a plurality of sub-recognition models for respectively identifying a number of specified types of infrastructure; and classify the objects whose recognition confidences are not higher than the confidence threshold into a suspected infrastructure data set.

8. An electronic device comprising: At least one processor, a memory, and a computer program stored in the memory and executable on the at least one processor, wherein the processor executes the computer program to implement the method according to any one of claims 1 to 5.

9. A computer storage medium storing a computer program, characterized in that: The computer program is executed by a processor to implement the method according to any one of claims 1 to 5.

10. A computer program product, characterized in that: The computer program product includes computer codes, and when the computer codes are executed by a processor of an electronic device, the method according to any one of claims 1 to 5 is implemented.

Images