Map correction method, device, computer equipment and storage medium

Abstract

The present application relates to a map correction method, device, computer equipment and storage medium. It relates to the field of intelligent inspection technology. The method comprises: in the inspection process of a mobile terminal, comparing the first recognition result of the passability of a target object in a first composition with the second recognition result of a second composition; when the first and second recognition results are inconsistent, obtaining the actual passability result of the mobile terminal for the target object; when the actual passability result indicates that the first recognition result is incorrect, correcting the first composition so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content of the second composition; or, when the actual passability result indicates that the second recognition result is incorrect, correcting the second composition so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content of the first composition. The use of this method can improve the accuracy of the navigation map of the inspection equipment.

Description

Technical Field

[0001] The present application relates to the field of intelligent inspection technology, and in particular to a map correction method, device, computer equipment and storage medium. Background Art

[0002] In the field of traditional inspection technology, inspection equipment such as robots and drones usually use laser point cloud mapping to build maps. The advantage is that it can realize the construction of global maps and provide an important reference for the navigation and movement of inspection equipment.

[0003] However, the current laser point cloud mapping method requires a lot of computing resources and time, especially in large-scale environments, and the above method is inefficient; and if there is a moving target object in the active environment of the inspection equipment, the point cloud map that has been obtained will be disturbed, which will lead to a decrease in the accuracy and reliability of the final map. Therefore, the accuracy of the map obtained by the current inspection equipment based on laser point cloud mapping is low. Summary of the invention

[0004] Based on this, the present application provides a map correction method, apparatus, computer device, computer-readable storage medium and computer program product.

[0005] In a first aspect, the present application provides a map correction method, comprising:

[0006] During the inspection process of the mobile terminal, a first recognition result of the passability of the target object in the first composition is compared with a second recognition result of the passability of the target object in the second composition; the first composition is a map constructed based on point cloud data captured by a first device carried by the mobile terminal, and the second composition is a map constructed based on point cloud data captured by a second device carried by the mobile terminal; an area covered by the first composition at least partially overlaps with an area covered by the second composition;

[0007] When the first recognition result and the second recognition result are inconsistent, obtaining an actual passability result of the mobile terminal over the target object;

[0008] In a case where the actual pass result indicates that the first recognition result is incorrect, the first composition is corrected so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content in the second composition; or, in a case where the actual pass result indicates that the second recognition result is incorrect, the second composition is corrected so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content in the first composition.

[0009] In one of the embodiments, when the first recognition result and the second recognition result are inconsistent, obtaining the actual passability result of the mobile terminal for the target object includes: when the first recognition result is passable and the second recognition result is unpassable, controlling the mobile terminal to move along a pre-planned navigation path to obtain the actual passability result of the mobile terminal for the target object; wherein the pre-planned navigation path is generated based on the first composition; when the first recognition result is unpassable and the second recognition result is passable, controlling the mobile terminal to move along a re-planned navigation path to obtain the actual passability result of the mobile terminal for the target object; wherein the re-planned navigation path is generated based on the second recognition result.

[0010] In one of the embodiments, the method further includes: when the actual passability result indicates that the first recognition result is incorrect, saving the point cloud data corresponding to the first recognition result, and marking the actual passability result of the point cloud data as a feedback sample of the first composition model; retraining the first composition model based on the feedback sample of the first composition model; the first composition model is used to construct a map based on the point cloud data captured by the first device.

[0011] In one of the embodiments, when the actual passability result indicates that the first recognition result is incorrect, the point cloud data corresponding to the first recognition result is saved as a feedback sample of the first composition model, including: when the first recognition result is passable, the second recognition result is unpassable and the actual passability result is unpassable, the point cloud data corresponding to the first recognition result is saved, and the point cloud data is marked as unpassable as the first feedback sample of the first composition model; when the first recognition result is unpassable, the second recognition result is passable and the actual passability result is passable, the point cloud data corresponding to the first recognition result is saved, and the point cloud data is marked as passable as the second feedback sample of the first composition model; the retraining of the first composition model based on the feedback samples of the first composition model includes: retraining the first composition model based on the first feedback samples and / or the second feedback samples of the first composition model.

[0012] In one of the embodiments, the method further includes: when the actual passability result indicates that the second recognition result is incorrect, saving the point cloud data corresponding to the second recognition result and marking the actual passability result of the point cloud data as a feedback sample of the second composition model; retraining the second composition model based on the feedback sample of the second composition model; the second composition model is used to construct a map according to the point cloud data captured by the second device.

[0013] In one of the embodiments, when the actual passability result indicates that the second recognition result is incorrect, the point cloud data corresponding to the second recognition result is saved as a feedback sample of the second composition model, including: when the second recognition result is passable, the first recognition result is unpassable and the actual passability result is unpassable, the point cloud data corresponding to the second recognition result is saved, and the point cloud data is marked as unpassable as the first feedback sample of the second composition model; when the second recognition result is unpassable, the first recognition result is passable and the actual passability result is passable, the point cloud data corresponding to the second recognition result is saved, and the point cloud data is marked as passable as the second feedback sample of the second composition model; the retraining of the second composition model based on the feedback samples of the second composition model includes: retraining the second composition model based on the first feedback samples and / or the second feedback samples of the second composition model.

[0014] In one embodiment, the first device is a laser radar, and the second device is a camera; the first composition or the second composition is constructed in the following manner: acquiring point cloud data captured by the laser radar or the camera; the capture range of the camera is smaller than the capture range of the laser radar; extracting key features in the point cloud data; the key features include at least one of corner features, edge features, and plane point features; based on the key features, point cloud registration is performed on the point cloud data acquired twice adjacently to acquire the posture information of the laser radar or the camera; based on the posture information and the point cloud data, the first composition or the second composition is obtained; the area covered by the second composition is a partial area of ​​the area covered by the first composition.

[0015] In a second aspect, the present application also provides a map correction device, comprising:

[0016] A recognition result comparison module, used for comparing, during the inspection process of the mobile terminal, a first recognition result of the passability of a target object in a first composition and a second recognition result of the passability of the target object in a second composition; the first composition is a map constructed based on point cloud data captured by a first device carried by the mobile terminal, and the second composition is a map constructed based on point cloud data captured by a second device carried by the mobile terminal; an area covered by the first composition at least partially overlaps with an area covered by the second composition;

[0017] An actual passability result acquisition module, configured to acquire an actual passability result of the mobile terminal for the target object when the first recognition result and the second recognition result are inconsistent;

[0018] A correction module is used to correct the first composition when the actual pass result indicates that the first recognition result is incorrect, so that the inconsistent content between the first composition and the second composition is replaced by the corresponding content of the second composition; or, when the actual pass result indicates that the second recognition result is incorrect, correct the second composition so that the inconsistent content with the first composition is replaced by the corresponding content of the first composition.

[0019] In a third aspect, the present application further provides a computer device, the computer device comprising a memory and a processor, the memory storing a computer program, and the processor implementing the following steps when executing the computer program:

[0020] During the inspection process of the mobile terminal, a first recognition result of the passability of the target object in the first composition is compared with a second recognition result of the passability of the target object in the second composition; the first composition is a map constructed based on point cloud data captured by a first device carried by the mobile terminal, and the second composition is a map constructed based on point cloud data captured by a second device carried by the mobile terminal; an area covered by the first composition at least partially overlaps with an area covered by the second composition;

[0021] When the first recognition result and the second recognition result are inconsistent, obtaining an actual passability result of the mobile terminal over the target object;

[0022] In a case where the actual pass result indicates that the first recognition result is incorrect, the first composition is corrected so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content in the second composition; or, in a case where the actual pass result indicates that the second recognition result is incorrect, the second composition is corrected so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content in the first composition.

[0023] In a fourth aspect, the present application further provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the following steps are implemented:

[0024] During the inspection process of the mobile terminal, a first recognition result of the passability of the target object in the first composition is compared with a second recognition result of the passability of the target object in the second composition; the first composition is a map constructed based on point cloud data captured by a first device carried by the mobile terminal, and the second composition is a map constructed based on point cloud data captured by a second device carried by the mobile terminal; an area covered by the first composition at least partially overlaps with an area covered by the second composition;

[0025] When the first recognition result and the second recognition result are inconsistent, obtaining an actual passability result of the mobile terminal over the target object;

[0026] In a case where the actual pass result indicates that the first recognition result is incorrect, the first composition is corrected so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content in the second composition; or, in a case where the actual pass result indicates that the second recognition result is incorrect, the second composition is corrected so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content in the first composition.

[0027] In a fifth aspect, the present application further provides a computer program product, the computer program product comprising a computer program, which implements the following steps when executed by a processor:

[0028] During the inspection process of the mobile terminal, a first recognition result of the passability of the target object in the first composition is compared with a second recognition result of the passability of the target object in the second composition; the first composition is a map constructed based on point cloud data captured by a first device carried by the mobile terminal, and the second composition is a map constructed based on point cloud data captured by a second device carried by the mobile terminal; an area covered by the first composition at least partially overlaps with an area covered by the second composition;

[0029] When the first recognition result and the second recognition result are inconsistent, obtaining an actual passability result of the mobile terminal over the target object;

[0030] In a case where the actual pass result indicates that the first recognition result is incorrect, the first composition is corrected so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content in the second composition; or, in a case where the actual pass result indicates that the second recognition result is incorrect, the second composition is corrected so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content in the first composition.

[0031] The map correction method, device, computer equipment, storage medium and computer program product provided in the present application are applied to a mobile terminal. During the inspection process of the mobile terminal, the first recognition result of the passability of the target object in the first composition and the second recognition result of the passability of the target object in the second composition are first compared; the first composition is a map constructed based on point cloud data captured by a lidar, and the second composition is a map constructed based on point cloud data captured by a camera; then, when it is determined that there is an inconsistency between the first recognition result and the second recognition result, the actual passability result of the mobile terminal for the target object is obtained; finally, when the actual passability result indicates that the first recognition result is incorrect, the first composition is corrected so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content of the second composition; or, when the actual passability result indicates that the second recognition result is incorrect, the second composition is corrected so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content of the first composition. In the above process, by comparing the recognition results of the passability of the target object in the map constructed by comparing the point cloud data captured by two different devices, when the two recognition results are consistent, the passability of the target object can be predicted as passable or impassable. When the two recognition results are inconsistent, the actual passability result of the target object by the mobile terminal is further obtained, and based on the actual passability result, it is determined whether the first composition needs to be corrected, so as to avoid the mobile terminal being further misled by the inaccurate first composition, colliding with an impassable obstacle or causing the inspection path to detour, thereby optimizing the inspection path of the mobile terminal. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] In order to more clearly illustrate the technical solutions in the embodiments of the present application or the related technologies, the drawings required for use in the embodiments or the related technical descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0033] Figure 1 A schematic diagram of a flow chart of a map correction method in one embodiment;

[0034] Figure 2is a schematic flow chart of a method for constructing a first composition or a second composition in one embodiment;

[0035] Figure 3 is a detailed flowchart of a map correction method in an embodiment;

[0036] Figure 4 is a schematic diagram of a map correction system in one embodiment;

[0037] Figure 5 is a structural block diagram of a map correction device in one embodiment;

[0038] Figure 6 FIG. 4 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0039] In order to make the purpose, technical solution and advantages of the present application more clearly understood, the present application is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present application and are not used to limit the present application.

[0040] In one embodiment, Figure 1 As shown, a map correction method is provided. This embodiment uses the method applied to a terminal as an example. It can be understood that the method can also be applied to a server corresponding to the terminal, and can also be applied to a system including a terminal and a server, and is implemented through the interaction between the terminal and the server; wherein the terminal can be a mobile terminal, such as a robot, a drone, an unmanned vehicle, etc. In this embodiment, the method includes the following steps S102 to S106. Among them:

[0041] Step S102, during the inspection process of the mobile terminal, compare the first recognition result of the passability of the target object in the first composition with the second recognition result of the passability of the target object in the second composition; the first composition is a map constructed based on the point cloud data captured by the first device carried by the mobile terminal, and the second composition is a map constructed based on the point cloud data captured by the second device carried by the mobile terminal; the area covered by the first composition at least partially overlaps with the area covered by the second composition.

[0042] Among them, the mobile terminal is such as a robot, a drone, an unmanned vehicle, etc., and the mobile terminal is equipped with a first device and a second device for collecting environmental data. The environmental data scanning ranges of the first device and the second device can be the same or different. When the scanning ranges of the two are different, illustratively, the range of the first composition constructed based on the point cloud data captured by the first device is greater than or equal to the range of the second composition constructed based on the point cloud data captured by the second device.

[0043] As an example, the mobile terminal is equipped with a laser radar and a camera, which can capture environmental point cloud data at different angles and in different ranges. As a specific example, the laser radar scans the point cloud data of a larger range of environments, and builds a navigation map based on the point cloud data scanned by the laser radar. The camera is used to scan the point cloud data of a smaller range of nearby environments, which is used to verify the navigation map and navigation path of the mobile terminal.

[0044] Among them, inspection may refer to regular or on-demand inspection of the target area by the mobile terminal; the target object may be any object in the target area, specifically an object that needs to be judged for passability when generating the navigation map of the mobile terminal; passability refers to whether the target object hinders the passage of the mobile terminal; the first recognition result refers to the passability result of the target object obtained by the first composition, which may include pass and fail; correspondingly, the second recognition result refers to the passability result of the target object obtained by the second composition, which may also include pass and fail; point cloud data is a representation of three-dimensional data composed of a large number of points. In this embodiment, the mobile terminal can generate a map using the point cloud data captured by the onboard laser radar, or it can generate a map based on the point cloud data captured by the onboard camera, and perform path planning and obstacle avoidance of the mobile terminal based on the map.

[0045] As an example, the movable terminal may be an inspection robot. Specifically, the robot may move for inspection on a pre-planned path, automatically collect required data in the target area during the inspection process, and simultaneously compare in real time or periodically a map constructed based on point cloud data captured by an onboard lidar and a map constructed based on point cloud data captured by a camera, for the recognition results of the passability of the same target object at the same location, and then recognize in real time or periodically whether the passability recognition results of the two are consistent.

[0046] Step S104, when the first recognition result and the second recognition result are inconsistent, obtaining the actual passability result of the mobile terminal over the target object.

[0047] Among them, the situation where the first recognition result and the second recognition result are inconsistent may include that the first recognition result is passable but the second recognition result is not passable, or the first recognition result is not passable but the second recognition result is passable; the actual passability result refers to controlling the robot to try to cross the target object and obtaining the robot's passing status of the target object, so as to verify the accuracy of the aforementioned first recognition result and the second recognition result.

[0048] As an example, the actual passability of the robot over the target object can be assisted by a contact sensor installed on the robot. When the robot crosses an obstacle, if the robot touches the target object, the contact sensor will feedback the contact information. The robot's passing status over the target object can also be confirmed through the collaborative use of lidar and cameras.

[0049] Step S106, when the actual pass result indicates that the first recognition result is incorrect, correct the first composition so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content of the second composition; or, when the actual pass result indicates that the second recognition result is incorrect, correct the second composition so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content of the first composition.

[0050] Among them, the actual passability result indicates that the first recognition result is incorrect, which can be understood as the actual passability result being inconsistent with the first recognition result. Specifically, it may include that the first recognition result is passable but the actual passability result is unpassable, or the first recognition result is unpassable but the actual passability result is passable; correction refers to adjusting the first composition to ensure the accuracy of the composition data.

[0051] As an example, when the first recognition result and the second recognition result are inconsistent, and the actual passability result indicates that the first recognition result is incorrect, it can be determined that the second recognition result of the target object in the second composition is correct. Based on this, the specific correction method may be: obtain the inconsistent passability recognition content results for the same object in the first composition relative to the second composition, replace the corresponding content of the object in the first composition with the content of the object in the second composition, and in addition, the corresponding point cloud data of the object in the point cloud data corresponding to the first composition may be replaced with the point cloud data corresponding to the second composition.

[0052] Among them, when the actual passability result indicates that the second recognition result is incorrect, the situation of correcting the second composition is similar to the principle of correcting the first composition when the actual passability result indicates that the first recognition result is incorrect, and will not be repeated.

[0053] In some embodiments, when the coverage area of ​​the first composition is larger than the coverage area of ​​the second composition, and the mobile terminal performs route planning based on the first composition by default, the first composition is corrected based on the aforementioned actual passability result, and the second composition does not need to be corrected; otherwise, the second composition is corrected based on the aforementioned actual passability result, and the first composition does not need to be corrected.

[0054] The above-mentioned map correction method is applied to a mobile terminal. During the inspection process of the mobile terminal, the first recognition result of the passability of the target object in the first composition and the second recognition result of the passability of the target object in the second composition are first compared; the first composition is a map constructed based on point cloud data captured by a laser radar, and the second composition is a map constructed based on point cloud data captured by a camera; then, when it is determined that there is an inconsistency between the first recognition result and the second recognition result, the actual passability result of the mobile terminal for the target object is obtained; finally, when the actual passability result indicates that the first recognition result is incorrect, the first composition is corrected so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content of the second composition; or, when the actual passability result indicates that the second recognition result is incorrect, the second composition is corrected so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content of the first composition. In the above process, by comparing the recognition results of the target object in the maps constructed based on the point cloud data of two different devices, when the two recognition results are consistent, the passability of the target object can be predicted as passable or impassable. When the two recognition results are inconsistent, the actual passability result of the target object by the mobile terminal is further obtained, and based on the actual passability result, it is determined whether the first composition needs to be corrected, so as to avoid the mobile terminal being further misled by the inaccurate first composition, colliding with an impassable obstacle or causing the inspection path to detour, thereby optimizing the inspection path of the mobile terminal.

[0055] In an exemplary embodiment, step S104, when the first recognition result and the second recognition result are inconsistent, obtaining the actual passability result of the mobile terminal over the target object includes:

[0056] When the first recognition result is passable and the second recognition result is unpassable, the mobile terminal is controlled to move along a pre-planned navigation path to obtain an actual passability result of the mobile terminal for the target object; wherein the pre-planned navigation path is generated based on the first composition; when the first recognition result is unpassable and the second recognition result is passable, the mobile terminal is controlled to move along a re-planned navigation path to obtain an actual passability result of the mobile terminal for the target object; wherein the re-planned navigation path is generated based on the second recognition result.

[0057] Among them, "passable" indicates that the target object does not hinder the passage of the mobile terminal, and correspondingly, "impassable" indicates that the target object hinders the passage of the mobile terminal; the pre-planned navigation path refers to a navigation path that has been executed before, and the path is generated according to the information of the first composition, that is, it is generated according to the point cloud data captured by the first device carried by the mobile terminal; the re-planned navigation path movement refers to a path that needs to be dynamically adjusted and calculated, and the path can be generated according to the information of the second composition or partially according to the information of the second composition, that is, the second composition is a map constructed based on the point cloud data captured by the second device carried by the mobile terminal.

[0058] As an example, the movable terminal may be a robot, the first device carried by the movable terminal may be a laser radar, and the second device carried by the movable terminal may be a camera. In the default state, the robot navigates the inspection path based on a map constructed based on point cloud data captured by the laser radar. When the map constructed based on the point cloud data captured by the laser radar determines that the robot can pass, that is, the target object does not hinder the passage of the robot, and the map constructed based on the point cloud data captured by the camera determines that the robot cannot pass, that is, the target object hinders the passage of the robot, the robot is controlled to move according to the pre-planned navigation path. If it can pass, it means that the actual passability result is the accurate first recognition result. If it cannot pass, it means that the actual passability result is the accurate second recognition result.

[0059] More specifically, as an example, when the map constructed based on the point cloud data captured by the lidar determines that the robot cannot pass, that is, the target object obstructs the robot's passage, and the map constructed based on the point cloud data captured by the camera determines that the robot can pass, that is, the target object does not obstruct the robot's passage, the navigation path is replanned with reference to the map constructed based on the point cloud data captured by the camera, and the robot is controlled to move according to the replanned navigation path. If it cannot pass, it means that the actual passability result is that the first recognition result is accurate; if it can pass, it means that the actual passability result is that the second recognition result is accurate.

[0060] In this embodiment, by dynamically adjusting the travel path of the movable terminal according to different comparison results, the movable terminal can better cope with different situations. When two recognition results are inconsistent, the corresponding navigation decision is adjusted to allow the robot to try to cross the target object, thereby obtaining the actual passability of the target object.

[0061] Furthermore, in one embodiment, the map correction method also includes: when the actual passability result indicates that the first recognition result is incorrect, saving the point cloud data corresponding to the first recognition result, and marking the actual passability result of the point cloud data as a feedback sample of the first composition model; retraining the first composition model based on the feedback sample of the first composition model; the first composition model is used to construct a map according to the point cloud data captured by the first device.

[0062] Among them, the actual passability result characterizing that the first recognition result is incorrect includes: the first recognition result is passable but the actual passability result is unpassable, or the first recognition result is unpassable but the actual passability result is passable; the incorrect first recognition result indicates that the passability recognition result for the target object in the map constructed based on the point cloud data captured by the first device is incorrect.

[0063] As an example, when the first recognition result is passable but the actual passability result is not passable, that is, the map constructed based on the point cloud data captured by the laser radar has a passability recognition result for the target object, but the actual passability result is not passable, then the point cloud data corresponding to the first recognition result is saved, and the actual passability result of the point cloud data is marked as a feedback sample of the first composition model; correspondingly, when the first recognition result is not passable but the actual passability result is passable, that is, the map constructed based on the point cloud data captured by the laser radar has a passability recognition result for the target object, but the actual passability result is passable, the point cloud data corresponding to the first recognition result is also saved, and the actual passability result of the point cloud data is marked as a feedback sample of the first composition model. Therefore, when the actual passability result indicates that the first recognition result is incorrect, it means that the map constructed based on the point cloud data captured by the laser radar is inaccurate, and it is necessary to save the corresponding point cloud data and mark the actual passability result corresponding to the point cloud data, and then train the first composition model to adjust and improve it.

[0064] In this embodiment, when the actual passability result indicates that the first recognition result is incorrect, by saving the point cloud data corresponding to the first recognition result and marking the actual passability result of the point cloud data as a feedback sample, the first composition model can continuously learn new data to obtain an updated first composition model, which can more accurately convert the point cloud data collected by the first device into a map, thereby enhancing the accuracy of the map.

[0065] In one embodiment, when the actual passability result indicates that the first recognition result is incorrect, the point cloud data corresponding to the first recognition result is saved as a feedback sample of the first composition model, including:

[0066] When the first recognition result is passable, the second recognition result is unpassable, and the actual passability result is unpassable, the point cloud data corresponding to the first recognition result is saved, and the point cloud data is marked as unpassable as a first feedback sample of the first composition model; when the first recognition result is unpassable, the second recognition result is passable, and the actual passability result is passable, the point cloud data corresponding to the first recognition result is saved, and the point cloud data is marked as passable as a second feedback sample of the first composition model; based on the feedback samples of the first composition model, retraining the first composition model, including: retraining the first composition model based on the first feedback samples and / or the second feedback samples of the first composition model.

[0067] Furthermore, based on the situation that the point cloud data corresponding to the first recognition result is impassable, the first composition model is retrained, and when encountering similar new point cloud data again, the actual passability result of the target object corresponding to the new point cloud data can be accurately judged.

[0068] In this embodiment, a feedback mechanism is established by comparing the first recognition result and the second recognition result, and the point cloud data is marked according to the actual passability results, specifically marked as impassable and passable, which are the first feedback sample of the first composition model and the second feedback sample of the first composition model, respectively. The first composition model is retrained based on the feedback samples, thereby improving the first composition model's learning ability for new point cloud data and the accuracy of actual passability judgment of new point cloud data.

[0069] Furthermore, in one embodiment, the map correction method also includes: when the actual passability result indicates that the second recognition result is incorrect, saving the point cloud data corresponding to the second recognition result, and marking the actual passability result of the point cloud data as a feedback sample of the second composition model; retraining the second composition model based on the feedback sample of the second composition model; the second composition model is used to construct a map according to the point cloud data captured by the second device.

[0070] Among them, the actual passability result characterizing that the second recognition result is incorrect includes: the second recognition result is passable but the actual passability result is unpassable, or the second recognition result is unpassable but the actual passability result is passable; the incorrect second recognition result indicates that the map constructed based on the point cloud data captured by the second device carried by the mobile terminal has an incorrect passability recognition result for the target object.

[0071] As an example, when the second recognition result is passable but the actual passability result is not passable, that is, the map constructed based on the point cloud data captured by the camera has a passability recognition result for the target object, but the actual passability result is not passable, then the point cloud data corresponding to the second recognition result is saved, and the actual passability result of the point cloud data is marked as a feedback sample of the second composition model; correspondingly, when the second recognition result is not passable but the actual passability result is passable, that is, the map constructed based on the point cloud data captured by the camera has a passability recognition result for the target object, but the actual passability result is passable, the point cloud data corresponding to the second recognition result is also saved, and the actual passability result of the point cloud data is marked as a feedback sample of the second composition model. Therefore, in the case where the actual passability result indicates that the second recognition result is incorrect, it means that the map constructed based on the point cloud data captured by the camera has an incorrect passability recognition result for the target object, that is, the map constructed based on the point cloud data captured by the camera is inaccurate, then it is necessary to save the corresponding point cloud data, and mark the actual passability result corresponding to the point cloud data, and then train the second composition model to adjust and improve it.

[0072] In this embodiment, by saving the point cloud data corresponding to the second recognition result and marking the actual pass result of the point cloud data as a feedback sample, the second composition model can continuously learn new data to obtain an updated second composition model, which can more accurately convert the point cloud data collected by the second device into a map, thereby enhancing the accuracy of the map.

[0073] In an exemplary embodiment, when the actual passability result indicates that the second recognition result is incorrect, the point cloud data corresponding to the second recognition result is saved as a feedback sample of the second composition model, including: when the second recognition result is passable, the first recognition result is unpassable and the actual passability result is unpassable, the point cloud data corresponding to the second recognition result is saved, and the point cloud data is marked as unpassable as the first feedback sample of the second composition model; when the second recognition result is unpassable, the first recognition result is passable and the actual passability result is passable, the point cloud data corresponding to the second recognition result is saved, and the point cloud data is marked as passable as the second feedback sample of the second composition model; based on the feedback samples of the second composition model, the second composition model is retrained, including: based on the first feedback samples and / or the second feedback samples of the second composition model, the second composition model is retrained.

[0074] As an example, based on the situation that the point cloud data corresponding to the second recognition result is impassable, the second composition model is retrained. When encountering similar new point cloud data again, the actual passability result of the target object corresponding to the new point cloud data can be accurately judged.

[0075] In this embodiment, a feedback mechanism is established by comparing the first recognition result and the second recognition result, and the point cloud data is marked according to the actual passability results, specifically marked as impassable and passable, which are the first feedback sample of the second composition model and the second feedback sample of the second composition model, respectively. The second composition model is retrained based on the feedback samples; the learning ability of the second composition model for new point cloud data and the accuracy of the actual passability judgment of the new point cloud data are improved.

[0076] In one embodiment, Figure 2 As shown, the first device is a laser radar, and the second device is a camera; the first composition or the second composition can be constructed in the following manner:

[0077] Step S202, obtaining point cloud data captured by a laser radar or a camera; the capture range of the camera is smaller than the capture range of the laser radar.

[0078] Among them, laser SLAM (Simultaneous Localization and Mapping) can be used for mapping, which can be LOAM (Lidar Odometry and Mapping) based on 3D laser radar and LeGO-LOAM (Lidar Enhanced Google Cartographer (LEGO), which usually includes a voxel grid filter method to filter, denoise and downsample the point cloud data captured by the laser radar to reduce the amount of data and improve processing efficiency.

[0079] More, such as:

[0080]

[0081] Among them, P centroid represents the coordinates of the centroid (center point) of a point in a voxel, m is the number of points in the voxel. xi, yi, zi are the X, Y, and Z coordinates of the i-th point in the voxel, respectively.

[0082] In addition, the mapping and processing methods of cameras and lidar are similar, but the sensors for collecting data are different. 3D ToF (3D Time of Flight) cameras can be used to detect environmental changes during actual robot inspections.

[0083] Step S204, extracting key features from the point cloud data; the key features include at least one of corner point features, edge features, and plane point features.

[0084] Specifically, the FPFH (Fast Point Feature Histograms) and PFH (Point Feature Histograms) algorithms can be used to identify key corner points and plane points through feature extraction.

[0085] Step S206, performing point cloud registration on the point cloud data acquired twice adjacently based on the key features to obtain the position and posture information of the laser radar or camera.

[0086] Among them, the iterative closest point (ICP) algorithm is used to align the point clouds of adjacent scans to estimate the pose change. The ICP algorithm finds the best rotation and translation by minimizing the distance between the point clouds to align the two point clouds, as shown in the formula:

[0087]

[0088] Where pi is a point in the source point cloud, pi′ is the closest point in the target point cloud, R and t are the rotation matrix and translation vector to be estimated. In the camera-based point cloud registration process, adjacent scanned point clouds can also be aligned through NDT (Normal Distributions Transform).

[0089] More often, the pose information is obtained through pose estimation. First, the direct linear transformation method (DLT) is used to make a preliminary pose estimate, and then a nonlinear optimization method is used for further optimization to estimate the pose of the lidar. DLT solves the camera's rotation matrix R and translation vector t by constructing a set of linear equations, as shown in the formula: , where A is a matrix constructed from matching point pairs, [R|t] is an unknown transformation matrix, and nonlinear optimization methods are usually used after the initial pose estimation, such as:

[0090] .

[0091] in is a robust kernel function used to handle outliers. K usually represents the intrinsic parameter matrix of the camera or lidar. i represents a point in three-dimensional space, P iw Represents a point in a specific coordinate system. In addition, the pose of a camera or lidar can be estimated by solving the PnP (Perspective-n-point) problem, which involves estimating the pose of the camera based on the 2D pixel coordinates of feature points in the image and their corresponding 3D spatial coordinates.

[0092] Step S208, obtaining a first composition or a second composition according to the posture information and the point cloud data; the area covered by the second composition is a partial area of ​​the area covered by the first composition.

[0093] Specifically, during the whole process, the backend optimization is continuously running, using the existing graph optimization method to represent the robot's posture and map feature points as nodes, and the constraints measured by the lidar as edges. The position of the nodes is optimized by minimizing the error of all edges in the graph, so as to obtain a globally consistent map and posture estimation. Commonly used graph optimization libraries include g2o (General Graph Optimization), CeresSolver (an open source C++ library), GTSAM (an open source library), etc., which can further improve the consistency and accuracy of the map.

[0094] Finally, the map is stored in an appropriate data structure for subsequent use. The implementation of the above algorithms requires accurate sensor data synchronization, efficient feature matching strategies, and powerful back-end optimization capabilities to ensure robust positioning and accurate map construction in complex environments.

[0095] In this embodiment, through an effective point cloud data processing method, a three-dimensional scene can be reconstructed with high precision, which is beneficial to map construction, improves the accuracy of the first composition or the second composition, and further provides a basis for determining the actual passability results of subsequent target objects.

[0096] This application provides a map correction method. In order to better understand the process of the above map correction method, combined with Figure 3 As shown, the specific process of a map correction method for a patrol robot is described in detail below, including the following steps:

[0097] Step S301, obtaining a first composition constructed based on a laser radar of the robot.

[0098] The first composition is to obtain geographical data in a large area based on lidar and construct a large-scale map centered on the robot. The map constructed at this time is relatively rough and cannot reflect the specific location information of each target object in the map.

[0099] Step S302: Perform preliminary path planning of the robot based on the first composition.

[0100] Preliminary path planning is performed based on the large-scale map built by LiDAR. Since the large-scale map is not detailed enough, the path planning performed at this time cannot effectively avoid the target object during the planning process, which may cause the robot to be unable to pass.

[0101] Step S303, obtaining a second composition based on a small-range visual scan by the robot's camera.

[0102] The second composition is to obtain geographical data within a small range based on the camera and construct a small-scale map centered on the robot. The map constructed at this time is more detailed and can reflect the specific location information of each target object in the map.

[0103] Step S304: performing consistency judgment based on the first composition and the second composition to obtain a judgment result.

[0104] The large-scale map built by LiDAR and the small-scale map built by camera simultaneously judge whether the robot can travel a certain distance smoothly without being blocked by the target object. If the judgment results are consistent, the robot will proceed along the original route; if the judgment results are inconsistent, an actual passability attempt will be made. The range of the small-scale map is included in the large-scale map.

[0105] Step S305 , when the consistency judgment result between the first composition and the second composition is inconsistent, an actual passability attempt of the robot is performed.

[0106] If the lidar determines that the robot can pass through the target object, but the camera determines that the robot cannot pass through the target object, the robot will attempt to cross the target object according to the first composition constructed by the lidar. If it fails to pass, it means that the camera's judgment result is accurate. Similarly, if the lidar determines that the robot cannot pass through the target object, but the camera determines that the robot can pass through the target object, the robot will plan a new path according to the second composition constructed by the camera, and attempt to cross the target object based on the new path. If it fails to pass, it means that the camera's judgment result is inaccurate.

[0107] Step S306, recording the point cloud image to the algorithm training library.

[0108] The point cloud map that records the actual passability attempt results is stored in the algorithm training library. The point cloud map includes the point cloud map of the original route whose actual passability attempt result is passable, and the point cloud map of the map updated after the new route planning is impassable.

[0109] Step S307: retrain the first composition model and the second composition model according to the point cloud data in the point cloud image.

[0110] In the above embodiments, by combining the robot's camera and lidar, more efficient and accurate map construction and path planning can be achieved. In an environment with moving obstacles, the map and path can be dynamically adjusted to improve the robot's navigation and obstacle avoidance capabilities; in large-scale environments, the efficiency is relatively high; in the field of computer vision, real-time perception of environmental changes by visual sensors and combined with lidar technology for map and route correction can achieve higher-precision environmental perception and navigation; in addition, the robot's map and path can be dynamically adjusted to improve the efficiency of the robot's work.

[0111] Furthermore, in one embodiment, Figure 4 As shown, a schematic diagram of the robot's map correction system is provided, and the system includes: an AI system, a control system, a map system, and a perception system; further, the perception system includes a laser radar and a vision (camera) system. The perception system provides raw data, that is, point cloud data to the map system, and the map system performs path planning based on the point cloud data, and provides the results of the path planning to the control system, and then performs passability verification of the target object. After obtaining the verification results, the results are further fed back to the perception system through the AI ​​system or the map data is modified according to the verification results, and the modified map data is input into the map system to complete the map update.

[0112] It should be understood that, although the various steps in the flowcharts involved in the above-mentioned embodiments are displayed in sequence according to the indication of the arrows, these steps are not necessarily executed in sequence according to the order indicated by the arrows. Unless there is a clear explanation in this article, the execution of these steps does not have a strict order restriction, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-mentioned embodiments can include multiple steps or multiple stages, and these steps or stages are not necessarily executed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a part of the steps or stages in other steps.

[0113] Based on the same inventive concept, the embodiment of the present application also provides a map correction device for implementing the map correction method involved above. The implementation solution provided by the device to solve the problem is similar to the implementation solution recorded in the above method, so the specific limitations in one or more map correction device embodiments provided below can refer to the limitations of the map correction method above, and will not be repeated here.

[0114] In an exemplary embodiment, Figure 5As shown, a map correction device is provided, comprising: a recognition result comparison module 501, an actual passability result acquisition module 502 and a correction module 503, wherein:

[0115] The recognition result comparison module 501 is used to compare the first recognition result of the passability of the target object in the first composition with the second recognition result of the passability of the target object in the second composition during the inspection process of the mobile terminal; the first composition is a map constructed based on the point cloud data captured by the first device carried by the mobile terminal, and the second composition is a map constructed based on the point cloud data captured by the second device carried by the mobile terminal; the area covered by the first composition at least partially overlaps with the area covered by the second composition.

[0116] The actual passability result acquisition module 502 is used to acquire the actual passability result of the mobile terminal for the target object when the first recognition result and the second recognition result are inconsistent.

[0117] The correction module 503 is used to correct the first composition when the actual pass result indicates that the first recognition result is incorrect, so that the inconsistent content between the first composition and the second composition is replaced by the corresponding content of the second composition; or, when the actual pass result indicates that the second recognition result is incorrect, correct the second composition so that the inconsistent content with the first composition is replaced by the corresponding content of the first composition.

[0118] Based on the map correction device of the above embodiment, the recognition result comparison module 501 compares the recognition results of the passability of the same target object in different compositions during the inspection process of the mobile terminal; then, the actual passability result acquisition module 502 obtains the actual passability result of the mobile terminal for the target object when it is determined that there is inconsistency between the two recognition results; then, the correction module 503 corrects the first composition when the actual passability result indicates that the first recognition result is incorrect, so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content of the second composition. In the above process, it is determined whether the first composition needs to be corrected based on the actual passability result, so as to avoid the mobile terminal being further misled by the inaccurate first composition, colliding with an obstacle that cannot be passed, or causing the inspection path to detour, thereby optimizing the inspection path of the mobile terminal.

[0119] Furthermore, in one embodiment, the actual passability result acquisition module 502 is also used to control the mobile terminal to move according to a pre-planned navigation path when the first recognition result is passable and the second recognition result is unpassable, so as to obtain the actual passability result of the mobile terminal for the target object; wherein the pre-planned navigation path is generated based on the first composition; and when the first recognition result is unpassable and the second recognition result is passable, control the mobile terminal to move according to the re-planned navigation path to obtain the actual passability result of the mobile terminal for the target object; wherein the re-planned navigation path is generated based on the second recognition result.

[0120] Furthermore, in one embodiment, the correction module 503 is also used to control the mobile terminal to move according to a pre-planned navigation path when the first recognition result is passable and the second recognition result is unpassable, so as to obtain the actual passability result of the mobile terminal for the target object; wherein the pre-planned navigation path is generated based on the first composition; and when the first recognition result is unpassable and the second recognition result is passable, control the mobile terminal to move according to the re-planned navigation path to obtain the actual passability result of the mobile terminal for the target object; wherein the re-planned navigation path is generated based on the second recognition result.

[0121] Furthermore, in one embodiment, the correction module 503 is also used to save the point cloud data corresponding to the first recognition result and mark the point cloud data as unpassable as a first feedback sample of the first composition model when the first recognition result is passable, the second recognition result is unpassable, and the actual passability result is unpassable; save the point cloud data corresponding to the first recognition result and mark the point cloud data as passable as a second feedback sample of the first composition model when the first recognition result is unpassable, the second recognition result is passable, and the actual passability result is passable; retrain the first composition model based on the feedback samples of the first composition model, including: retraining the first composition model based on the first feedback samples and / or the second feedback samples of the first composition model.

[0122] Furthermore, in one embodiment, the actual passability result acquisition module 502 is also used to save the point cloud data corresponding to the second recognition result and mark the actual passability result of the point cloud data as a feedback sample of the second composition model when the actual passability result indicates that the second recognition result is incorrect; the second composition model is retrained based on the feedback sample of the second composition model; and the second composition model is used to construct a map according to the point cloud data captured by the second device.

[0123] Furthermore, in one embodiment, the correction module 503 is also used to save the point cloud data corresponding to the second recognition result and mark the point cloud data as unpassable as a first feedback sample of the second composition model when the second recognition result is passable, the first recognition result is unpassable, and the actual passability result is unpassable; save the point cloud data corresponding to the second recognition result and mark the point cloud data as passable as a second feedback sample of the second composition model when the second recognition result is unpassable, the first recognition result is passable, and the actual passability result is passable; retrain the second composition model based on the feedback samples of the second composition model, including: retraining the second composition model based on the first feedback samples and / or the second feedback samples of the second composition model.

[0124] Furthermore, in one embodiment, the correction module 503 is also used to obtain point cloud data captured by a laser radar or a camera; the capture range of the camera is smaller than the capture range of the laser radar; key features in the point cloud data are extracted; the key features include at least one of corner features, edge features, and plane point features; based on the key features, point cloud registration is performed on the point cloud data acquired twice adjacently to obtain the posture information of the laser radar or the camera; based on the posture information and the point cloud data, a first composition or a second composition is obtained; the area covered by the second composition is a partial area of ​​the area covered by the first composition.

[0125] Each module in the above-mentioned map correction device can be implemented in whole or in part by software, hardware or a combination thereof. Each module can be embedded in or independent of a processor in a computer device in the form of hardware, or can be stored in a memory in a computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0126] In an exemplary embodiment, a computer device is provided. The computer device may be a server, and its internal structure diagram may be as shown in FIG. Figure 6As shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, referred to as I / O) and a communication interface. The processor, the memory and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store map correction data. The input / output interface of the computer device is used to exchange information between the processor and an external device. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, a map correction method is implemented.

[0127] Those skilled in the art will understand that Figure 6 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than those shown in the figure, or combine certain components, or have a different arrangement of components.

[0128] In an exemplary embodiment, a computer device is provided, including a memory and a processor, wherein a computer program is stored in the memory, and the processor implements the steps in the above-mentioned method embodiments when executing the computer program.

[0129] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0130] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0131] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0132] Those skilled in the art can understand that all or part of the processes in the above-mentioned embodiment methods can be completed by instructing the relevant hardware through a computer program, and the computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above-mentioned methods. Among them, any reference to the memory, database or other medium used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. As an illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The database involved in each embodiment provided in this application may include at least one of a relational database and a non-relational database. Non-relational databases may include distributed databases based on blockchains, etc., but are not limited to this. The processor involved in each embodiment provided in this application may be a general-purpose processor, a central processing unit, a graphics processor, a digital signal processor, a programmable logic device, a data processing logic device based on quantum computing, etc., but are not limited to this.

[0133] The technical features of the above embodiments may be arbitrarily combined. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0134] The above-described embodiments only express several implementation methods of the present application, and the descriptions thereof are relatively specific and detailed, but they cannot be understood as limiting the scope of the present application. It should be pointed out that, for a person of ordinary skill in the art, several variations and improvements can be made without departing from the concept of the present application, and these all belong to the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the attached claims.

Claims

1. A map correction method, characterized in that: The method comprises: During the inspection process of the mobile terminal, a first recognition result of the passability of the target object in the first composition is compared with a second recognition result of the passability of the target object in the second composition; the first composition is a map constructed based on point cloud data captured by a first device carried by the mobile terminal, and the second composition is a map constructed based on point cloud data captured by a second device carried by the mobile terminal; an area covered by the first composition at least partially overlaps with an area covered by the second composition; When the first recognition result and the second recognition result are inconsistent, obtaining an actual passability result of the mobile terminal over the target object; In a case where the actual pass result indicates that the first recognition result is incorrect, the first composition is corrected so that the content in the first composition that is inconsistent with the second composition is replaced by the corresponding content in the second composition; or, in a case where the actual pass result indicates that the second recognition result is incorrect, the second composition is corrected so that the content in the second composition that is inconsistent with the first composition is replaced by the corresponding content in the first composition.

2. The method according to claim 1, characterized in that When the first recognition result and the second recognition result are inconsistent, obtaining an actual passability result of the mobile terminal over the target object includes: When the first recognition result is passable and the second recognition result is not passable, controlling the mobile terminal to move according to a pre-planned navigation path to obtain an actual passability result of the mobile terminal for the target object; wherein the pre-planned navigation path is generated based on the first composition; When the first recognition result is not passable and the second recognition result is passable, the movable terminal is controlled to move according to the re-planned navigation path to obtain an actual passability result of the movable terminal for the target object; wherein the re-planned navigation path is generated based on the second recognition result.

3. The method according to claim 1, characterized in that: The method further comprises: When the actual passability result indicates that the first recognition result is incorrect, saving point cloud data corresponding to the first recognition result, and marking the actual passability result of the point cloud data as a feedback sample of the first composition model; Based on the feedback samples of the first composition model, the first composition model is retrained; the first composition model is used to construct a map according to the point cloud data captured by the first device.

4. The method according to claim 3, characterized in that: When the actual passability result indicates that the first recognition result is incorrect, saving the point cloud data corresponding to the first recognition result as a feedback sample of the first composition model includes: When the first recognition result is passable, the second recognition result is unpassable, and the actual passability result is unpassable, saving point cloud data corresponding to the first recognition result, and marking the point cloud data as unpassable as a first feedback sample of the first composition model; When the first recognition result is unacceptable, the second recognition result is acceptable, and the actual passability result is acceptable, saving the point cloud data corresponding to the first recognition result, and marking the point cloud data as acceptable as a second feedback sample of the first composition model; The retraining of the first composition model based on the feedback samples of the first composition model includes: retraining the first composition model based on the first feedback samples and / or the second feedback samples of the first composition model.

5. The method according to claim 1, characterized in that The method further comprises: When the actual passability result indicates that the second recognition result is incorrect, saving point cloud data corresponding to the second recognition result, and marking the actual passability result of the point cloud data as a feedback sample of the second composition model; Based on the feedback samples of the second composition model, the second composition model is retrained; the second composition model is used to construct a map according to the point cloud data captured by the second device.

6. The method according to claim 5, characterized in that When the actual passability result indicates that the second recognition result is incorrect, saving point cloud data corresponding to the second recognition result as a feedback sample of the second composition model includes: When the second recognition result is passable, the first recognition result is unpassable, and the actual passability result is unpassable, saving point cloud data corresponding to the second recognition result, and marking the point cloud data as unpassable as a first feedback sample of the second composition model; When the second recognition result is unacceptable, the first recognition result is acceptable, and the actual passability result is acceptable, saving point cloud data corresponding to the second recognition result, and marking the point cloud data as acceptable as a second feedback sample of the second composition model; The retraining of the second composition model based on the feedback samples of the second composition model includes: retraining the second composition model based on the first feedback samples and / or the second feedback samples of the second composition model.

7. The method according to any one of claims 1 to 6, characterized in that: The first device is a laser radar, and the second device is a camera; The first composition or the second composition is constructed in the following manner: Acquire point cloud data captured by the laser radar or the camera; the capture range of the camera is smaller than the capture range of the laser radar; Extracting key features from the point cloud data; the key features include at least one of corner point features, edge features, and plane point features; Performing point cloud registration on the point cloud data acquired twice adjacently based on the key features to obtain the position and posture information of the laser radar or the camera; Obtaining the first composition or the second composition according to the posture information and the point cloud data; The area covered by the second composition is a partial area of ​​the area covered by the first composition.

8. A map correction device, characterized in that: The device comprises: A recognition result comparison module, used for comparing, during the inspection process of the mobile terminal, a first recognition result of the passability of a target object in a first composition and a second recognition result of the passability of the target object in a second composition; the first composition is a map constructed based on point cloud data captured by a first device carried by the mobile terminal, and the second composition is a map constructed based on point cloud data captured by a second device carried by the mobile terminal; an area covered by the first composition at least partially overlaps with an area covered by the second composition; An actual passability result acquisition module, configured to acquire an actual passability result of the mobile terminal for the target object when the first recognition result and the second recognition result are inconsistent; A correction module is used to correct the first composition when the actual pass result indicates that the first recognition result is incorrect, so that the inconsistent content between the first composition and the second composition is replaced by the corresponding content of the second composition; or, when the actual pass result indicates that the second recognition result is incorrect, correct the second composition so that the inconsistent content with the first composition is replaced by the corresponding content of the first composition.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images