A spatial intelligent indoor navigation instruction ambiguity active clarification method

Abstract

This invention discloses a method for proactively clarifying ambiguities in indoor navigation commands using spatial intelligence, relating to the field of spatial intelligence. It solves the problem of poor clarification effect in existing proactive clarification methods for navigation commands. The method includes the following steps: Step S1: Acquiring indoor navigation commands; acquiring the indoor space corresponding to the indoor navigation commands to obtain a pre-selected command navigation space; collecting laser point clouds of regional subspaces within the pre-selected command navigation space to obtain command region analysis data; Step S2: Performing 3D reconstruction of the pre-selected command navigation space to obtain a 3D model of the command region; semantically segmenting the indoor navigation commands to obtain multiple location description terms; classifying the location description terms by type using the 3D model of the command region to obtain location term analysis data; Step S3: Semantically clarifying ambiguous location terms based on the 3D model of the command region. This invention can improve the efficiency of clarification of navigation command ambiguities.

Description

Technical Field

[0001] This invention belongs to the field of space and relates to semantic recognition technology, specifically a method for proactively clarifying ambiguities in intelligent indoor navigation commands. Background Technology

[0002] Existing methods for proactively clarifying command ambiguities have the following drawbacks when clarifying ambiguities in indoor navigation commands: 1. Existing methods for proactively clarifying ambiguities in indoor navigation commands often struggle to select appropriate model creation methods based on the characteristics of the indoor scene. They tend to use relatively fixed modeling techniques to process various subspaces, which can easily generate navigation paths that do not conform to actual traffic rules. This increases the probability of ambiguity and requires users to perform multiple clarification interactions, significantly reducing the processing efficiency of navigation commands and the user experience. 2. Existing active ambiguity resolution methods often struggle to flexibly adapt to the semantic resolution approach based on the actual semantic content obtained from the parsing. They typically employ relatively fixed processing strategies. In some scenarios where inference disambiguation can be completed based on localization and spatial topology, unnecessary user queries may be triggered, increasing user interaction overhead. This limits the adaptability of semantic resolution to different ambiguity scenarios.

[0003] Therefore, this application proposes a method for proactively clarifying ambiguities in spatial intelligent indoor navigation commands. Summary of the Invention

[0004] To address the shortcomings of existing technologies, the purpose of this invention is to provide a method for proactively clarifying ambiguities in intelligent indoor navigation commands, thereby improving the effectiveness of proactive clarification of ambiguities in indoor navigation commands.

[0005] To achieve the above objectives, the present invention adopts the following technical solution: a method for proactively clarifying ambiguities in intelligent indoor navigation commands, comprising: Step S1: Obtain indoor navigation instructions and acquire the indoor space corresponding to the indoor navigation instructions to obtain the pre-selected instruction navigation space. Perform laser point cloud acquisition on each subspace of the pre-selected instruction navigation space, analyze and obtain the horizontal distribution complexity and vertical distribution complexity of the point cloud, and obtain instruction area analysis data. Step S2: Based on the instruction area analysis data, perform three-dimensional reconstruction of the pre-selected instruction navigation space to obtain a three-dimensional model of the instruction area. Semantically segment the indoor navigation instructions to obtain multiple location description terms. Classify the location description terms by type through the instruction area three-dimensional model to obtain location term analysis data. Step S3: Clarify the semantics of ambiguous terms based on the three-dimensional model of the instruction area.

[0006] Furthermore, in step S1, the specific steps are as follows: Step S11: Obtain the indoor navigation commands output by the user in real time to obtain the real-time navigation output commands, and obtain the indoor space associated with the real-time navigation output commands to obtain the pre-selected command navigation space. Step S12: Divide the preselected instruction navigation space into several regional subspaces, and arbitrarily select one T1 subspace from the divided regional subspaces; Step S13: Collect and analyze laser point cloud data in the T1 subspace, and obtain the horizontal and vertical distribution complexity of the point cloud corresponding to the T1 subspace based on the analysis results. Step S14: Obtain the horizontal distribution complexity and vertical distribution complexity of the point cloud corresponding to each region subspace to obtain the instruction region analysis data.

[0007] Furthermore, in step S13, the specific steps are as follows: Step S131: Use a laser scanning terminal to perform laser scanning on the T1 subspace, and obtain the three-dimensional point cloud set corresponding to the T1 subspace based on the scanning results; Step S132: Obtain the point cloud envelope space corresponding to the three-dimensional point cloud set to obtain the pre-selected point cloud envelope space; Step S133: In the preselected point cloud envelope space, mark the first preselected spatial plane and the second preselected spatial plane; Step S134: In the pre-selected point cloud envelope space, create several spatial sections parallel to the first pre-selected spatial plane to obtain multiple horizontal spatial sections. Perform position analysis on the laser sampling points in the horizontal spatial sections and obtain the horizontal distribution complexity of the point cloud based on the analysis results. Step S135: In the pre-selected point cloud envelope space, create several spatial sections parallel to the second pre-selected spatial plane to obtain multiple longitudinal spatial sections. Perform position analysis on the laser sampling points in the longitudinal spatial sections and obtain the longitudinal distribution complexity of the point cloud based on the analysis results.

[0008] Furthermore, in step S134, the specific steps are as follows: Arbitrarily select a Q1 spatial section from the obtained spatial sections, and acquire the laser sampling points contained in the Q1 spatial section to obtain multiple cross-sectional laser sampling points; Arbitrarily select a characteristic laser sampling point from the obtained cross-sectional laser sampling points, and use the characteristic laser sampling point as the center to create the largest inscribed circle within the Q1 spatial cross-section, thus obtaining the sampling neighborhood radiation circle. Mark multiple preset points on the circumference of the sampling neighborhood radiation circle. Connect each preset point on the circumference with the feature laser sampling point to obtain multiple circumference center lines. Randomly select one sample center line from the obtained circumference center lines. The laser sampling points of the cross section covered by the line connecting the sample centers are obtained, resulting in multiple line-covered sampling points. The line distance value between each line-covered sampling point and the feature laser sampling point is obtained, and the minimum line distance value is marked as the neighborhood sampling distance.

[0009] Furthermore, in step S134, the specific steps are as follows: Obtain the neighborhood sampling distance corresponding to each line connecting the centers of the circumference, and calculate the average of the multiple neighborhood sampling distances to obtain the local neighborhood average distance. Obtain the local neighborhood average distance corresponding to the laser sampling point of the cross section, and calculate the variance of the obtained local neighborhood average distance to obtain the neighborhood sampling distance variance; The neighborhood sampling distance variance is obtained by collecting spatial cross sections in history, resulting in multiple historical neighborhood distance variances. The largest historical neighborhood distance variance is marked as the neighborhood comparison distance variance. The ratio of the neighborhood sampling distance variance to the neighborhood comparison distance variance is calculated to obtain the historical peak ratio of neighborhood variance. The number of laser sampling points contained in the Q1 spatial section is counted to obtain the cumulative value of laser sampling. The product of the cumulative value of laser sampling and the ratio of the historical peak value of the neighborhood variance is calculated to obtain the laser sampling complexity. Obtain the laser sampling complexity corresponding to each horizontal spatial section, compare the obtained laser sampling complexity numerically, and mark the maximum laser sampling complexity as the horizontal distribution complexity of the point cloud.

[0010] Furthermore, step S2 includes the following steps: Obtain instruction region analysis data, and based on the instruction region analysis data, obtain the horizontal distribution complexity and vertical distribution complexity of the point cloud corresponding to each region subspace; Create a complex metric coordinate system. In the complex metric coordinate system, coordinate points are created with the horizontal distribution complexity of the point cloud as the x-axis and the vertical distribution complexity of the point cloud as the y-axis, resulting in multiple complexity distribution coordinate points. Obtain the distance values ​​between the coordinate points of each complexity distribution and the origin, and obtain the spatial complexity quantification value corresponding to each region subspace.

[0011] Furthermore, step S2 includes the following steps: A spatial complexity baseline is created. If the spatial complexity quantification value is greater than the spatial complexity baseline value, the point cloud mesh modeling method is used to reconstruct the three-dimensional region subspace. If the spatial complexity quantification value is less than or equal to the spatial complexity baseline value, the inverse solid modeling method is used to reconstruct the three-dimensional region subspace. The three-dimensional reconstruction model corresponding to each region subspace is obtained and combined to obtain the three-dimensional model of the instruction region. Obtain indoor navigation commands, perform natural language recognition on the indoor navigation commands, and obtain several location description terms based on the recognition results; In the acquired location description terms, arbitrarily select a feature location term. If there is a unique location in the 3D model of the instruction area that coincides with the location referred to by the feature location term, then mark the feature location term as a clear location term. If there is no unique location in the 3D model of the instruction area that coincides with the location referred to by the feature location term, then mark the feature location term as an ambiguous location term. Each location description term is categorized to obtain location term analysis data.

[0012] Furthermore, in step S3, the specific steps are as follows: Step S31: Obtain indoor navigation instructions, sort the location description terms according to their order of appearance in the indoor navigation instructions, and obtain the location description time series; Step S32: Obtain location terminology analysis data, acquire ambiguous location terms and clear location terms based on the location terminology analysis data, and perform semantic clarification on ambiguous location terms.

[0013] Furthermore, in step S32, the specific steps are as follows: Step S321: Randomly select an ambiguous term P1 from the collected ambiguous location terms, mark the ambiguous term P1 as an ambiguous segmentation node in the location description time series, mark the location description time series segment before the ambiguous segmentation node as the first location subsequence, and mark the location description time series segment after the ambiguous segmentation node as the second location subsequence. Step S322: If any position subsequence in the first position subsequence or the second position subsequence does not contain a clear position term, the control mobile terminal adopts a completely passive interaction mode to complete the ambiguity clarification of the P1 ambiguous term with the instruction initiating user until the P1 ambiguous term becomes a clear position term. Step S323: If there are explicit positional terms in both the first positional subsequence and the second positional subsequence, perform model positional analysis on the ambiguous term P1. Based on the analysis results, clarify the semantics of the ambiguous term P1 through conditional passive interaction until the ambiguous term P1 is an explicit positional term.

[0014] Furthermore, in step S323, the specific steps are as follows: Obtain the last explicitly stated position term in the first position subsequence to obtain the first explicitly stated position term; obtain the first explicitly stated position term in the second position subsequence to obtain the second explicitly stated position term. Obtain the spatial location of the first explicit location term in the three-dimensional model of the instruction area, and obtain the location referred to by the first term. Obtain the spatial location of the second explicit location term in the three-dimensional model of the instruction area, and obtain the location referred to by the second term. In the 3D model of the instruction area, a model path is generated from the location referred to by the first term to the location referred to by the second term, resulting in multiple preset model paths. The model locations contained in each preset model path are collected to obtain multiple candidate instruction locations. Based on the candidate command location, the control mobile terminal clarifies the ambiguity of P1 ambiguous terms with the command user through conditional passive interaction.

[0015] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are: 1. This invention can select an appropriate model creation method based on the characteristics of the indoor scene. It can adaptively select reverse entity modeling or point cloud mesh modeling according to the complexity of the distribution of sampling points in the indoor subspace, reduce the component fitting deviation and loss of structured semantics, improve the matching effect between location description terms and spatial points, and help improve the reliability of the ambiguity clarification process. 2. Based on the parsed speech semantic content, this invention flexibly selects the semantic clarification method, distinguishing between active reasoning-based clarification, conditional passive interaction, and completely passive interaction. When conditions permit, it minimizes unnecessary user inquiries and reasonably enables user interaction in ambiguous scenarios that are not suitable for autonomous reasoning, thereby reducing the risk of incorrectly deducing the target location and improving the adaptability of semantic clarification in different ambiguous scenarios. Attached Figure Description

[0016] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0017] Figure 1 This is a diagram illustrating the implementation steps of the present invention; Figure 2 This is a schematic diagram of the sampling neighborhood radiation circle of the present invention; Figure 3 This is a schematic diagram of the included angle between the centers of neighboring points in this invention. Detailed Implementation

[0018] The technical solution of the present invention will be clearly and completely described below with reference to the embodiments. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0019] Example 1 Please see Figure 1 This invention provides a technical solution: a method for proactively clarifying ambiguities in intelligent indoor navigation commands, comprising the following steps: Step S1: Obtain indoor navigation instructions and acquire the indoor space corresponding to the indoor navigation instructions to obtain the pre-selected instruction navigation space. Perform laser point cloud acquisition on each subspace of the pre-selected instruction navigation space and analyze the corresponding horizontal distribution complexity and vertical distribution complexity of the point cloud to obtain instruction area analysis data. In step S1, the specific steps are as follows: The system acquires the real-time indoor navigation commands output by the user, obtains the real-time navigation output commands, and acquires the indoor space associated with the real-time navigation output commands, obtains the pre-selected command navigation space. It should be noted here that: In this application, the pre-selected instruction navigation space specifically refers to the spatial range involved in this navigation being determined based on the internal real-time navigation instruction after the indoor navigation system completes instruction ambiguity disambiguation and generates a definite navigation task instruction.

[0020] The preselected instruction navigation space is divided into several regional subspaces, and one T1 subspace is arbitrarily selected from the divided regional subspaces; It should be noted here that: In this application, the regional subspace referred to herein specifically refers to multiple independent yet interconnected local spatial units obtained by dividing the preselected instruction navigation space according to spatial connectivity and geographical boundaries.

[0021] Laser point cloud acquisition and analysis were performed on the T1 subspace. Based on the analysis results, the horizontal distribution complexity and vertical distribution complexity of the point cloud corresponding to the T1 subspace were obtained. Specifically as follows: The T1 subspace is scanned using a laser scanning terminal, and the three-dimensional point cloud set corresponding to the T1 subspace is obtained based on the scanning results. It should be noted here that: In this application, the laser scanning terminal referred to herein is a lidar.

[0022] The point cloud envelope space corresponding to the three-dimensional point cloud set is obtained to obtain the pre-selected point cloud envelope space; It should be noted here that: In this application, the point cloud envelope space is a three-dimensional closed space formed by the maximum and minimum values ​​of the coordinates of all laser sampling points in the three-dimensional point cloud set on the X, Y, and Z coordinate axes. All point cloud sampling points are located inside the point cloud envelope space.

[0023] In the preselected point cloud envelope space, the bottom surface of the space is marked as the first preselected spatial plane. A plane perpendicular to the first preselected spatial plane is constructed to obtain the second preselected spatial plane. It should be noted here that: In this application, the bottom surface of the preselected point cloud envelope space corresponds to the ground surface in the T1 subspace.

[0024] In the pre-selected point cloud envelope space, several spatial sections parallel to the first pre-selected spatial plane are created to obtain multiple lateral spatial sections. The position of the laser sampling points in the lateral spatial sections is analyzed, and the lateral distribution complexity of the point cloud is obtained based on the analysis results.

[0025] Specifically as follows: Arbitrarily select a Q1 spatial section from the obtained spatial sections, and acquire the laser sampling points contained in the Q1 spatial section to obtain multiple cross-sectional laser sampling points; Please see Figure 2 Arbitrarily select a feature laser sampling point from the obtained cross-sectional laser sampling points, and use the feature laser sampling point as the center to create the largest inscribed circle of the Q1 spatial cross-section, thus obtaining the sampling neighborhood radiation circle. Mark multiple preset points on the circumference of the sampling neighborhood radiation circle. It should be noted here that: Please see Figure 3 In this application, the lines connecting two adjacent circumferential preset points and the feature laser sampling point are respectively marked as the first circle center line and the second circle center line. The angle formed by the first circle center line and the second circle center line at the feature laser sampling point is marked as the neighborhood point circle center angle. The angle values ​​corresponding to the circle center angles formed by different adjacent circumferential preset points are all constant values.

[0026] Connect each preset point on the circumference with the feature laser sampling point to obtain multiple circumference center lines. Randomly select one sample center line from the obtained circumference center lines.

[0027] The laser sampling points of the cross section covered by the sample center line are obtained, resulting in multiple line-covered sampling points. The line distance value between each line-covered sampling point and the feature laser sampling point is obtained, and the minimum line distance value is marked as the neighborhood sampling distance corresponding to the sample center line. Repeatedly sample the neighborhood distance corresponding to the center line of the sample, obtain the neighborhood sampling distance corresponding to each center line of the circle, and calculate the average of the obtained multiple neighborhood sampling distances to obtain the local neighborhood average distance corresponding to the feature laser sampling point. Repeat the process of obtaining the average distance of the local neighborhood corresponding to the feature laser sampling point, obtain the average distance of the local neighborhood corresponding to the cross-sectional laser sampling point respectively, and calculate the variance of the obtained average distance of the local neighborhood to obtain the variance of the neighborhood sampling distance.

[0028] The neighborhood sampling distance variance is obtained by collecting spatial cross sections in history, resulting in multiple historical neighborhood distance variances. The largest historical neighborhood distance variance is marked as the neighborhood comparison distance variance. The ratio of the neighborhood sampling distance variance to the neighborhood comparison distance variance is calculated to obtain the historical peak ratio of the neighborhood variance corresponding to the Q1 spatial cross section. The number of laser sampling points contained in the Q1 spatial section is counted to obtain the cumulative value of laser sampling. The product of the cumulative value of laser sampling and the ratio of the historical peak value of the neighborhood variance is calculated to obtain the laser sampling complexity corresponding to the Q1 spatial section. Repeat the process of obtaining the laser sampling complexity corresponding to the Q1 spatial section, obtain the laser sampling complexity corresponding to each horizontal spatial section, and compare the obtained laser sampling complexity numerically. Mark the maximum laser sampling complexity as the horizontal distribution complexity of the point cloud.

[0029] In the pre-selected point cloud envelope space, several spatial sections parallel to the second pre-selected spatial plane are created to obtain multiple longitudinal spatial sections. The position of the laser sampling points in the longitudinal spatial sections is analyzed, and the longitudinal distribution complexity of the point cloud is obtained based on the analysis results. Repeat the process of obtaining the horizontal and vertical distribution complexity of the point cloud corresponding to the T1 subspace, and obtain the horizontal and vertical distribution complexity of the point cloud corresponding to each region subspace to obtain the instruction region analysis data.

[0030] Step S2: Based on the instruction area analysis data, perform three-dimensional reconstruction of the pre-selected instruction navigation space to obtain a three-dimensional model of the instruction area. Semantically segment the indoor navigation instructions to obtain multiple location description terms. Classify the location description terms by type through the instruction area three-dimensional model to obtain location term analysis data. Step S2 includes the following steps: Obtain instruction region analysis data, and based on the instruction region analysis data, obtain the horizontal distribution complexity and vertical distribution complexity of the point cloud corresponding to each region subspace.

[0031] Create a complex metric coordinate system. In the complex metric coordinate system, coordinate points are created with the horizontal distribution complexity of the point cloud as the x-axis and the vertical distribution complexity of the point cloud as the y-axis, resulting in multiple complexity distribution coordinate points. Obtain the distance values ​​between the coordinate points of each complexity distribution and the origin, and obtain the spatial complexity quantification value corresponding to each region subspace; A spatial complexity baseline is created. If the spatial complexity quantification value is greater than the spatial complexity baseline value, the point cloud mesh modeling method is used to reconstruct the three-dimensional region subspace. If the spatial complexity quantification value is less than or equal to the spatial complexity baseline value, the inverse solid modeling method is used to reconstruct the three-dimensional region subspace. The three-dimensional reconstruction model corresponding to each region subspace is obtained and combined to obtain the three-dimensional model of the instruction region.

[0032] It should be noted here that: In this application, spatial complexity quantification values ​​are obtained for the historical region subspace that has been reconstructed using the reverse entity modeling method, resulting in multiple historical spatial complexities. The maximum historical spatial complexity is set as the spatial complexity benchmark value.

[0033] In this application, the reverse solid modeling method takes 3D point cloud data obtained by laser survey as input, extracts the geometric features of building components such as walls, floors, doors and windows from the point cloud envelope space, and generates parametric building components with independent attributes through plane fitting and contour extraction. The components carry structured information such as location, thickness, floor, and room number. Non-building objects such as furniture and miscellaneous items are filtered out, and a structured interior model is output. The point cloud mesh modeling method processes the original 3D laser point cloud, and generates triangular meshes with semantic labels after noise reduction and downsampling of the point cloud. It does not construct independent parametric building component objects, and fully preserves all geometric details of on-site equipment, partitions, and miscellaneous items.

[0034] Obtain indoor navigation commands, perform natural language recognition on the indoor navigation commands, and obtain several location description terms based on the recognition results.

[0035] It should be noted here that: In this application, the location description terms involved are semantic fragments extracted after performing natural language recognition on indoor navigation instructions, used to represent spatial location, reference objects or orientation relationships, and used to match points and areas in the indoor space model, and do not contain action words such as walking or going to. If the indoor navigation command is "Go to Conference Room 1 on the fourth floor", then the location terms that can be identified here are "fourth floor" and "Conference Room 1".

[0036] In the acquired location description terms, arbitrarily select a feature location term. If there is a unique location in the 3D model of the instruction area that coincides with the location referred to by the feature location term, then mark the feature location term as a clear location term. If there is no unique location in the 3D model of the instruction area that coincides with the location referred to by the feature location term, then mark the feature location term as an ambiguous location term. It should be noted here that: In this application, the specific location term is defined as follows: in the three-dimensional model of the instruction area, there is only one spatial location that can match and overlap with the target location expressed by the feature location term, and there will be no multiple candidate points.

[0037] Repeat the process of classifying the terminology for the characteristic location, classifying each location description term separately to obtain location terminology analysis data.

[0038] Step S3: Clarify the semantics of ambiguous terms based on the 3D model of the instruction area; In step S3, the specific steps are as follows: Obtain indoor navigation instructions, sort the location description terms according to their order of appearance in the indoor navigation instructions, and obtain the location description time series; Acquire location terminology analysis data, and based on the location terminology analysis data, obtain ambiguous location terms and explicit location terms, and arbitrarily select one P1 ambiguous term from the collected ambiguous location terms; The ambiguous term P1 is marked as an ambiguous segmentation node in the location description time series. The location description time series segment before the ambiguous segmentation node is marked as the first location subsequence, and the location description time series segment after the ambiguous segmentation node is marked as the second location subsequence. If any position subsequence in the first position subsequence or the second position subsequence does not contain a clear position term, the controlling mobile terminal adopts a completely passive interaction mode to complete the ambiguity clarification of the P1 ambiguous term with the user who initiated the instruction until the P1 ambiguous term becomes a clear position term. If there are explicit positional terms in both the first positional subsequence and the second positional subsequence, then model positional analysis is performed on the ambiguous term P1. Based on the analysis results, the ambiguous term P1 is semantically clarified through conditional passive interaction until the ambiguous term P1 is an explicit positional term.

[0039] It should be noted here that: In this application, the completely passive interaction method involved here is specifically that when multiple candidate locations are obtained by matching the location description terms, the system cannot rely on internal information such as spatial model, user positioning, and spatial topology to perform any filtering and screening on the candidate set, and cannot narrow down the candidate range. The control mobile terminal adopts a completely passive interaction method to initiate an inquiry to the user, and all candidate locations are handed over to the user for identification and selection. Based on the user's feedback, the ambiguity of the P1 ambiguous terms is clarified, and the unique navigation target location is determined. The conditional passive interaction method involved here is as follows: when multiple candidate locations are obtained by matching the location description terms, the system can use the existing spatial information to filter out some candidate locations. However, after the filtering process, two or more candidate command locations are still left, and it is impossible to converge to obtain a unique target. At this time, the control mobile terminal uses the conditional passive interaction method to output the query content to the command user, and then completes the ambiguity clarification based on the feedback results given by the user to determine the final navigation target location.

[0040] Specifically as follows: Obtain the last explicitly stated position term in the first position subsequence to obtain the first explicitly stated position term; obtain the first explicitly stated position term in the second position subsequence to obtain the second explicitly stated position term. Obtain the spatial location of the first explicit location term in the three-dimensional model of the instruction area, and obtain the location referred to by the first term. Obtain the spatial location of the second explicit location term in the three-dimensional model of the instruction area, and obtain the location referred to by the second term. In the 3D model of the instruction region, a model path is generated from the location referred to by the first term to the location referred to by the second term, resulting in multiple preset model paths. The model locations contained in each preset model path are collected to obtain multiple candidate instruction locations. Based on the candidate command location, the control mobile terminal clarifies the ambiguity of P1 ambiguous terms with the command user through conditional passive interaction.

[0041] It should be noted here that: The mobile terminal is used to receive indoor navigation commands issued by the user. The mobile terminal can conduct voice interaction with the user, and can collect and receive the voice navigation commands issued by the user, and can also output voice inquiry content to push to the user to clarify ambiguous terms. It can also receive voice response information from the user to provide interactive data for subsequent indoor navigation processing.

[0042] Repeat the process of semantic clarification for the ambiguous terms in P1, and perform semantic clarification for each ambiguous term separately.

[0043] Compared to the problems described in the background art, the present invention can select an appropriate model creation method based on the characteristics of the indoor scene. It can adaptively select reverse entity modeling or point cloud mesh modeling according to the complexity of the distribution of sampling points in the indoor subspace, reduce the component fitting deviation and loss of structured semantics, improve the matching effect between location description terms and spatial points, and help improve the reliability of the ambiguity clarification process. Furthermore, this invention flexibly selects semantic clarification methods based on the parsed speech semantic content, distinguishing between active reasoning-based clarification, conditional passive interaction, and completely passive interaction. When conditions permit, it minimizes unnecessary user inquiries, while reasonably enabling user interaction in ambiguous scenarios that are not suitable for autonomous reasoning, reducing the risk of incorrectly deducing the target location and improving the adaptability of semantic clarification in different ambiguous scenarios.

[0044] The preferred embodiments of the present invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to specific implementation methods. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention. The present invention is limited only to the technical solutions and their full scope and equivalents.

Claims

1. A method for proactively clarifying ambiguities in spatial intelligent indoor navigation commands, characterized in that, include: Step S1: Obtain indoor navigation instructions and acquire the indoor space corresponding to the indoor navigation instructions to obtain the pre-selected instruction navigation space. Perform laser point cloud acquisition on each subspace of the pre-selected instruction navigation space, analyze and obtain the horizontal distribution complexity and vertical distribution complexity of the point cloud, and obtain instruction area analysis data. Step S2: Based on the instruction area analysis data, perform three-dimensional reconstruction of the pre-selected instruction navigation space to obtain a three-dimensional model of the instruction area. Semantically segment the indoor navigation instructions to obtain multiple location description terms. Classify the location description terms by type through the instruction area three-dimensional model to obtain location term analysis data. Step S3: Clarify the semantics of ambiguous terms based on the three-dimensional model of the instruction area.

2. The method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 1, characterized in that, The specific steps in step S1 are as follows: Step S11: Obtain the indoor navigation commands output by the user in real time to obtain the real-time navigation output commands, and obtain the indoor space associated with the real-time navigation output commands to obtain the pre-selected command navigation space. Step S12: Divide the preselected instruction navigation space into several regional subspaces, and arbitrarily select one T1 subspace from the divided regional subspaces; Step S13: Collect and analyze laser point cloud data in the T1 subspace, and obtain the horizontal and vertical distribution complexity of the point cloud corresponding to the T1 subspace based on the analysis results. Step S14: Obtain the horizontal distribution complexity and vertical distribution complexity of the point cloud corresponding to each region subspace to obtain the instruction region analysis data.

3. The method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 2, characterized in that, In step S13, the specific steps are as follows: Step S131: Use a laser scanning terminal to perform laser scanning on the T1 subspace, and obtain the three-dimensional point cloud set corresponding to the T1 subspace based on the scanning results; Step S132: Obtain the point cloud envelope space corresponding to the three-dimensional point cloud set to obtain the pre-selected point cloud envelope space; Step S133: In the preselected point cloud envelope space, mark the first preselected spatial plane and the second preselected spatial plane; Step S134: In the pre-selected point cloud envelope space, create several spatial sections parallel to the first pre-selected spatial plane to obtain multiple horizontal spatial sections. Perform position analysis on the laser sampling points in the horizontal spatial sections to obtain the horizontal distribution complexity of the point cloud. Step S135: In the pre-selected point cloud envelope space, create several spatial sections parallel to the second pre-selected spatial plane to obtain multiple longitudinal spatial sections. Perform position analysis on the laser sampling points in the longitudinal spatial sections to obtain the longitudinal distribution complexity of the point cloud.

4. The method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 3, characterized in that, In step S134, the specific steps are as follows: Arbitrarily select a Q1 spatial section from the obtained spatial sections, and acquire the laser sampling points contained in the Q1 spatial section to obtain multiple cross-sectional laser sampling points; Arbitrarily select a characteristic laser sampling point from the obtained cross-sectional laser sampling points, and use the characteristic laser sampling point as the center to create the largest inscribed circle within the Q1 spatial cross-section, thus obtaining the sampling neighborhood radiation circle. Mark multiple preset points on the circumference of the sampling neighborhood radiation circle. Connect each preset point on the circumference with the feature laser sampling point to obtain multiple circumference center lines. Randomly select one sample center line from the obtained circumference center lines. The laser sampling points of the cross section covered by the line connecting the sample centers are obtained, resulting in multiple line-covered sampling points. The line distance value between each line-covered sampling point and the feature laser sampling point is obtained, and the minimum line distance value is marked as the neighborhood sampling distance.

5. The method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 4, characterized in that, In step S134, the specific steps are as follows: Obtain the neighborhood sampling distance corresponding to each line connecting the centers of the circumference, and calculate the average of the multiple neighborhood sampling distances to obtain the local neighborhood average distance. Obtain the local neighborhood average distance corresponding to the laser sampling point of the cross section, and calculate the variance of the obtained local neighborhood average distance to obtain the neighborhood sampling distance variance; The neighborhood sampling distance variance is obtained by collecting spatial cross sections in history, resulting in multiple historical neighborhood distance variances. The largest historical neighborhood distance variance is marked as the neighborhood comparison distance variance. The ratio of the neighborhood sampling distance variance to the neighborhood comparison distance variance is calculated to obtain the historical peak ratio of neighborhood variance. The number of laser sampling points contained in the Q1 spatial section is counted to obtain the cumulative value of laser sampling. The product of the cumulative value of laser sampling and the ratio of the historical peak value of the neighborhood variance is calculated to obtain the laser sampling complexity. Obtain the laser sampling complexity corresponding to each horizontal spatial section, compare the obtained laser sampling complexity numerically, and mark the maximum laser sampling complexity as the horizontal distribution complexity of the point cloud.

6. The method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 1, characterized in that, Step S2 includes the following steps: Obtain instruction region analysis data, and based on the instruction region analysis data, obtain the horizontal distribution complexity and vertical distribution complexity of the point cloud corresponding to each region subspace; Create a complex metric coordinate system. In the complex metric coordinate system, coordinate points are created with the horizontal distribution complexity of the point cloud as the x-axis and the vertical distribution complexity of the point cloud as the y-axis, resulting in multiple complexity distribution coordinate points. Obtain the distance values ​​between the coordinate points of each complexity distribution and the origin, and obtain the spatial complexity quantification value corresponding to each region subspace.

7. The method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 6, characterized in that, In step S2 The steps include the following: A spatial complexity baseline is created. If the spatial complexity quantification value is greater than the spatial complexity baseline value, the point cloud mesh modeling method is used to reconstruct the three-dimensional region subspace. If the spatial complexity quantification value is less than or equal to the spatial complexity baseline value, the reverse solid modeling method is used to reconstruct the three-dimensional region subspace. The three-dimensional reconstruction model corresponding to each region subspace is obtained and combined to obtain the three-dimensional model of the instruction region. Obtain indoor navigation commands, perform natural language recognition on the indoor navigation commands, and obtain several location description terms; In the acquired location description terms, arbitrarily select a feature location term. If there is a unique location in the 3D model of the instruction area that coincides with the location referred to by the feature location term, then mark the feature location term as a clear location term. If there is no unique location in the 3D model of the instruction area that coincides with the location referred to by the feature location term, then mark the feature location term as an ambiguous location term. Each location description term is categorized to obtain location term analysis data.

8. The method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 1, characterized in that, The specific steps in step S3 are as follows: Step S31: Obtain indoor navigation instructions, sort the location description terms according to their order of appearance in the indoor navigation instructions, and obtain the location description time series; Step S32: Obtain location terminology analysis data, acquire ambiguous location terms and clear location terms based on the location terminology analysis data, and perform semantic clarification on ambiguous location terms.

9. The method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 8, characterized in that, The specific steps in step S32 are as follows: Step S321: Randomly select an ambiguous term P1 from the collected ambiguous location terms, mark the ambiguous term P1 as an ambiguous segmentation node in the location description time series, mark the location description time series segment before the ambiguous segmentation node as the first location subsequence, and mark the location description time series segment after the ambiguous segmentation node as the second location subsequence. Step S322: If any position subsequence in the first position subsequence or the second position subsequence does not contain a clear position term, the control mobile terminal adopts a completely passive interaction mode to complete the ambiguity clarification of the P1 ambiguous term with the instruction initiating user until the P1 ambiguous term becomes a clear position term. Step S323: If there are explicit positional terms in both the first positional subsequence and the second positional subsequence, then perform model positional analysis on the ambiguous term P1, and clarify the semantics of the ambiguous term P1 through conditional passive interaction until the ambiguous term P1 is an explicit positional term.

10. A method for proactively clarifying ambiguities in intelligent indoor navigation commands according to claim 9, characterized in that, In step S323, the specific steps are as follows: Obtain the last explicitly stated position term in the first position subsequence to obtain the first explicitly stated position term; obtain the first explicitly stated position term in the second position subsequence to obtain the second explicitly stated position term. Obtain the spatial location of the first explicit location term in the three-dimensional model of the instruction area, and obtain the location referred to by the first term. Obtain the spatial location of the second explicit location term in the three-dimensional model of the instruction area, and obtain the location referred to by the second term. In the 3D model of the instruction area, a model path is generated from the location referred to by the first term to the location referred to by the second term, resulting in multiple preset model paths. The model locations contained in each preset model path are collected to obtain multiple candidate instruction locations. Based on the candidate command location, the control mobile terminal clarifies the ambiguity of P1 ambiguous terms with the command user through conditional passive interaction.

Images