A high-precision point cloud reconstruction system with spatial intelligence depth consistency and geometric correction

Abstract

The application discloses a kind of space intelligence depth consistent and geometric correction high-precision point cloud reconstruction system, it is related to space intelligence field, solve the problem of insufficient accuracy of present point cloud model reconstruction, including: multi-modal perception module: data acquisition is carried out to space environment;Consistency modeling module: space environment is segmented, and unit model is obtained;Consistency optimization module: unit model is analyzed to overlap, and unit model is optimized;Self-adapting correction module: construct space residual graph, generate compensation field, and carry out self-adapting adjustment based on compensation field;Structure fusion and balance module: construct overall point cloud model, and optimize fusion is carried out to model;Feedback optimization module: point cloud reconstruction is carried out to overall point cloud model, compare the residual change of space residual graph, and feedback optimization is carried out to point cloud reconstruction;Check output module: consistency check is carried out to overall point cloud model;The application can effectively improve the accuracy of model reconstruction.

Description

Technical Field

[0001] This invention belongs to the field of spatial intelligence and relates to three-dimensional reconstruction technology. Specifically, it is a high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction. Background Technology

[0002] Existing point cloud reconstruction methods for spatial environments have the following specific drawbacks:

[0003] 1. Traditional multi-view Figure 3 3D reconstruction methods (such as algorithms based on stereo matching or voxel fusion) typically rely on disparity estimation and pixel-by-pixel depth prediction, but the geometric consistency between depth maps from different viewpoints is difficult to guarantee, which in turn affects the overall accuracy and spatial consistency of the reconstruction results.

[0004] 2. Multimodal sensors (such as RGB cameras, depth cameras, and LiDAR) are susceptible to acquisition anomalies (such as lens distortion, attitude shift, and depth noise) during data acquisition, resulting in insufficient accuracy of the acquired data. This makes it difficult for point cloud models to guarantee the accuracy and stability of modeling in high-precision scenarios (such as industrial inspection and digital twin modeling).

[0005] 3. Existing point cloud models lack error feedback and self-correction mechanisms. Current systems mainly focus on error backpropagation for error processing and perform geometric correction through error backpropagation. In practical applications, the feedback update efficiency of backpropagation error is insufficient, and the error will accumulate and amplify in multiple rounds of fusion, ultimately affecting the overall geometric consistency and spatial alignment accuracy.

[0006] To address this, we propose a high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction. Summary of the Invention

[0007] To address the shortcomings of existing technologies, the purpose of this invention is to provide a high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction, aiming to improve the accuracy of spatial modeling.

[0008] To achieve the above objectives, the present invention adopts the following technical solution: a high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction, the specific working process of each module is as follows:

[0009] Multimodal perception module: Collects data on the space environment to obtain space observation information, and fuses the space observation information to obtain a space feature tensor;

[0010] Consistency Modeling Module: Semantically segment the spatial environment to obtain spatial units; extract the spatial feature tensor of each spatial unit; and model the spatial unit based on the spatial feature tensor to obtain the unit model;

[0011] Consistency Optimization Module: Based on the spatial environment, a world coordinate system is constructed; the unit model is mapped onto the world coordinate system; overlap analysis is performed on the unit model to obtain the overlapping distortion region, and the unit model is optimized.

[0012] Adaptive correction module: It performs distortion modeling based on overlapping distortion regions to obtain a spatial residual map; it generates a compensation field based on the spatial residual map and performs adaptive adjustment based on the compensation field;

[0013] Structure fusion and balancing module: Collects point cloud data from all unit models to construct an overall point cloud model; analyzes the overall point cloud model and optimizes and fuses the model.

[0014] Feedback optimization module: Based on the optimization of the unit model and the overall point cloud model, the module performs point cloud reconstruction on the overall point cloud model, compares the residual changes of the spatial residual map, and performs feedback optimization on the point cloud reconstruction.

[0015] Verification output module: Performs consistency verification on the overall point cloud model.

[0016] Furthermore, data collection on the space environment will be conducted, specifically as follows:

[0017] The system statistically analyzes the sensors within the space environment, synchronizes the data collected by the sensors using a high-precision clock, and obtains the time-series data of each sensor. Statistical analysis of the time-series data yields the space observation information of the space environment.

[0018] The spatial observation information is traversed to extract data features, resulting in geometric and semantic features.

[0019] The geometric distribution of the spatial environment is recorded based on geometric features, semantic features are associated with the geometric distribution, and semantic features and geometric features are combined to construct the spatial feature tensor of the spatial environment.

[0020] Furthermore, the spatial units are modeled as follows:

[0021] Semantic features are extracted based on the spatial feature tensor; the spatial environment is semantically segmented based on the semantic features, dividing the spatial environment into multiple spatial units; spatial feature tensors are collected based on the spatial units, and the sensors corresponding to the spatial feature tensors are recorded; the sensors are statistically analyzed to construct a multi-view list.

[0022] The sensors are traversed using a multi-view list to extract monitoring data for each spatial unit. Spatial models are then created for each spatial unit based on the monitoring data, resulting in local models. These local models are statistically analyzed, and the local models corresponding to the monitoring data from all sensors are fused. The local models corresponding to different sensors are then input, and a preliminary depth estimate is generated using a deep network. A unified 3D depth field is constructed, and the local models of each sensor are projected onto the 3D depth field for alignment. Consistent modeling of the spatial units is then performed to obtain the unit model.

[0023] Furthermore, the unit model is optimized as follows:

[0024] Based on the spatial environment, a coordinate system is constructed to obtain the world coordinate system; unit models are obtained and mapped to the world coordinate system to obtain the mapped models; the mapped models of different unit models are statistically analyzed to determine the overlap relationship of unit models in the world coordinate system, and the overlap distortion region is obtained based on the overlap relationship between unit models.

[0025] The overlapping element models are extracted based on the overlapping distortion regions to obtain the boundary positions of the element models. The boundary positions of the element models are combined with the overlapping distortion regions to perform local geometric reconstruction of the element models and optimize them.

[0026] Furthermore, local geometric reconstruction is performed on the unit model, as follows:

[0027] The sensors corresponding to each unit model are acquired, and their positions in the world coordinate system are collected to obtain reference coordinates. The monitoring data of the sensors on the unit model are acquired, and based on the reference coordinates and the monitoring data, the unit model is mapped to the world coordinate system. The coordinate mapping of the unit model is statistically analyzed to obtain the mapped model.

[0028] The mapping model is traversed to extract the overlapping parts of the mapping model, thus obtaining the overlapping distortion region; the mapping model is then searched based on the overlapping distortion region to extract the associated model of the overlapping distortion region.

[0029] Obtain the edge coordinates of the overlapping distortion region to obtain the distortion boundary; obtain the edge coordinates of the associated model to obtain the model boundary; perform continuous reconstruction of the overlapping region of the associated model based on the model boundary, and offset the distortion boundary to the model boundary; perform boundary offset on all associated models in the overlapping distortion region, perform local geometric reconstruction of the model based on the offset result, and optimize the unit model based on the local geometric reconstruction.

[0030] Furthermore, adaptive adjustments are made based on the compensation field, as follows:

[0031] The mapping model on the world coordinate system is traversed to collect the overlapping distortion regions of the mapping model; the coordinate points of each overlapping distortion region on the world coordinate system are statistically analyzed, and the overlapping distortion regions are modeled based on the coordinate points to obtain the spatial residual map.

[0032] Obtain the geometric reconstruction results of the unit model, extract the coordinate changes of the overlapping distortion region based on the geometric reconstruction of the unit model, combine the spatial residual map corresponding to the overlapping distortion region, generate the supplementary field of the spatial residual map, feed the supplementary field back to the sensor, and adaptively adjust the supplementary field.

[0033] Furthermore, the supplementary field for generating the spatial residual map is as follows:

[0034] Based on the geometric reconstruction of the unit model, a compensation field is generated for the spatial residual map. The coordinate compensation of each coordinate on the spatial residual map is recorded through the compensation field of the spatial residual map. The sensors corresponding to the coordinates on the spatial residual map are extracted. Data feedback is given to the sensors based on the coordinate compensation.

[0035] Multiple data acquisitions are performed on the sensor. Spatial modeling is carried out based on the data acquisition results. The coordinate compensation results at different time series are recorded. The coordinate compensation is statistically analyzed to obtain a time series compensation list. Linear regression is performed on the coordinate compensation based on the time series compensation list. Based on the regression results, the coordinate compensation is adjusted in time series. The coordinate compensation on the spatial residual map is statistically analyzed, and the compensation field is adaptively adjusted.

[0036] Furthermore, the model is optimized and fused, as follows:

[0037] Data is collected from all unit models based on the compensation field to obtain point cloud data of the unit models; the point cloud data of the unit models is mapped to the world coordinate system, and the mapped point cloud data is statistically analyzed to construct the overall point cloud model.

[0038] The point clouds between unit models are analyzed based on the overall point cloud model. The point cloud data within the same unit model are traversed, and the coordinate distance between the point cloud data is extracted. Based on the coordinate distance, the point cloud density of the unit model is obtained. The surface point cloud of the same unit model is collected, and the surface curvature of the unit model is collected based on the surface point cloud.

[0039] The unit model is traversed, and the fusion status between adjacent unit models is judged based on the point cloud density and surface curvature of the unit model. Point cloud adjustment is performed based on the fusion status between unit models, point cloud alignment is performed on the unit models, and point cloud optimization is performed on the overall point cloud model based on the point cloud alignment results.

[0040] Furthermore, feedback optimization is performed on the point cloud reconstruction, as follows:

[0041] The time-series data of the sensors are acquired, and the unit model and the overall point cloud model are optimized based on the time sequence of the sensors. The point cloud is reconstructed based on the optimization results. The spatial residual map after each reconstruction is extracted according to the point cloud reconstruction. The reconstructed spatial residual map is spatially compared with the spatial residual map before reconstruction to obtain the residual change of the spatial residual map.

[0042] The compensation field of the spatial residual map is obtained, and the compensation field of the spatial residual map is adjusted based on the residual changes. The compensation field provides feedback to the overall point cloud model, thereby optimizing the point cloud reconstruction of the overall point cloud model.

[0043] Furthermore, a consistency check is performed on the overall point cloud model, as follows:

[0044] The process involves acquiring spatial observation information of the space environment, extracting geometric features of the space environment based on the spatial observation information, comparing the geometric features of the space environment with the overall point cloud model, verifying the geometric consistency of the overall point cloud model, extracting semantic features of the space environment based on the spatial observation information, combining the semantic features of the space environment with the corresponding geometric features, retrieving the overall point cloud environment, and verifying the semantic consistency of the overall point cloud model.

[0045] The overall point cloud model is output based on the results of geometric consistency verification and semantic consistency verification.

[0046] In summary, due to the adoption of the above technical solution, the beneficial effects of the present invention are:

[0047] 1. This invention traverses the sensors, applies depth consistency constraints based on the sensor monitoring data, establishes a global geometric consistency mapping between different viewpoints, aligns the multi-view depth results in a unified world coordinate system, improves the accuracy of model construction, and ensures the reconstruction efficiency and stability of the model.

[0048] 2. This invention analyzes the unit model, judges the sensing distortion based on the overlap relationship between the unit models, learns and corrects the distortion characteristics, feeds back the sensing distortion to the sensor, and performs high-precision control on the sensor's sensing data to ensure the accuracy of the sensing data.

[0049] 3. This invention reconstructs the model based on the temporal relationship of sensors, feeds back the model error based on the reconstructed model, and analyzes the model error in conjunction with the sensor to realize an error feedback loop optimization mechanism. After each round of reconstruction, it automatically analyzes the global and local error distribution, generates residual information and performs feedback optimization, and performs geometric consistency verification and semantic consistency verification on the overall model to ensure the geometric accuracy and semantic accuracy of the model, and ensure the accuracy of the output. Attached Figure Description

[0050] To facilitate understanding by those skilled in the art, the present invention will be further described below with reference to the accompanying drawings.

[0051] Figure 1 This is a functional block diagram of the present invention;

[0052] Figure 2 This is a schematic diagram of the model construction in this invention;

[0053] Figure 3 This is a schematic diagram of the distortion processing of the present invention; Detailed Implementation

[0054] It should be understood that the specific embodiments described herein are merely illustrative of the invention and are not intended to limit the invention.

[0055] This application provides a high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction. The executing entity of this high-precision point cloud reconstruction system includes, but is not limited to, at least one of the following electronic devices that can be configured to execute the system provided in this application: a server, a terminal, etc. In other words, the high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction can be executed by software or hardware installed on a terminal device or a server device. The server includes, but is not limited to, a single server, a server cluster, a cloud server, or a cloud server cluster.

[0056] Reference Figure 1 The diagram shown is a functional block diagram of a high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction provided in an embodiment of the present invention. In this embodiment, the high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction includes: a multimodal perception module, a consistency modeling module, a consistency optimization module, an adaptive correction module, a structure fusion and balancing module, a feedback optimization module, and a verification output module. The module described in this invention can also be called a unit, which refers to a series of computer program segments that can be executed by the processor of an electronic device and can perform a fixed function, and are stored in the memory of the electronic device.

[0057] In this embodiment of the invention, the functions of each module / unit are as follows:

[0058] Multimodal perception module: Collects data on the space environment through multi-source sensors to obtain space observation information (including visible light images, depth maps, laser point clouds, IMU attitude, GNSS positioning and semantic tags), extracts geometric and semantic features from the space observation information, and fuses the geometric and semantic features to obtain a spatial feature tensor;

[0059] The specific workflow of the multimodal perception module is as follows:

[0060] The system statistically analyzes the sensors within the space environment, synchronizes the data collected by the sensors using a high-precision clock, and obtains the time-series data of each sensor. Statistical analysis of the time-series data yields the space observation information of the space environment.

[0061] The spatial observation information is traversed to extract data features, resulting in geometric and semantic features.

[0062] It should be noted that geometric features include depth gradient, surface normal, and rate of change of curvature; semantic features include category labels, material properties, and spatial function descriptions.

[0063] The geometric distribution of the spatial environment is recorded based on geometric features, semantic features are associated with the geometric distribution, and semantic features and geometric features are combined to construct the spatial feature tensor of the spatial environment.

[0064] It should be noted that the spatial feature tensor refers to the feature set containing semantic features and their corresponding geometric features; it reflects the semantic diversity and geometric correlation of the spatial environment.

[0065] Consistency Modeling Module: Based on semantic features, the spatial environment is semantically segmented to obtain spatial units. The spatial feature tensor of each spatial unit is extracted. Based on the spatial feature tensor, the geometric features of the spatial units are fused to perform consistency modeling on the spatial units to obtain unit models.

[0066] The specific workflow of the consistency modeling module is as follows:

[0067] Please see Figure 2 Based on the spatial feature tensor, semantic features are extracted; based on the semantic features, the spatial environment is semantically segmented into multiple spatial units, and spatial feature tensors are collected based on the spatial units to record the sensors corresponding to the spatial feature tensors; the sensors are statistically analyzed to construct a multi-view list.

[0068] The sensors are traversed using a multi-view list to extract monitoring data for each spatial unit. Spatial models are then created for each spatial unit based on the monitoring data, resulting in local models. These local models are statistically analyzed, and the local models corresponding to the monitoring data from all sensors are fused. The local models corresponding to different sensors are then input, and a preliminary depth estimate is generated using a deep network. A unified 3D depth field is constructed, and the local models of each sensor are projected onto the 3D depth field for alignment. Consistent modeling of the spatial units is then performed to obtain the unit model.

[0069] Consistency Optimization Module: Based on the spatial environment, a world coordinate system is constructed; the unit model is mapped onto the world coordinate system using sensors; overlap analysis is performed on the mapped unit model to obtain the overlapping distortion region; based on the overlapping distortion region, local geometric reconstruction is performed in conjunction with the overlapping unit model to optimize the unit model;

[0070] The specific workflow of the consistency optimization module is as follows:

[0071] Please see Figure 3 Based on the spatial environment, a coordinate system is constructed to obtain the world coordinate system; unit models are obtained and mapped to the world coordinate system to obtain the mapping model; the mapping models of different unit models are statistically analyzed to determine the overlap relationship of unit models in the world coordinate system, and the overlap distortion region is obtained based on the overlap relationship between unit models.

[0072] The overlapping element models are extracted based on the overlapping distortion regions to obtain the boundary positions of the element models. The boundary positions of the element models are combined with the overlapping distortion regions to perform local geometric reconstruction of the element models and optimize them.

[0073] The local geometric reconstruction of the unit model is performed as follows:

[0074] The sensors corresponding to each unit model are acquired, and their positions in the world coordinate system are collected to obtain reference coordinates. The monitoring data of the sensors on the unit model are acquired, and based on the reference coordinates and the monitoring data, the unit model is mapped to the world coordinate system. The coordinate mapping of the unit model is statistically analyzed to obtain the mapped model.

[0075] The mapping model is traversed to extract the overlapping parts of the mapping model, thus obtaining the overlapping distortion region; the mapping model is then searched based on the overlapping distortion region to extract the associated model of the overlapping distortion region.

[0076] It should be noted that the associated model refers to all mapping models involved in the overlapping distortion region.

[0077] Obtain the edge coordinates of the overlapping distortion region to obtain the distortion boundary; obtain the edge coordinates of the associated model to obtain the model boundary; perform continuous reconstruction of the overlapping region of the associated model based on the model boundary, and offset the distortion boundary to the model boundary; perform boundary offset on all associated models in the overlapping distortion region, perform local geometric reconstruction of the model based on the offset result, and optimize the unit model based on the local geometric reconstruction.

[0078] Adaptive correction module: It collects data on overlapping distortion regions, performs distortion modeling based on these regions, and obtains a spatial residual map; it generates a compensation field based on the spatial residual map; it correlates the compensation field with the sensor and adaptively adjusts the compensation field based on changes in the sensor parameters.

[0079] The specific workflow of the adaptive correction module is as follows:

[0080] The mapping model on the world coordinate system is traversed to collect the overlapping distortion regions of the mapping model; the coordinate points of each overlapping distortion region on the world coordinate system are statistically analyzed, and the overlapping distortion regions are modeled based on the coordinate points to obtain the spatial residual map.

[0081] Obtain the geometric reconstruction results of the unit model, extract the coordinate changes of the overlapping distortion region based on the geometric reconstruction of the unit model, combine the spatial residual map corresponding to the overlapping distortion region, generate the supplementary field of the spatial residual map, feed the supplementary field back to the sensor, and adaptively adjust the supplementary field.

[0082] The supplementary field for generating the spatial residual map is as follows:

[0083] Based on the geometric reconstruction of the unit model, a compensation field is generated for the spatial residual map. The coordinate compensation of each coordinate on the spatial residual map is recorded through the compensation field of the spatial residual map. The sensors corresponding to the coordinates on the spatial residual map are extracted. Data feedback is given to the sensors based on the coordinate compensation.

[0084] Multiple data acquisitions are performed on the sensor. Spatial modeling is carried out based on the data acquisition results. The coordinate compensation results at different time series are recorded. The coordinate compensation is statistically analyzed to obtain a time series compensation list. Linear regression is performed on the coordinate compensation based on the time series compensation list. Based on the regression results, the coordinate compensation is adjusted in time series. The coordinate compensation on the spatial residual map is statistically analyzed, and the compensation field is adaptively adjusted.

[0085] It should be noted that by analyzing the monitoring data from the sensors, the accuracy of the sensor data is improved, ensuring the accuracy of data acquisition, thereby improving the modeling accuracy of the unit model. Through the high-precision modeling, the overall spatial environment can be accurately and effectively analyzed and judged.

[0086] The structural fusion and balancing module collects point cloud data from all unit models, maps the point cloud data to the world coordinate system, and constructs an overall point cloud model. Based on the overall point cloud model, it performs point cloud analysis, samples and adjusts the overall point cloud model, and optimizes and fuses the overall point cloud model.

[0087] The specific workflow of the structural integration and balancing module is as follows:

[0088] Data is collected from all unit models based on the compensation field to obtain point cloud data of the unit models; the point cloud data of the unit models is mapped to the world coordinate system, and the mapped point cloud data is statistically analyzed to construct the overall point cloud model.

[0089] The point clouds between unit models are analyzed based on the overall point cloud model. The point cloud data within the same unit model are traversed, and the coordinate distance between the point cloud data is extracted. Based on the coordinate distance, the point cloud density of the unit model is obtained. The surface point cloud of the same unit model is collected, and the surface curvature of the unit model is collected based on the surface point cloud.

[0090] The unit model is traversed, and the fusion status between adjacent unit models is judged based on the point cloud density and surface curvature of the unit model. Point cloud adjustment is performed based on the fusion status between unit models, point cloud alignment is performed on the unit models, and point cloud optimization is performed on the overall point cloud model based on the point cloud alignment results.

[0091] Feedback optimization module: Based on the optimization of the unit model and the overall point cloud model, the module performs point cloud reconstruction on the overall point cloud model, records the spatial residual map of each point cloud reconstruction, and compares the residual changes of the spatial residual map after reconstruction; and performs feedback optimization on point cloud reconstruction based on the residual changes.

[0092] The specific workflow of the feedback optimization module is as follows:

[0093] The time-series data of the sensors are acquired, and the unit model and the overall point cloud model are optimized based on the time sequence of the sensors. The point cloud is reconstructed based on the optimization results. The spatial residual map after each reconstruction is extracted according to the point cloud reconstruction. The reconstructed spatial residual map is spatially compared with the spatial residual map before reconstruction to obtain the residual change of the spatial residual map.

[0094] The compensation field of the spatial residual map is obtained, and the compensation field of the spatial residual map is adjusted based on the residual changes. The compensation field provides feedback to the overall point cloud model, thereby optimizing the point cloud reconstruction of the overall point cloud model.

[0095] Verification output module: Performs consistency verification on the overall point cloud model and outputs the overall point cloud model based on the verification results;

[0096] The specific workflow of the verification output module is as follows:

[0097] The process involves acquiring spatial observation information of the space environment, extracting geometric features of the space environment based on the spatial observation information, comparing the geometric features of the space environment with the overall point cloud model, verifying the geometric consistency of the overall point cloud model, extracting semantic features of the space environment based on the spatial observation information, combining the semantic features of the space environment with the corresponding geometric features, retrieving the overall point cloud environment, and verifying the semantic consistency of the overall point cloud model.

[0098] The overall point cloud model is output based on the results of geometric consistency verification and semantic consistency verification.

[0099] Compared to the problems described in the background technology, this invention improves the accuracy of model construction and ensures the reconstruction efficiency and stability of the model by traversing the sensors and applying depth consistency constraints based on the sensor monitoring data. It establishes a global geometric consistency mapping between different viewpoints, aligning multi-view depth results in a unified world coordinate system. Furthermore, this invention analyzes the unit models, judges sensor distortion based on the overlap relationship between unit models, learns and corrects distortion features, and feeds back sensor distortion to the sensors, performing high-precision control of sensor data to ensure its accuracy. Finally, this invention reconstructs the model based on the temporal relationship of the sensors, feeds back model errors based on the reconstructed model, and analyzes model errors in conjunction with sensor data to achieve an error feedback loop optimization mechanism. After each round of reconstruction, it automatically analyzes the global and local error distribution, generates residual information, and performs feedback optimization. It also performs geometric consistency and semantic consistency checks on the overall model, ensuring the geometric and semantic accuracy of the model and the accuracy of the output. Therefore, the high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction provided by this invention can improve the accuracy of model reconstruction.

[0100] It will be apparent to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential characteristics of the present invention.

[0101] Finally, it should be noted that deleting any one of the above embodiments does not affect the technical solutions of other embodiments. The above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention.

Claims

1. A high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction, characterized in that, include: Multimodal perception module: Collects data on the space environment to obtain space observation information, and fuses the space observation information to obtain a space feature tensor; Consistency Modeling Module: Semantically segment the spatial environment to obtain spatial units; extract the spatial feature tensor of each spatial unit; and model the spatial unit based on the spatial feature tensor to obtain the unit model; Consistency optimization module: Constructs a world coordinate system based on the spatial environment; Map the element model onto the world coordinate system; perform overlap analysis on the element model to obtain the overlapping distortion region, and optimize the element model; Adaptive correction module: Based on the overlapping distortion region, distortion modeling is performed to obtain a spatial residual map; a compensation field is generated based on the spatial residual map, and adaptive adjustment is performed based on the compensation field; Adaptive adjustment is based on the compensation field, as detailed below: The mapping model on the world coordinate system is traversed to collect the overlapping distortion regions of the mapping model; the coordinate points of each overlapping distortion region on the world coordinate system are statistically analyzed, and the overlapping distortion regions are modeled based on the coordinate points to obtain the spatial residual map. Obtain the geometric reconstruction results of the unit model, extract the coordinate changes of the overlapping distortion region based on the geometric reconstruction of the unit model, combine the spatial residual map corresponding to the overlapping distortion region, generate the supplementary field of the spatial residual map, feed the supplementary field back to the sensor, and adaptively adjust the supplementary field. The supplementary field for generating the spatial residual map is as follows: Based on the geometric reconstruction of the unit model, a compensation field is generated for the spatial residual map. The coordinate compensation of each coordinate on the spatial residual map is recorded through the compensation field of the spatial residual map. The sensors corresponding to the coordinates on the spatial residual map are extracted. Data feedback is given to the sensors based on the coordinate compensation. Multiple data acquisitions are performed on the sensor. Spatial modeling is carried out based on the data acquisition results. The coordinate compensation results under different time series are recorded. The coordinate compensation is statistically analyzed to obtain a time series compensation list. Linear regression is performed on the coordinate compensation based on the time series compensation list. Based on the regression results, the coordinate compensation is adjusted over time. The coordinate compensation on the spatial residual map is statistically analyzed, and the compensation field is adaptively adjusted. Structural Fusion and Balancing Module: Collects point cloud data from all unit models to construct an overall point cloud model; analyzes the overall point cloud model and optimizes and fuses the model. Feedback optimization module: Based on the optimization of the unit model and the overall point cloud model, the module performs point cloud reconstruction on the overall point cloud model, compares the residual changes of the spatial residual map, and performs feedback optimization on the point cloud reconstruction. Verification output module: Performs consistency verification on the overall point cloud model.

2. The high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction according to claim 1, characterized in that, Data collection of the space environment is carried out as follows: Statistical analysis is performed on the sensors within the space environment, and the data collected by the sensors is synchronized in time to obtain the time-series data of each sensor; statistical analysis is performed on the time-series data to obtain the space observation information of the space environment. The space observation information is traversed to extract data features, resulting in geometric and semantic features. The geometric distribution of the spatial environment is recorded based on geometric features, semantic features are associated with the geometric distribution, and semantic features and geometric features are combined to construct the spatial feature tensor of the spatial environment.

3. The high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction according to claim 2, characterized in that, The spatial unit is modeled as follows: Semantic features are extracted based on the spatial feature tensor; Semantic segmentation of the spatial environment is performed, dividing the spatial environment into multiple spatial units. Spatial feature tensors are collected based on the spatial units, and the sensors corresponding to the spatial feature tensors are recorded. Statistical analysis of the sensors is performed to construct a multi-view list; The sensors are traversed using a multi-view list to extract monitoring data for each spatial unit. Spatial models are then created for each spatial unit based on the monitoring data, resulting in local models. These local models are statistically analyzed, and the local models corresponding to the monitoring data from all sensors are fused. The local models corresponding to different sensors are then input, and a preliminary depth estimate is generated using a deep network. A unified 3D depth field is constructed, and the local models of each sensor are projected onto the 3D depth field for alignment. Consistent modeling of the spatial units is then performed to obtain the unit model.

4. The high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction according to claim 3, characterized in that, The unit model is optimized as follows: Based on the spatial environment, a coordinate system is constructed to obtain the world coordinate system; unit models are obtained and mapped to the world coordinate system to obtain the mapped models; the mapped models of different unit models are statistically analyzed to determine the overlap relationship of unit models in the world coordinate system, and the overlap distortion region is obtained based on the overlap relationship between unit models. The overlapping element models are extracted based on the overlapping distortion regions to obtain the boundary positions of the element models. The boundary positions of the element models are combined with the overlapping distortion regions to perform local geometric reconstruction of the element models and optimize them.

5. A high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction according to claim 4, characterized in that, Local geometric reconstruction of the unit model is performed as follows: The sensors corresponding to each unit model are acquired, and their positions in the world coordinate system are collected to obtain reference coordinates. The monitoring data of the sensors on the unit model are acquired, and based on the reference coordinates and the monitoring data, the unit model is mapped to the world coordinate system. The coordinate mapping of the unit model is statistically analyzed to obtain the mapped model. The mapping model is traversed to extract the overlapping parts of the mapping model, thus obtaining the overlapping distortion region; the mapping model is then searched based on the overlapping distortion region to extract the associated model of the overlapping distortion region. Obtain the edge coordinates of the overlapping distortion region to obtain the distortion boundary; obtain the edge coordinates of the associated model to obtain the model boundary; perform continuous reconstruction of the overlapping region of the associated model based on the model boundary, and offset the distortion boundary to the model boundary; perform boundary offset on all associated models in the overlapping distortion region, perform local geometric reconstruction of the model based on the offset result, and optimize the unit model based on the local geometric reconstruction.

6. The high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction according to claim 1, characterized in that, The models are optimized and fused as follows: Data is collected from all unit models based on the compensation field to obtain point cloud data of the unit models; the point cloud data of the unit models is mapped to the world coordinate system, and the mapped point cloud data is statistically analyzed to construct the overall point cloud model. The point cloud between unit models is analyzed, the point cloud data within the same unit model is traversed, the coordinate distance between the point cloud data is extracted, and the point cloud density of the unit model is obtained based on the coordinate distance. The surface point cloud of the same unit model is collected, and the surface curvature of the unit model is also collected. The unit model is traversed, and the fusion status between adjacent unit models is judged based on the point cloud density and surface curvature of the unit model. Point cloud regulation is performed based on the fusion status between unit models, point cloud alignment is performed on unit models, and point cloud optimization is performed on the overall point cloud model.

7. A high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction according to claim 6, characterized in that, Feedback optimization is performed on point cloud reconstruction, as follows: The time-series data of the sensors are acquired, and the unit model and the overall point cloud model are optimized based on the time sequence of the sensors. The overall point cloud model is then reconstructed. Based on the point cloud reconstruction, the spatial residual map after each reconstruction is extracted. The reconstructed spatial residual map is then compared with the original spatial residual map to obtain the residual change of the spatial residual map. The compensation field of the spatial residual map is obtained, and the compensation field of the spatial residual map is adjusted based on the residual changes. The compensation field provides feedback to the overall point cloud model, thereby optimizing the point cloud reconstruction of the overall point cloud model.

8. The high-precision point cloud reconstruction system with spatial intelligent depth consistency and geometric correction according to claim 1, characterized in that, The consistency of the overall point cloud model is verified as follows: The process involves acquiring spatial observation information of the space environment, extracting geometric features of the space environment based on the spatial observation information, comparing the geometric features of the space environment with the overall point cloud model, verifying the geometric consistency of the overall point cloud model, extracting semantic features of the space environment based on the spatial observation information, combining the semantic features of the space environment with the corresponding geometric features, retrieving the overall point cloud environment, and verifying the semantic consistency of the overall point cloud model. The overall point cloud model is output based on the results of geometric consistency verification and semantic consistency verification.

Images