A method for reconstructing a three-dimensional building using a radar point cloud

Abstract

This invention discloses a method for reconstructing 3D buildings using radar point clouds. The invention includes the following steps: First, the timestamps of the point cloud and the image are aligned using visual offsets to obtain a mapping relationship between the point cloud and the image; next, the point cloud with continuous time trajectories is stitched and cropped according to pose to obtain the final building point cloud; the point cloud is filtered using a reflectivity threshold, and then architectural semantic features are extracted and distinguished using various architectural prior knowledge to generate a structured facade information representation; finally, the facade information is converted into an architectural information model, and texture features are projected to ultimately form a textured 3D building model. This invention enables the direct generation of 3D building models from point clouds, combining real-time processing capabilities with innovative processing strategies. It is widely applicable to multiple fields such as architectural design, urban planning, and cultural heritage protection, providing a new, efficient, and accurate approach for 3D building reconstruction.

Description

Technical Field

[0001] This invention relates to a point cloud 3D reconstruction method in the field of 3D modeling technology, specifically a method for rapid 3D building reconstruction using radar point clouds. Background Technology

[0002] In the field of modern 3D spatial data processing, point clouds serve as the core carrier, providing detailed geometric and topological information for the accurate modeling of complex buildings. The main obstacles to this technology include the massive scale of data, noise interference, and the difficulty in analyzing complex structures. This drives researchers to continuously innovate and develop efficient and robust algorithms to optimize point cloud processing workflows and ensure the accuracy of 3D reconstruction. For example, traditional methods based on geometric features have gradually transitioned to using deep learning techniques, especially Graph Convolutional Networks (GCNs), which effectively capture the local structural features of point clouds, demonstrating particularly outstanding performance on large-scale datasets.

[0003] In terms of point cloud acquisition technology, Structure from Motion (SfM) and Multi-View Stereo (MVS) have become mainstream. The former utilizes multi-image feature matching to reconstruct sparse point clouds, while the latter generates high-density point clouds through stereo matching of multi-view images, jointly supporting the basic framework of architectural 3D models. Building upon this foundation, advanced technologies such as Poisson surface reconstruction and texture mapping can directly construct highly detailed 3D mesh models from point clouds. Although these models still have shortcomings, with advancements in semantic understanding and post-optimization techniques, their integrity is being significantly improved.

[0004] After acquiring point clouds, most work involves semantic segmentation of the point clouds before reconstruction. However, the reconstruction results are often composed of multiple stacked cubes, and the surface texture is achieved only through simple projection, which is still far from a high-quality building information model.

[0005] Digital building models are demonstrating immense potential in numerous cutting-edge applications, such as digital urban management, game development, navigation map construction, and autonomous driving simulation training. To enhance the realism and usability of these models, research is focusing on the deep fusion of images and point clouds. For instance, deep learning-driven image feature extraction and texture mapping strategies are being used to directly integrate high-resolution textures into digital building models, significantly improving their realism and market application value. Furthermore, the integration of real-time rendering technology and ray tracing algorithms is further pushing the boundaries of model interactivity and visual effects, providing users with an immersive experience. Summary of the Invention

[0006] To address the problems existing in the background technology, this invention provides a method for reconstructing three-dimensional buildings using radar point clouds, used for constructing three-dimensional building models. This invention is applicable to building data with existing LiDAR point clouds and images, and is particularly suitable for applications in architectural design, urban planning, and cultural heritage preservation.

[0007] The technical solution adopted in this invention includes the following steps:

[0008] I. A method for reconstructing 3D buildings using radar point clouds

[0009] S1: By utilizing the offset effect between point cloud and image, timestamp matching is performed between point cloud and image to obtain the mapping relationship between point cloud and image; texture features of image are also extracted.

[0010] S2: Based on the pose information on the continuous time trajectory, the point cloud is stitched and cropped to extract the complete building point cloud; then the reflectivity threshold filtering mechanism is used to filter the complete building point cloud to obtain the processed building point cloud.

[0011] S3: A noise-resistant density-based spatial clustering algorithm is used to perform semantic recognition on the processed building point cloud to obtain various building semantic features; then, combined with the building structural attributes and the identified building semantic features, the facade structure of the processed building point cloud is refined to obtain complete building facade information.

[0012] S4: Convert the complete building facade information into a building information model, and then add texture features to the building information model according to the mapping relationship between point cloud and image, thereby generating a three-dimensional building model with realistic texture.

[0013] In step S1, feature points in the image and feature points in the point cloud are identified, the feature points in the image are matched with the corresponding features in the point cloud, and then the spatiotemporal deviation between the two is minimized by an optimization algorithm to obtain the mapping relationship between the point cloud and the image.

[0014] In step S2, a reflectivity threshold filtering mechanism is used to filter the complete building point cloud to obtain a processed building point cloud, including:

[0015] The reflectance value I in the complete building point cloud p Point cloud p with reflectivity less than the reflectivity threshold T is removed.

[0016] In step S3, the facade structure of the processed building point cloud is refined by combining the building structural attributes and the identified building semantic features to obtain complete building facade information, specifically:

[0017] Based on the repetition and equal spacing of architectural semantic features in the building structure, the missing and erroneous features in the facade structure of the processed building point cloud are fine-tuned to obtain complete building facade information.

[0018] In step S4, the complete building facade information is converted into a building information model using the Blender modeling tool.

[0019] In step S4, based on the mapping relationship between point cloud and image, the texture features of the image are attached to the corresponding surface of the BIM model using image projection. Then, the realism of the model is enhanced by adjusting the lighting model and materials, thereby generating a three-dimensional building model with realistic texture.

[0020] II. A computer device

[0021] The computer device includes a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the method.

[0022] III. A computer-readable storage medium

[0023] A computer program is stored on a computer-readable storage medium, which, when executed by a processor, implements the steps of the method.

[0024] IV. A computer program product

[0025] The computer program product includes a computer program / instruction that, when executed by a processor, implements the steps of the method.

[0026] The beneficial effects of this invention are:

[0027] Precise spatiotemporal matching enhances texture mapping quality: By applying a visual offset effect between point clouds and images, this invention achieves high-precision matching between point clouds and image timestamps, establishing a fine correspondence between point clouds and images. This not only ensures the accuracy of model texture mapping but also greatly enhances the realism and detail reproduction of the reconstructed model, especially in complex lighting and dynamic environments.

[0028] Efficient point cloud processing and structure optimization: Guided by continuous time trajectories, the point cloud is intelligently stitched and cropped. Combined with a reflectivity threshold filtering mechanism, noise interference is effectively filtered out, preserving pure building structure information. This strategy not only improves the quality of the point cloud but also accelerates subsequent processing, resulting in a clearer and more accurate reconstructed model.

[0029] Advanced semantic understanding and refined modeling: The advanced noise-resistant density clustering algorithm DBSCAN is used to cluster point clouds that may contain specific semantic features such as doors, windows, and roofs. Based on the point clouds and architectural expertise, the building structure is further constructed and the facade details are further refined. This not only deepens the structural understanding of the model, but also provides a richer information foundation for subsequent architectural design, analysis and management.

[0030] Realistic Texture Enhancement in BIM Model Generation: Utilizing professional modeling tools such as Blender and combining image projection technology, this invention can precisely match textures and apply them to the corresponding surfaces of a BIM model. Through meticulous adjustments to lighting and materials, the realism and visual appeal of the model are greatly enhanced. This process not only simplifies the tedious steps of traditional BIM texture addition but also gives the model greater visual expressiveness and practicality.

[0031] Automated processing: The method proposed in this invention can be transformed into an executable software solution through computer equipment and computer-readable storage media, which greatly promotes the automation process of 3D building modeling, reduces the cost and error rate of manual intervention, and improves overall work efficiency. Attached Figure Description

[0032] Figure 1 This is a general flowchart of a method for reconstructing a three-dimensional building using radar point clouds, according to an embodiment of the present invention.

[0033] Figure 2 This is a schematic diagram illustrating the image point cloud timestamp alignment implemented by the present invention.

[0034] Figure 3 This is a rendering of the model reconstructed by the present invention. Detailed Implementation

[0035] The technical solution of the present invention will now be clearly and completely described with reference to the accompanying drawings. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0036] To make the objectives, technical solutions, and advantages of the present invention clearer, the embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0037] like Figure 1 As shown, the present invention includes the following steps:

[0038] S1: Utilize the visual offset effect between point cloud and image to accurately match the timestamps of point cloud and image, thereby obtaining the mapping relationship between point cloud and image to support texture mapping of subsequent models; and extract texture features of image.

[0039] Specifically: First, build a platform that can simultaneously visualize point clouds and images. Then, adapt it to a program capable of displaying both point cloud sequences and image sequences, aligning them according to their respective timestamps, such as... Figure 2 As shown. Through observation, there is a significant misalignment between the timestamps of the point cloud sequence and the salience sequence. Assuming their difference is Δt, the relationship between the timestamps of the point cloud and the image can be expressed as:

[0040] T cam =T lidar +Δt

[0041] Among them, T cam For the timestamp of the camera shot, T lidar This is the timestamp for the lidar.

[0042] Since the camera pose is calculated based on the radar pose and the radar-to-camera transformation matrix, the camera pose will change when the timestamp of the camera relative to the radar changes. Here, the radar pose is interpolated to index the radar pose associated with the camera from a finer-grained timestamp. Finally, the camera pose can be represented as:

[0043] P cam =P lidar ×T l2c

[0044] min|T cam -T lidar -Δt|

[0045] Among them, P cam It refers to the camera's pose, P. lidar It is the radar's pose, T l2c This is the transformation matrix from radar to camera, where || represents the absolute value. The selection of the radar-camera timestamp difference Δt depends on how closely the texture features of the image are closest to the texture features of the radar point cloud at different time intervals.

[0046] S2: Based on the pose information on the continuous time trajectory, the point cloud is stitched and cropped to extract the complete building point cloud; then the reflectivity threshold filtering mechanism is used to filter the complete building point cloud to reduce noise interference and obtain the processed building point cloud.

[0047] Specifically, the point cloud is subjected to octree downsampling and pruning operations, where the point cloud is pruned to retain only the central building. The point cloud filtering step based on the reflectivity threshold screening mechanism further includes:

[0048] I p <T

[0049] Among them, I p This represents the reflectance value of a single point p in the point cloud, where T is the set reflectance threshold.

[0050] S3: A noise-resistant density-based spatial clustering algorithm is used to perform semantic recognition on the processed building point cloud to obtain various building semantic features. At the same time, the building semantic features are distinguished, including doors, windows, roofs, etc. Then, combined with the building structural attributes and the identified building semantic features, the facade structure of the processed building point cloud is refined to obtain complete building facade information.

[0051] Specifically:

[0052] First, the RANSAC (Random Sample Consensus) algorithm is used to extract two planes, the building facade and the ground, from the initial point cloud, and the parameters of these two planes are retained.

[0053] Next, the DBSCAN algorithm is used to automatically discover and distinguish point cloud clusters within different density regions by setting appropriate neighborhood distances and density thresholds, effectively differentiating the different components of a building. The DBSCAN process is as follows:

[0054] P1,P2,...,P n =DBSCAN(P,ε,minPts)

[0055] Where P represents the complete point cloud output, ε represents the neighborhood size, and minPts is the minimum number of points in the neighborhood required to define a point as a core point, P1, P2, ..., P n This refers to the n point clouds that are finally clustered.

[0056] After completing the point cloud clustering process, the implementation steps of this invention involve projecting the optimized and integrated point cloud onto a two-dimensional building facade plane to construct a basic facade schematic diagram of the building. Subsequently, by identifying and analyzing the significant vertical gaps between different levels in this facade schematic diagram, different floor structures are systematically defined. For these defined floors, the inherent regularities of the building facade, namely repetitive patterns and equal spacing characteristics, are utilized to further refine and segment window elements. Notably, this method specifically marks the bottom floor of the building as the entrance area, while the top floor corresponds to the roof structure, thus ensuring a clear semantic distinction between the first and last floors. Finally, based on this refined structured information, this invention generates a comprehensive JSON file that fully and comprehensively encodes the structural features and semantic tags of the entire building, providing highly standardized and information-based data support for subsequent analysis, simulation, and visualization.

[0057] S4: Using the Blender modeling tool, the complete building facade information is converted into a Building Information Model (BIM). Then, based on the mapping relationship between point clouds and images, texture features are added to the BIM to generate a 3D building model with realistic textures. Specifically, based on the mapping relationship between point clouds and images, image projection is used to apply the texture features of the image to the corresponding surfaces of the BIM model. Then, lighting and material adjustments are made to enhance the realism of the model, thus generating a 3D building model with realistic textures. Figure 3 As shown.

[0058] In texture mapping, due to the large number of images involved, a portion of the images were randomly selected for texture mapping. The image poses used in texture mapping were optimized in step S1.

[0059] The present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps of the method for reconstructing a three-dimensional building using radar point clouds.

[0060] The present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method for reconstructing a three-dimensional building using radar point clouds.

[0061] This invention enables the direct generation of 3D building models from point clouds. It combines real-time processing capabilities with innovative processing strategies, making it widely applicable to various fields such as architectural design, urban planning, and cultural heritage protection. It provides a new approach for 3D building reconstruction that is both efficient and accurate.

[0062] Finally, it should be noted that the above embodiments and descriptions are only used to illustrate the technical solutions of the present invention and not to limit it. 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 disclosure of the technical solutions of the present invention, and all such modifications and substitutions should be covered within the protection scope of the claims of the present invention.

Claims

1. A method for reconstructing a three-dimensional building using radar point clouds, characterized in that, Includes the following steps: S1: By utilizing the offset effect between point cloud and image, timestamp matching is performed between point cloud and image to obtain the mapping relationship between point cloud and image; texture features of image are also extracted. S2: Based on the pose information on the continuous time trajectory, the point cloud is stitched and cropped to extract the complete building point cloud; then the reflectivity threshold filtering mechanism is used to filter the complete building point cloud to obtain the processed building point cloud. S3: A noise-resistant density-based spatial clustering algorithm is used to perform semantic recognition on the processed building point cloud to obtain various building semantic features; then, combined with the building structural attributes and the identified building semantic features, the facade structure of the processed building point cloud is refined to obtain complete building facade information. The facade structure of the processed building point cloud is refined by combining building structural attributes and the identified building semantic features to obtain complete building facade information, specifically as follows: Based on the repetition and equal spacing of architectural semantic features in the architectural structure, the missing and erroneous features in the facade structure of the processed architectural point cloud are fine-tuned to obtain complete architectural facade information. S4: Convert the complete building facade information into a building information model, and then add texture features to the building information model according to the mapping relationship between point cloud and image, thereby generating a three-dimensional building model with realistic texture. In step S4, based on the mapping relationship between point cloud and image, the texture features of the image are attached to the corresponding surface of the BIM model using image projection. Then, the realism of the model is enhanced by adjusting the lighting model and materials, thereby generating a three-dimensional building model with realistic texture.

2. The method for reconstructing a three-dimensional building using radar point clouds according to claim 1, characterized in that, In step S1, feature points in the image and feature points in the point cloud are identified, the feature points in the image are matched with the corresponding features in the point cloud, and then the spatiotemporal deviation between the two is minimized by an optimization algorithm to obtain the mapping relationship between the point cloud and the image.

3. The method for reconstructing a three-dimensional building using radar point clouds according to claim 1, characterized in that, In step S2, a reflectivity threshold filtering mechanism is used to filter the complete building point cloud to obtain a processed building point cloud, including: Remove point cloud elements (i) whose reflectance values ​​are less than the reflectance threshold from the complete building point cloud.

4. The method for reconstructing a three-dimensional building using radar point clouds according to claim 1, characterized in that, In step S4, the complete building facade information is converted into a building information model using the Blender modeling tool.

5. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method according to any one of claims 1 to 4.

6. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the steps of the method according to any one of claims 1 to 4.

7. A computer program product comprising a computer program / instructions, characterized in that, When the computer program / instructions are executed by the processor, they implement the steps of the method described in any one of claims 1 to 4.

Images