Environment situation scanning three-dimensional reconstruction method based on infrared and SLAM

Abstract

The invention discloses an environment situation scanning three-dimensional reconstruction method based on infrared and SLAM, and belongs to the technical field of computer vision and three-dimensional reconstruction. According to the method, thermal radiation information and environment image information of a target environment are obtained through infrared sensing and visual imaging respectively, after feature extraction and matching processing are conducted on the two kinds of information, synchronous positioning and map construction are completed, and environment pose data and environment map data are generated; a sensing uncertainty probability model is established, after multi-source data denoising is completed through Bayesian filtering, feature level fusion is carried out to obtain fusion environment data, and finally a three-dimensional reconstruction model of a target environment is constructed based on the fusion environment data. According to the method, the full-process closed loop of environment situation scanning and three-dimensional reconstruction is realized, the application limitation of single sensing information is made up, the comprehensiveness of environment sensing data and the scene adaptability of three-dimensional reconstruction are improved, and the digital reconstruction requirements in various complex environments can be met.

Description

Technical Field

[0001] This invention relates to the field of computer vision and 3D reconstruction technology, and in particular to a 3D reconstruction method based on infrared and SLAM environmental situation scanning. Background Technology

[0002] With the rapid development of artificial intelligence, robotics, and the drone industry, the market demand for environmental situational awareness and 3D spatial digital reconstruction technologies in complex environments continues to expand. These technologies have formed large-scale application scenarios in various industries such as spatial mapping, emergency inspection, urban digital modeling, and resource exploration, becoming one of the core underlying technologies supporting the digital application of spatial information. Currently, environmental situational awareness and 3D reconstruction technologies have formed an industry pattern of parallel development across multiple technical paths. Visual imaging-based 3D reconstruction technology, infrared sensing-based environmental feature perception technology, and SLAM-based simultaneous localization and mapping technology have all formed complete theoretical research systems and engineering application solutions. Various technologies have completed multiple rounds of technical iteration and scenario adaptation optimization in their corresponding sub-application scenarios. At the same time, multi-sensor information fusion and multi-technology collaborative applications have become important research and development directions in this field. Various multi-source sensor fusion environmental perception schemes and multi-technology linked 3D reconstruction processes are continuously proposed and verified, providing diversified technical implementation paths for the full-dimensional and refined acquisition of environmental spatial information, and also promoting the continuous improvement of related industries' capabilities in environmental situational awareness and visualization.

[0003] In the development and application of existing technologies, various environmental situation scanning and 3D reconstruction technologies mostly focus on the performance optimization of single technical links, making it difficult to form a closed-loop collaborative system integrating multi-source sensor information and multiple technical links. They fail to deeply integrate infrared sensor information acquisition, visual imaging information acquisition, SLAM simultaneous localization and mapping, and 3D model reconstruction into a streamlined process. This results in significant disconnects between the environmental information acquisition, processing, fusion, and model building stages, making it impossible to complete the entire process from target environmental information acquisition to 3D reconstruction model output through a unified technical workflow. Furthermore, existing technologies lack systematic solutions for handling uncertainties in the sensor acquisition process during multi-source environmental information processing. They cannot uniformly suppress and optimize noise in the acquired environmental information, pose data generated by simultaneous localization, and map data. This leads to a lack of reliable data foundation for the fusion of multi-source data, making it difficult to achieve effective feature-level fusion of multi-source environmental information and fully leverage the complementary advantages of different sensor information in environmental perception. Furthermore, the existing 3D reconstruction process is difficult to achieve a deep technical connection with the synchronous positioning and mapping process. It is impossible to directly build a high-precision 3D model based on the output of positioning and mapping and multi-source fused environmental data. As a result, the reconstructed 3D model is difficult to completely and accurately restore the actual situation of the target environment, and cannot meet the application requirements of environmental situation scanning and 3D reconstruction in multiple scenarios. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM.

[0005] The objective of this invention is achieved through the following technical solution: A method for 3D reconstruction of environmental situation scanning based on infrared and SLAM is provided, which includes the following steps: S1. Acquire thermal radiation information and environmental image information corresponding to the target environment. The thermal radiation information is acquired through infrared sensing, and the environmental image information is acquired through visual imaging. S2. Extract features from thermal radiation information and environmental image information to obtain image feature points, perform matching processing on image feature points, and perform synchronous localization and map construction based on the matched image feature points to generate environmental pose data and environmental map data. S3. Establish a sensing uncertainty probability model, and perform Bayesian filtering on thermal radiation information, environmental image information, environmental pose data and environmental map data based on the sensing uncertainty probability model to complete the denoising operation of multi-source data. Then, perform fusion processing on the denoised multi-source data to obtain fused environmental data. S4. Construct a 3D environmental model based on the fused environmental data to generate a 3D reconstruction model corresponding to the target environment.

[0006] Furthermore, step S1 includes the following sub-steps: S1.1. Conduct full-range information collection on the target environment. During the collection process, a synchronous triggering method is used to simultaneously acquire thermal radiation information and environmental image information, ensuring that the collection frame rate of thermal radiation information and environmental image information remains consistent. S1.2. Timestamp the acquired thermal radiation information and environmental image information, and perform timestamp alignment processing on the thermal radiation information and environmental image information based on a unified time reference, so that the thermal radiation information and environmental image information maintain frame-by-frame correspondence in the time dimension. S1.3. Based on the pre-completed joint sensor calibration parameters, perform spatial coordinate transformation processing on the time-stamp aligned thermal radiation information and environmental image information, so that the thermal radiation information and environmental image information are mapped to the same spatial coordinate system, and complete the spatial coordinate matching processing.

[0007] Furthermore, step S2 includes the following sub-steps: S2.1. Perform image preprocessing on thermal radiation information and environmental image information. The preprocessing includes image distortion correction, image grayscale conversion, and image enhancement. Image distortion correction is performed based on pre-calibrated sensor intrinsic parameters. Image enhancement is performed to adjust the contrast of thermal radiation information and environmental image information under low light conditions. S2.2. Feature point detection is performed on the preprocessed thermal radiation information and environmental image information to generate an initial feature point set. Non-maximum suppression processing and response threshold filtering processing are performed on the initial feature point set to obtain effective image feature points. Feature descriptors corresponding to the effective image feature points are generated simultaneously. S2.3. Perform similarity matching processing on the feature descriptors of effective image feature points corresponding to adjacent frame images to obtain initial matching feature point pairs. Perform mismatch removal processing on the initial matching feature point pairs to obtain matching feature point pairs. S2.4. Initialize the matched feature point pairs. After the initialization is successful, calculate the basic matrix corresponding to the image. Based on the basic matrix, calculate the rotation matrix and translation vector corresponding to the image to generate environmental pose data. Based on the environmental pose data and the matched feature point pairs, perform triangulation to generate map points. Simultaneously complete the construction of environmental map data. During the construction process, local optimization and global optimization are performed simultaneously.

[0008] Furthermore, step S3 includes the following sub-steps: S3.1. Based on the acquisition process error parameters of infrared sensing and visual imaging, establish infrared sensing uncertainty probability model and visual imaging uncertainty probability model respectively, and integrate the infrared sensing uncertainty probability model and visual imaging uncertainty probability model to generate sensing uncertainty probability model. S3.2. Input thermal radiation information, environmental image information, environmental pose data and environmental map data into the sensing uncertainty probability model, complete the prior prediction and posterior update of the data through Bayesian filtering, remove the noise components contained in the data, and obtain the denoised multi-source data. S3.3. Based on the particle filtering algorithm, particle sampling processing is performed on the denoised multi-source data. The sampled particles are weighted and updated to complete the particle resampling process. Based on the resampled particles, the fusion weights corresponding to the multi-source data are determined. Feature-level fusion processing of the multi-source data is performed according to the assigned fusion weights to obtain the fused environment data.

[0009] Furthermore, step S4 includes the following sub-steps: S4.1. Perform feature point matching and pose alignment processing on the fused environment data. Based on the aligned environment pose data and effective image feature points, perform triangulation processing to generate sparse 3D point cloud data corresponding to the target environment. Based on the pixel information and depth information in the fused environment data, perform densification processing to generate dense 3D point cloud data corresponding to the target environment. S4.2. Perform multi-frame point cloud registration processing on dense 3D point cloud data, complete the stitching and fusion of multi-frame point clouds, perform statistical filtering and radius filtering processing on the stitched 3D point cloud data, remove outliers and noise points, and obtain optimized 3D point cloud data. S4.3. Based on the optimized 3D point cloud data, perform 3D surface reconstruction processing to generate a 3D mesh model corresponding to the target environment. Extract texture data from the environmental image information, map the texture data onto the surface of the 3D mesh model, complete the texture mapping processing, and generate a 3D reconstruction model corresponding to the target environment.

[0010] Furthermore, in step S1, the environmental image information is acquired by a multi-view camera or a monocular camera combined with a fisheye lens. Before the acquisition operation begins, the hardware acquisition parameters of infrared sensing and visual imaging are jointly calibrated. The calibration process covers lens distortion parameters and sensor installation offset parameters. After calibration, the intrinsic and extrinsic parameter data corresponding to the acquisition process are generated. The intrinsic and extrinsic parameter data are used throughout the spatial coordinate matching processing of thermal radiation information and environmental image information. During the acquisition process, the acquisition quality of thermal radiation information and environmental image information is continuously detected, and invalid acquisition data that is blurred or overexposed is removed.

[0011] Furthermore, in step S2, during the synchronous localization and map construction process, keyframe filtering is performed on the acquired continuous image frames to extract image frames with effective feature points as keyframes. Environmental pose data calculation and environmental map data construction are completed only based on keyframes. Simultaneously, local optimization and global optimization are performed on all keyframes. The optimization process minimizes the error between the coordinates of all map points projected onto the image and the actual coordinates of the detected feature points by adjusting the environmental pose data and map point coordinates.

[0012] Furthermore, in step S3, during the multi-source data fusion process, feature-level fusion of thermal radiation information, environmental image information, environmental pose data, and environmental map data is completed according to preset fusion rules. The preset fusion rules first unify the feature dimensions of each type of data, then normalize the feature data of the same dimension, and finally assign different fusion weights according to the data acquisition quality and scene adaptability. During the fusion process, the temperature features of thermal radiation information and the texture features of environmental image information are retained, while the spatial location features of environmental pose data and the topological structure features of environmental map data are integrated.

[0013] Furthermore, in step S4, after the 3D reconstruction model is generated, the thermal radiation information and environmental image information of the target environment are continuously collected in real time. The data collected in real time is processed according to the processing flow of steps S1 to S3 to obtain real-time fused environmental data. The environmental change features in the real-time fused environmental data are extracted, and the 3D reconstruction model is incrementally updated based on the environmental change features. During the update process, only the areas that have changed are reconstructed, while the areas that have not changed retain the original data of the 3D reconstruction model.

[0014] Furthermore, in step S4, during the construction of the 3D reconstruction model, the processing strategy for 3D reconstruction is adjusted according to the scene type of the target environment. For small object scenes, the density of feature point detection is increased and the accuracy of the densification processing of 3D point cloud data is optimized. For natural environment scenes, the filtering processing of point cloud data is strengthened to remove invalid point clouds caused by vegetation. For urban environment scenes, the construction rules of the 3D mesh model are optimized to restore the geometric structural features of buildings. The processing strategies for different scene types are adapted to the texture mapping processing flow to ensure that the texture restoration effect of the 3D reconstruction model matches the spatial structure restoration effect.

[0015] The beneficial effects of this invention are: (1) Integrate the entire process of environmental multi-source information acquisition, synchronous positioning and map construction, data denoising and fusion and three-dimensional model construction to realize target environmental situation scanning and three-dimensional reconstruction, and ensure the complete closed loop of the reconstruction process; (2) Joint processing and feature fusion of two types of environmental information, infrared thermal radiation and visual images, can make up for the application limitations of single sensor information and improve the comprehensiveness and reliability of environmental perception data. (3) The reconstruction processing strategy can be adapted and adjusted according to different application scenarios, and the incremental update of the three-dimensional model is supported to ensure the consistency between the reconstruction model and the actual environment. Attached Figure Description

[0016] Figure 1 A flowchart illustrating the steps of a three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM. Figure 2 The following is a flowchart illustrating the specific steps of a three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM, provided as an example. Detailed Implementation

[0017] 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.

[0018] Example 1 See Figure 1 This embodiment provides a three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM, which includes the following steps: S1. Acquire thermal radiation information and environmental image information corresponding to the target environment. The thermal radiation information is acquired through infrared sensing, and the environmental image information is acquired through visual imaging. S2. Extract features from thermal radiation information and environmental image information to obtain image feature points, perform matching processing on image feature points, and perform synchronous localization and map construction based on the matched image feature points to generate environmental pose data and environmental map data. S3. Establish a sensing uncertainty probability model, and perform Bayesian filtering on thermal radiation information, environmental image information, environmental pose data and environmental map data based on the sensing uncertainty probability model to complete the denoising operation of multi-source data. Then, perform fusion processing on the denoised multi-source data to obtain fused environmental data. S4. Construct a 3D environmental model based on the fused environmental data to generate a 3D reconstruction model corresponding to the target environment.

[0019] In some embodiments, step S1 includes the following sub-steps: S1.1. Conduct full-range information collection on the target environment. During the collection process, a synchronous triggering method is used to simultaneously acquire thermal radiation information and environmental image information, ensuring that the collection frame rate of thermal radiation information and environmental image information remains consistent. S1.2. Timestamp the acquired thermal radiation information and environmental image information, and perform timestamp alignment processing on the thermal radiation information and environmental image information based on a unified time reference, so that the thermal radiation information and environmental image information maintain frame-by-frame correspondence in the time dimension. S1.3. Based on the pre-completed joint sensor calibration parameters, perform spatial coordinate transformation processing on the time-stamp aligned thermal radiation information and environmental image information, so that the thermal radiation information and environmental image information are mapped to the same spatial coordinate system, and complete the spatial coordinate matching processing.

[0020] In some embodiments, step S2 includes the following sub-steps: S2.1. Perform image preprocessing on thermal radiation information and environmental image information. The preprocessing includes image distortion correction, image grayscale conversion, and image enhancement. Image distortion correction is performed based on pre-calibrated sensor intrinsic parameters. Image enhancement is performed to adjust the contrast of thermal radiation information and environmental image information under low light conditions. S2.2. Feature point detection is performed on the preprocessed thermal radiation information and environmental image information to generate an initial feature point set. Non-maximum suppression processing and response threshold filtering processing are performed on the initial feature point set to obtain effective image feature points. Feature descriptors corresponding to the effective image feature points are generated simultaneously. S2.3. Perform similarity matching processing on the feature descriptors of effective image feature points corresponding to adjacent frame images to obtain initial matching feature point pairs. Perform mismatch removal processing on the initial matching feature point pairs to obtain matching feature point pairs. S2.4. Initialize the matched feature point pairs. After the initialization is successful, calculate the basic matrix corresponding to the image. Based on the basic matrix, calculate the rotation matrix and translation vector corresponding to the image to generate environmental pose data. Based on the environmental pose data and the matched feature point pairs, perform triangulation to generate map points. Simultaneously complete the construction of environmental map data. During the construction process, local optimization and global optimization are performed simultaneously.

[0021] In some embodiments, step S3 includes the following sub-steps: S3.1. Based on the acquisition process error parameters of infrared sensing and visual imaging, establish infrared sensing uncertainty probability model and visual imaging uncertainty probability model respectively, and integrate the infrared sensing uncertainty probability model and visual imaging uncertainty probability model to generate sensing uncertainty probability model. S3.2. Input thermal radiation information, environmental image information, environmental pose data and environmental map data into the sensing uncertainty probability model, complete the prior prediction and posterior update of the data through Bayesian filtering, remove the noise components contained in the data, and obtain the denoised multi-source data. S3.3. Based on the particle filtering algorithm, particle sampling processing is performed on the denoised multi-source data. The sampled particles are weighted and updated to complete the particle resampling process. Based on the resampled particles, the fusion weights corresponding to the multi-source data are determined. Feature-level fusion processing of the multi-source data is performed according to the assigned fusion weights to obtain the fused environment data.

[0022] In some embodiments, step S4 includes the following sub-steps: S4.1. Perform feature point matching and pose alignment processing on the fused environment data. Based on the aligned environment pose data and effective image feature points, perform triangulation processing to generate sparse 3D point cloud data corresponding to the target environment. Based on the pixel information and depth information in the fused environment data, perform densification processing to generate dense 3D point cloud data corresponding to the target environment. S4.2. Perform multi-frame point cloud registration processing on dense 3D point cloud data, complete the stitching and fusion of multi-frame point clouds, perform statistical filtering and radius filtering processing on the stitched 3D point cloud data, remove outliers and noise points, and obtain optimized 3D point cloud data. S4.3. Based on the optimized 3D point cloud data, perform 3D surface reconstruction processing to generate a 3D mesh model corresponding to the target environment. Extract texture data from the environmental image information, map the texture data onto the surface of the 3D mesh model, complete the texture mapping processing, and generate a 3D reconstruction model corresponding to the target environment.

[0023] In some embodiments, in step S1, environmental image information is acquired by a multi-view camera or a monocular camera combined with a fisheye lens. Before the acquisition operation begins, the hardware acquisition parameters of infrared sensing and visual imaging are jointly calibrated. The calibration process covers lens distortion parameters and sensor installation offset parameters. After calibration, intrinsic and extrinsic parameter data corresponding to the acquisition process are generated. The intrinsic and extrinsic parameter data are used throughout the spatial coordinate matching processing of thermal radiation information and environmental image information. During the acquisition process, the acquisition quality of thermal radiation information and environmental image information is continuously detected, and invalid acquisition data that is blurred or overexposed is removed.

[0024] In some embodiments, in step S2, during the synchronous localization and map building process, key frame filtering is performed on the acquired continuous image frames to extract image frames with effective feature points as key frames. Environmental pose data calculation and environmental map data construction are completed only based on key frames. Local optimization and global optimization are performed on all key frames simultaneously. The optimization process minimizes the error between the coordinates of all map points projected onto the image and the actual coordinates of the detected feature points by adjusting the environmental pose data and map point coordinates.

[0025] In some embodiments, in step S3, during the fusion processing of multi-source data, feature-level fusion of thermal radiation information, environmental image information, environmental pose data, and environmental map data is completed according to preset fusion rules. The preset fusion rules first unify the feature dimensions of each type of data, then normalize the feature data of the same dimension, and finally assign different fusion weights according to the data acquisition quality and scene adaptability. During the fusion process, the temperature features of thermal radiation information and the texture features of environmental image information are retained, while the spatial location features of environmental pose data and the topological structure features of environmental map data are integrated.

[0026] In some embodiments, in step S4, after the three-dimensional reconstruction model is generated, the thermal radiation information and environmental image information of the target environment are continuously collected in real time. The data collected in real time is processed according to the processing flow of steps S1 to S3 to obtain real-time fused environmental data. The environmental change features in the real-time fused environmental data are extracted. The three-dimensional reconstruction model is incrementally updated based on the environmental change features. During the update process, only the areas that have changed are reconstructed, while the areas that have not changed retain the original data of the three-dimensional reconstruction model.

[0027] In some embodiments, during step S4, the processing strategy for 3D reconstruction is adjusted according to the scene type of the target environment during the construction of the 3D reconstruction model. For small object scenes, the density of feature point detection is increased and the accuracy of the densification processing of 3D point cloud data is optimized. For natural environment scenes, the filtering processing of point cloud data is strengthened to remove invalid point clouds caused by vegetation. For urban environment scenes, the construction rules of the 3D mesh model are optimized to restore the geometric structural features of buildings. The processing strategies for different scene types are adapted to the texture mapping processing flow to ensure that the texture restoration effect of the 3D reconstruction model matches the spatial structure restoration effect.

[0028] Example 2 This embodiment provides a specific implementation process for a three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM, such as... Figure 2 As shown, the specific implementation steps are as follows: S1. Environmental multi-source information collection: S1.1. Joint calibration and processing of sensor acquisition parameters: Before the data acquisition operation begins, the hardware acquisition parameters of the infrared sensor and visual imaging are jointly calibrated. The calibration process covers lens distortion parameters and sensor installation offset parameters. The calibration process first completes the single-sensor intrinsic parameter calibration for both infrared and visual imaging, obtaining the lens distortion coefficient, focal length parameters, and imaging origin parameters for each sensor. After completing the single-sensor intrinsic parameter calibration, the dual-sensor extrinsic parameter calibration is performed based on the calibration board, obtaining the spatial position offset parameters and rotation offset parameters between the infrared sensor and visual imaging. The single-sensor intrinsic parameters and the dual-sensor extrinsic parameters are integrated to generate the corresponding intrinsic and extrinsic parameter data for the acquisition process. These intrinsic and extrinsic parameter data are used throughout the spatial coordinate matching processing of thermal radiation information and environmental image information. Environmental image information is acquired through multi-view cameras or a monocular camera combined with a fisheye lens, while thermal radiation information is acquired through infrared sensors. The installation position and orientation of the acquisition equipment remain fixed after calibration until the entire single acquisition operation is completed.

[0029] In some embodiments, the joint calibration process can be repeated after a single acquisition operation is completed. The calibration process involves acquiring multiple sets of calibration images through a calibration board, and performing error verification on the generated intrinsic and extrinsic data based on the calibration images. When the verification error exceeds a preset range, the joint calibration process is completed again, and the intrinsic and extrinsic data are updated.

[0030] S1.2. Synchronous collection of information across the entire target environment: The system performs full-range information collection on the target environment, employing a synchronous triggering method to simultaneously acquire thermal radiation information and environmental image information, ensuring consistent frame rates for both. The collection process follows a pre-planned path to achieve full coverage of the target environment, guaranteeing that the entire target environment is within the sensor's coverage area with no blind spots. Continuous quality monitoring of the acquired thermal radiation and environmental image information is performed during the collection process. For each set of synchronously acquired thermal radiation and environmental image data, image sharpness and exposure are checked, blurry and overexposed invalid data are removed, and the areas corresponding to invalid data are re-acquired to ensure effective data coverage of the entire target environment.

[0031] S1.3. Timestamp marking and alignment of collected information: The acquired thermal radiation information and environmental image information are timestamped. Based on a unified time reference, the thermal radiation information and environmental image information are timestamped to ensure frame-by-frame correspondence in the time dimension. The timestamping process is completed simultaneously with the acquisition trigger. Each group of synchronously acquired thermal radiation information and environmental image information is marked with the same timestamp identifier, which is generated based on a unified hardware clock to ensure the uniqueness of the time reference. The timestamp alignment process first performs an initial matching of thermal radiation information and environmental image information based on the timestamp identifier, grouping thermal radiation information and environmental image information with the same timestamp identifier into the same data group. Then, frame synchronization verification is performed on the thermal radiation information and environmental image information within the same data group. After verification, frame-by-frame correspondence in the time dimension is completed, ensuring that the thermal radiation information and environmental image information within the same data group correspond to the same acquisition time and the same acquisition area of ​​the target environment.

[0032] S1.4. Spatial coordinate matching processing of collected information: Based on the pre-completed sensor joint calibration parameters, namely the intrinsic and extrinsic parameter data generated in S1.1, spatial coordinate transformation is performed on the time-stamp-aligned thermal radiation information and environmental image information to map them to the same spatial coordinate system, completing the spatial coordinate matching process. The spatial coordinate transformation process first corrects imaging distortion for both thermal radiation information and environmental image information based on the intrinsic parameter data, eliminating pixel coordinate offsets caused by lens distortion. Then, based on the extrinsic parameter data, it transforms the pixel coordinate system of the thermal radiation information to the pixel coordinate system of the visual imaging, ensuring that each pixel of the thermal radiation information and the pixel of the environmental image information form a spatial correspondence in the same pixel coordinate system. Finally, the two sets of data are uniformly mapped to the world coordinate system, completing the spatial coordinate matching process. After the spatial coordinate matching process is completed, spatial correspondence verification is performed on the thermal radiation information and environmental image information within the same data set to ensure that the two sets of data completely correspond to the same acquisition area in spatial dimension. Data sets that fail the verification are returned for re-performance of spatial coordinate transformation until the verification succeeds.

[0033] S2. Image Feature Processing and Simultaneous Localization and Mapping: S2.1. Image preprocessing for acquired information: Image preprocessing is performed on thermal radiation information and environmental image information. Preprocessing includes image distortion correction, image grayscale conversion, and image enhancement. Image distortion correction is performed based on pre-calibrated sensor intrinsic parameters, while image enhancement adjusts the contrast of thermal radiation information and environmental image information under low-light conditions. The image distortion correction process, based on the intrinsic parameter data generated in S1.1, corrects distortion for each pixel of the thermal radiation information and environmental image information, eliminating pixel position shifts caused by radial and tangential distortion of the lens, resulting in distorted image data. The image grayscale conversion process, for the distorted environmental image information, converts multi-channel color image data into single-channel grayscale image data. The conversion process is based on preset pixel weight allocation rules, preserving the image's texture and edge features. After grayscale conversion, both thermal radiation information and environmental image information are single-channel image data, unifying the data format for subsequent processing. The image enhancement process adaptively adjusts the image contrast based on thermal radiation and environmental image information acquired in low-light environments. This adjustment is based on the image's pixel grayscale distribution, improving detail in dark areas while suppressing overexposure in bright areas. It also enhances extractable features, ensuring a sufficient number of effective feature points are acquired in subsequent feature point detection. After preprocessing, the validity of the image data is verified by checking the feature richness. Image data with feature richness below a preset standard is marked as requiring supplementation, and data is collected in the corresponding areas to ensure all image data used for subsequent processing possesses sufficient extractable features.

[0034] S2.2. Image Feature Point Detection and Descriptor Generation: Feature point detection is performed on the preprocessed thermal radiation information and environmental image information to generate an initial feature point set. Non-maximum suppression (NMS) and response threshold filtering are then applied to this initial feature point set to obtain valid image feature points. Simultaneously, feature descriptors corresponding to these valid image feature points are generated. The feature point detection process detects corner and edge features for each set of preprocessed image data, identifying stable and discriminative pixels as initial feature points to generate the initial feature point set. The NMS process compares the response values ​​of each feature point in the initial feature point set within its neighborhood, retaining only the feature point with the highest response value in the neighborhood, eliminating duplicate feature points, and reducing feature point redundancy. The response threshold filtering process, based on a preset feature point response threshold, filters the feature points processed by NMS, removing those with response values ​​below the preset threshold and retaining those with response values ​​that meet the requirements as valid image feature points. After determining the valid image feature points, a corresponding feature descriptor is generated for each valid feature point. The feature descriptor is generated based on the pixel grayscale distribution within the feature point's neighborhood and is used to characterize the feature attributes of the corresponding feature point. This ensures that feature descriptors generated from feature points at the same spatial location in different frames have high similarity, and feature descriptors generated from feature points at different spatial locations have high discriminative power. After the feature descriptor generation is complete, the number of valid image feature points is verified. If the number of valid image feature points in a single frame is lower than a preset standard, it is marked as an invalid frame and will not proceed to the subsequent matching process. Data supplementation is then performed in the corresponding acquisition area.

[0035] In some embodiments, the response threshold for feature point detection can be adjusted according to the richness of texture in the image information during the feature point detection process. In image information with rich texture, the response threshold can be increased to filter out more stable feature points, while in image information with sparse texture, the response threshold can be decreased to ensure that a sufficient number of effective feature points can be extracted.

[0036] S2.3. Feature point matching processing of adjacent frames: The process involves similarity matching of feature descriptors for effective image feature points in adjacent frames to obtain initial matched feature point pairs. Mismatch removal is then performed on these initial matched feature point pairs to obtain fully matched feature point pairs. The feature point similarity matching process involves acquiring effective image feature points and feature descriptors for each of two consecutively acquired adjacent frames, calculating the similarity between the two sets of descriptors based on distance. Feature points whose distance meets a preset matching standard are grouped into an initial matched feature point pair. After matching all feature points, an initial matched feature point pair set is generated. The mismatch removal process involves using random sampling consistency to remove mismatched feature point pairs from the initial matched feature point pair set. This process involves fitting a matching model that conforms to the image pose transformation relationship through multiple random samplings. The initial matched feature point pairs are then validated based on the matching model. Feature point pairs that do not conform to the matching model are marked as mismatched and removed, while those that conform are retained, ultimately resulting in fully matched feature point pairs. After matching is completed, the number of valid matching feature point pairs is checked. If the number of valid matching feature point pairs between adjacent frames is lower than the preset standard, the subsequent pose calculation will not proceed. Instead, image data of the corresponding area will be collected and the matching process will be completed again.

[0037] S2.4. Synchronous Positioning and Map Data Construction Processing: The process begins by initializing the matched feature point pairs. If the initialization passes, the corresponding fundamental matrix of the image is calculated. Based on this matrix, the rotation matrix and translation vector are calculated to generate environmental pose data. Triangulation is then performed on the environmental pose data and the matched feature point pairs to generate map points, simultaneously constructing the environmental map data. Local and global optimization processes are performed concurrently during this construction process. During simultaneous localization and map construction, keyframe filtering is performed on consecutive acquired image frames. Image frames with valid feature points are extracted as keyframes, and environmental pose data calculation and environmental map data construction are completed solely based on these keyframes. The initialization process involves verifying the initial two image frames based on the matched feature point pairs between them. This verification includes checking the number of matched feature point pairs and disparity. If the verification passes, initialization is complete, and the process proceeds to the subsequent pose calculation stage. After initialization, for each subsequent set of adjacent keyframes, the fundamental matrix corresponding to the image is calculated based on the matched feature point pairs. The rotation matrix and translation vector are then decomposed to obtain the corresponding rotation matrix and translation vector. These rotation and translation vectors characterize the position and orientation of the acquisition device in the world coordinate system. Integrating the rotation matrices and translation vectors of all image frames generates environmental pose data. Simultaneously, triangulation is performed on the environmental pose data and matched feature point pairs to calculate the 3D spatial coordinates of the feature points, generating map points. The set of all map points constitutes the foundation of the environmental map data. During the construction of the environmental map data, local and global optimization processes are performed simultaneously on all keyframes. Local optimization adjusts the environmental pose data and map points corresponding to consecutive adjacent keyframes, while global optimization adjusts the environmental pose data and all map points corresponding to all keyframes. This optimization minimizes the sum of errors between the coordinates of all map points projected onto the image and the actual coordinates of the detected feature points by adjusting the environmental pose data and map point coordinates. After optimization, the consistency between environmental pose data and environmental map data is verified to ensure that the spatial correspondence between pose data and map point data is accurate. If the verification fails, the optimization process is repeated until the verification passes.

[0038] S3. Denoising and Fusion Processing of Multi-Source Data: S3.1. Construction of the sensing uncertainty probability model: Based on the acquisition process error parameters of infrared sensing and visual imaging, uncertainty probability models for infrared sensing and visual imaging are established separately. These two models are then integrated to generate a comprehensive sensing uncertainty probability model. The infrared sensing uncertainty probability model is based on the acquisition noise distribution of infrared sensing. The input is the acquired thermal radiation information, and the output is the noise probability distribution corresponding to the thermal radiation information. The model construction process is based on the factory error parameters of the infrared sensor, the error parameters obtained during calibration, and the noise statistics results from the actual acquisition process. The model can characterize the probability of noise occurrence and the noise distribution range corresponding to the value of each pixel in the thermal radiation information. The visual imaging uncertainty probability model is based on the acquisition distortion and pixel noise distribution of visual imaging. The input is the acquired environmental image information, and the output is the noise probability distribution corresponding to the environmental image information. The model construction process is based on the lens distortion parameters of visual imaging, the error parameters obtained during calibration, and the pixel noise statistics results from the actual acquisition process. The model can characterize the probability of noise occurrence and the noise distribution range corresponding to the value of each pixel in the environmental image information. The infrared sensing uncertainty probability model and the visual imaging uncertainty probability model are integrated in parallel to form a unified sensing uncertainty probability model. This model can simultaneously receive multi-source acquired data, environmental pose data, and environmental map data, and output the noise probability distribution results for the corresponding data, providing a probabilistic basis for subsequent denoising processing. After the model is built, it is validated using standard data collected for calibration to ensure that the noise probability distribution results output by the model are consistent with the noise distribution in the actual data. If the validation fails, the model parameters are readjusted until the validation passes.

[0039] S3.2. Multi-source data Bayesian filtering noise reduction processing: Thermal radiation information, environmental image information, environmental pose data, and environmental map data are input into the sensing uncertainty probability model. Bayesian filtering is used to perform prior prediction and posterior update of the data, removing noise components to obtain denoised multi-source data. The Bayesian filtering process first performs prior prediction, predicting the prior probability distribution of the data at the current time based on the filtering result of the previous time step and the noise probability distribution output by the sensing uncertainty probability model. Then, it performs posterior update, correcting the prior probability distribution based on the actual observation data collected at the current time step to obtain the posterior probability distribution of the data at the current time step. Based on the posterior probability distribution, it performs optimal estimation of the data, removing components that conform to the noise distribution and retaining the effective signal components, thus completing the denoising process for the multi-source data. The denoising process covers all data content of thermal radiation information, environmental image information, environmental pose data, and environmental map data, ensuring that all data entering the subsequent fusion processing stage undergoes noise removal, improving data reliability. After denoising, the signal-to-noise ratio (SNR) of the denoised data is checked. Data with an SNR lower than the preset standard undergoes denoising again until the check passes.

[0040] In some embodiments, the Bayesian filtering process can be performed using extended Kalman filtering to linearize the nonlinear acquisition error distribution, thereby improving the adaptability and accuracy of the denoising process.

[0041] S3.3. Feature-level fusion processing of multi-source data: The particle filtering algorithm is used to perform particle sampling on the denoised multi-source data. Weights are assigned and updated for the sampled particles, and particle resampling is then performed. Based on the resampled particles, the corresponding fusion weights for the multi-source data are determined. Feature-level fusion processing is then performed according to the assigned fusion weights to obtain fused environmental data. During the multi-source data fusion process, feature-level fusion of thermal radiation information, environmental image information, environmental pose data, and environmental map data is completed according to preset fusion rules. The preset fusion rules first unify the feature dimensions of each type of data, then normalize feature data of the same dimension, and finally assign different fusion weights based on the data acquisition quality and scene adaptability. The fusion process preserves the temperature features of thermal radiation information and the texture features of environmental image information, while also integrating the spatial location features of environmental pose data and the topological structure features of environmental map data. The particle sampling process involves sampling a preset number of particles from the denoised multi-source data. Each particle corresponds to a set of multi-source data fusion weight allocation schemes. The weight allocation and update process assigns weights to each particle based on the observation accuracy and feature matching degree of the multi-source data, and then updates the particle weights based on actual observation results, increasing the weights of particles with high matching degrees and decreasing the weights of particles with low matching degrees. The particle resampling process removes particles with weights below a preset threshold, while simultaneously replicating and supplementing particles with higher weights to ensure a stable total number of particles. After resampling, the optimal fusion weights for the multi-source data are calculated based on the weights of all particles and their corresponding fusion weight schemes. Based on the optimal fusion weights, feature-level fusion of the multi-source data is completed. The feature data from thermal radiation information, environmental image information, environmental pose data, and environmental map data are weighted and fused according to the fusion weights to generate unified fused environmental data. This fused environmental data simultaneously includes the temperature features, texture features, spatial location features, and topological structure features of the target environment, providing a complete environmental data foundation for subsequent 3D model construction. After the fusion process is completed, the integrity of the fused environment data is verified to ensure that the fused environment data covers the entire area of ​​the target environment and has no missing features. If the verification fails, the fusion process is repeated until the verification passes.

[0042] S4. Environmental 3D Model Construction and Update: S4.1. Generation and processing of 3D point cloud data: Feature point matching and pose alignment are performed on the fused environmental data. Based on the aligned environmental pose data and effective image feature points, triangulation is performed to generate sparse 3D point cloud data corresponding to the target environment. Then, based on the pixel and depth information in the fused environmental data, densification is performed to generate dense 3D point cloud data corresponding to the target environment. In the feature point matching and pose alignment process, cross-frame feature point matching and pose alignment are performed for all keyframes in the fused environmental data, ensuring that the pose data of all keyframes are unified in the same world coordinate system and that the spatial correspondence of feature points is accurate. In the triangulation process, based on the aligned environmental pose data, the 3D spatial coordinates of the same matched feature point in different keyframes are calculated to obtain the corresponding 3D spatial points. The set of all 3D spatial points constitutes the sparse 3D point cloud data corresponding to the target environment, which can represent the overall spatial structure and contour of the target environment. The densification process, based on pixel and depth information from the fused environmental data, performs pixel-level densification expansion on the sparse 3D point cloud data. It calculates the 3D spatial coordinates corresponding to each effective pixel in the image, generating dense 3D point cloud data covering the entire target environment. This dense 3D point cloud data can characterize the detailed spatial structure and surface features of the target environment. After the point cloud data is generated, the integrity of both the sparse and dense 3D point cloud data is verified to ensure that the point cloud data covers the entire target environment without missing spatial structures. If the verification fails, the point cloud generation process is repeated until it passes.

[0043] S4.2. Optimization and processing of 3D point cloud data: Multi-frame point cloud registration is performed on dense 3D point cloud data to stitch and fuse the points. Statistical filtering and radius filtering are then applied to the stitched 3D point cloud data to remove outliers and noise points, resulting in optimized 3D point cloud data. The multi-frame point cloud registration process performs coarse and fine registration on dense 3D point cloud data generated from different keyframes. Coarse registration uses feature descriptors to perform initial position alignment, while fine registration uses an iterative nearest-point algorithm to achieve precise registration. This ensures accurate stitching and fusion of point cloud data from different frames within the same world coordinate system, generating complete point cloud data covering the entire target environment. The statistical filtering process calculates the average distance from each point to all points in its neighborhood. Based on the statistical distribution of this average distance, outliers with average distances exceeding a preset range are removed. The radius filtering process counts the number of points in a preset radius neighborhood for each point and removes noise points with fewer points in the neighborhood than a preset threshold. After two filtering processes, outliers and noise points in the point cloud data are removed, resulting in optimized 3D point cloud data. This optimized point cloud data possesses higher accuracy and completeness, accurately representing the spatial structure of the target environment. Following point cloud optimization, the accuracy of the point cloud data is verified to ensure that the spatial positional error is within acceptable limits. Data that fails verification is re-registered and filtered until it passes verification.

[0044] In some embodiments, voxel downsampling can be used to simplify the stitched 3D point cloud data during the point cloud data optimization process. This reduces the redundancy of the point cloud data and improves the efficiency of subsequent processing while preserving the overall spatial structure characteristics of the point cloud data.

[0045] S4.3. 3D Mesh Model Construction and Texture Mapping Processing: Based on optimized 3D point cloud data, 3D surface reconstruction is performed to generate a 3D mesh model corresponding to the target environment. Texture data is extracted from the environmental image information and mapped onto the surface of the 3D mesh model to complete texture mapping processing and generate a 3D reconstructed model of the target environment. During the construction of the 3D reconstruction model, the 3D reconstruction processing strategy is adjusted according to the scene type of the target environment. For scenes with small objects, the density of feature point detection is increased and the accuracy of the densification processing of 3D point cloud data is optimized. For natural environment scenes, the filtering processing of point cloud data is strengthened to remove invalid point clouds from vegetation. For urban environment scenes, the construction rules of the 3D mesh model are optimized to restore the geometric structural features of buildings. The processing strategies for different scene types are adapted to the texture mapping processing flow to ensure that the texture restoration effect of the 3D reconstructed model matches the spatial structure restoration effect. The 3D surface reconstruction process is based on optimized 3D point cloud data and uses Poisson surface reconstruction to complete the construction of the 3D mesh model. The construction process first calculates the surface indicator function of the target environment based on the point cloud data, then extracts isosurfaces based on the surface indicator function to generate a 3D mesh model composed of triangular facets. The 3D mesh model can accurately represent the 3D geometric structure of the target environment. The texture data extraction process extracts image texture data corresponding to each keyframe from the environmental image information, completing distortion correction and color correction of the texture data to ensure accuracy and consistency. The texture mapping process, based on the pixel correspondence between the faces of the 3D mesh model and the environmental image information, maps the extracted texture data onto the corresponding face surfaces of the 3D mesh model, completing texture fusion and stitching, eliminating seams, and generating a 3D reconstructed model with complete texture information. After the 3D reconstructed model is generated, the geometric accuracy and texture restoration effect of the model are verified to ensure that the model's geometry matches the actual structure of the target environment and that the texture information matches the actual appearance of the target environment. If the verification fails, the mesh construction and texture mapping process are repeated until the verification passes.

[0046] S4.4. Incremental update processing of 3D reconstruction model: After the 3D reconstruction model is generated, real-time acquisition of thermal radiation and environmental image information of the target environment is continuously performed. The real-time acquired data is processed according to the aforementioned information acquisition, feature processing and localization mapping, denoising and fusion workflow to obtain real-time fused environmental data. Environmental change features are extracted from the real-time fused environmental data, and the 3D reconstruction model is incrementally updated based on these features. During the update process, only the areas that have changed are reconstructed; the unchanged areas retain the original data of the 3D reconstruction model. The real-time acquisition process is completed using the same parameters and rules as the initial acquisition, ensuring that the real-time acquired data has the same format and accuracy as the initial acquired data. The real-time data processing process fully follows the aforementioned workflow, completing the calibration, alignment, preprocessing, feature extraction, localization mapping, denoising and fusion processing of the real-time acquired data to generate real-time fused environmental data. The environmental change feature extraction process compares the real-time fused environmental data with the initial fused environmental data, identifying areas in the target environment where spatial structure and appearance features have changed, marking these as changed areas, and marking the unchanged areas as unchanging areas. The incremental update process only targets the changed areas, completing the corresponding point cloud data generation, optimization, mesh model construction, and texture mapping. This generates 3D model fragments corresponding to the changed areas, which replace the content of the corresponding areas in the original 3D reconstructed model, thus completing the model update. The 3D model content of the unchanging areas remains unchanged and does not require reconstruction. After the model update is complete, the consistency of the updated 3D reconstructed model is verified to ensure a smooth transition in model structure and texture information between the updated and unupdated areas, with no splicing errors. If the verification fails, the update process is repeated until it passes.

[0047] In some embodiments, the incremental update process of the 3D reconstruction model can be executed cyclically according to a preset time period, or it can be triggered when a significant change in the target environment is detected, so as to ensure that the 3D reconstruction model always remains consistent with the actual state of the target environment.

[0048] This embodiment employs a complete environmental situation scanning 3D reconstruction processing flow to achieve synchronous acquisition and matching processing of multi-source environmental information of the target environment, ensuring consistency of the acquired information in both temporal and spatial dimensions, and providing reliable basic data for subsequent processing. Through denoising and fusion processing of multi-source data, it integrates the characteristic advantages of different types of acquired information, improving the integrity and reliability of environmental data and fully preserving the multi-dimensional characteristic information of the target environment. Through synchronous positioning and map construction processing, it achieves pose calculation and environmental map construction during the acquisition process, ensuring the accuracy of pose and map data through optimization processing. Through 3D model construction processing after multi-source information fusion, it completes the 3D model reconstruction of the target environment, achieving a complete restoration of the target environment's situation. The above processing flow is adaptable to different types of target environments, adjusting processing strategies according to scene type to ensure the adaptability and processing effect of 3D reconstruction processing in different scenarios. After the 3D reconstruction model is generated, it can be incrementally updated based on real-time acquired environmental information to ensure consistency between the 3D reconstruction model and the actual state of the target environment. Meanwhile, all the steps of the above processing flow form a complete data flow link. The processing result of the previous step is directly used as the input data of the next step. A stable connection relationship is formed between each step, ensuring the continuity and stability of the entire processing flow. Those skilled in the art can completely reproduce the entire processing flow based on the above process to achieve three-dimensional reconstruction of the target environment.

[0049] The above description is merely a preferred embodiment of the present invention. It should be understood that the present invention is not limited to the forms disclosed herein and should not be construed as excluding other embodiments. It can be used in various other combinations, modifications, and environments, and can be altered within the scope of the concept described herein through the above teachings or related technologies or knowledge. Modifications and variations made by those skilled in the art that do not depart from the spirit and scope of the present invention should be within the protection scope of the appended claims.

Claims

1. A three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM, characterized in that, Includes the following steps: S1. Acquire thermal radiation information and environmental image information corresponding to the target environment. The thermal radiation information is acquired through infrared sensing, and the environmental image information is acquired through visual imaging. S2. Extract features from thermal radiation information and environmental image information to obtain image feature points, perform matching processing on image feature points, and perform synchronous localization and map construction based on the matched image feature points to generate environmental pose data and environmental map data. S3. Establish a sensing uncertainty probability model, and perform Bayesian filtering on thermal radiation information, environmental image information, environmental pose data and environmental map data based on the sensing uncertainty probability model to complete the denoising operation of multi-source data. Then, perform fusion processing on the denoised multi-source data to obtain fused environmental data. S4. Construct a 3D environmental model based on the fused environmental data to generate a 3D reconstruction model corresponding to the target environment.

2. The three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, Step S1 includes the following sub-steps: S1.

1. Conduct full-range information collection on the target environment. During the collection process, a synchronous triggering method is used to simultaneously acquire thermal radiation information and environmental image information, ensuring that the collection frame rate of thermal radiation information and environmental image information remains consistent. S1.

2. Timestamp the acquired thermal radiation information and environmental image information, and perform timestamp alignment processing on the thermal radiation information and environmental image information based on a unified time reference, so that the thermal radiation information and environmental image information maintain frame-by-frame correspondence in the time dimension. S1.

3. Based on the pre-completed joint sensor calibration parameters, perform spatial coordinate transformation processing on the time-stamp aligned thermal radiation information and environmental image information, so that the thermal radiation information and environmental image information are mapped to the same spatial coordinate system, and complete the spatial coordinate matching processing.

3. The three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, Step S2 includes the following sub-steps: S2.

1. Perform image preprocessing on thermal radiation information and environmental image information. The preprocessing includes image distortion correction, image grayscale conversion, and image enhancement. Image distortion correction is performed based on pre-calibrated sensor intrinsic parameters. Image enhancement is performed to adjust the contrast of thermal radiation information and environmental image information under low light conditions. S2.

2. Feature point detection is performed on the preprocessed thermal radiation information and environmental image information to generate an initial feature point set. Non-maximum suppression processing and response threshold filtering processing are performed on the initial feature point set to obtain effective image feature points. Feature descriptors corresponding to the effective image feature points are generated simultaneously. S2.

3. Perform similarity matching processing on the feature descriptors of effective image feature points corresponding to adjacent frame images to obtain initial matching feature point pairs. Perform mismatch removal processing on the initial matching feature point pairs to obtain matching feature point pairs. S2.

4. Initialize the matched feature point pairs. After the initialization is successful, calculate the basic matrix corresponding to the image. Based on the basic matrix, calculate the rotation matrix and translation vector corresponding to the image to generate environmental pose data. Based on the environmental pose data and the matched feature point pairs, perform triangulation to generate map points. Simultaneously complete the construction of environmental map data. During the construction process, local optimization and global optimization are performed simultaneously.

4. The three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, Step S3 includes the following sub-steps: S3.

1. Based on the acquisition process error parameters of infrared sensing and visual imaging, establish infrared sensing uncertainty probability model and visual imaging uncertainty probability model respectively, and integrate the infrared sensing uncertainty probability model and visual imaging uncertainty probability model to generate sensing uncertainty probability model. S3.

2. Input thermal radiation information, environmental image information, environmental pose data and environmental map data into the sensing uncertainty probability model, complete the prior prediction and posterior update of the data through Bayesian filtering, remove the noise components contained in the data, and obtain the denoised multi-source data. S3.

3. Based on the particle filtering algorithm, particle sampling processing is performed on the denoised multi-source data. The sampled particles are weighted and updated to complete the particle resampling process. Based on the resampled particles, the fusion weights corresponding to the multi-source data are determined. Feature-level fusion processing of the multi-source data is performed according to the assigned fusion weights to obtain the fused environment data.

5. The three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, Step S4 includes the following sub-steps: S4.

1. Perform feature point matching and pose alignment processing on the fused environment data. Based on the aligned environment pose data and effective image feature points, perform triangulation processing to generate sparse 3D point cloud data corresponding to the target environment. Based on the pixel information and depth information in the fused environment data, perform densification processing to generate dense 3D point cloud data corresponding to the target environment. S4.

2. Perform multi-frame point cloud registration processing on dense 3D point cloud data, complete the stitching and fusion of multi-frame point clouds, perform statistical filtering and radius filtering processing on the stitched 3D point cloud data, remove outliers and noise points, and obtain optimized 3D point cloud data. S4.

3. Based on the optimized 3D point cloud data, perform 3D surface reconstruction processing to generate a 3D mesh model corresponding to the target environment. Extract texture data from the environmental image information, map the texture data onto the surface of the 3D mesh model, complete the texture mapping processing, and generate a 3D reconstruction model corresponding to the target environment.

6. The three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, In step S1, environmental image information is acquired through multi-view cameras or a single-view camera combined with a fisheye lens. Before the acquisition operation begins, the hardware acquisition parameters of infrared sensing and visual imaging are jointly calibrated. The calibration process covers lens distortion parameters and sensor installation offset parameters. After calibration, intrinsic and extrinsic parameter data corresponding to the acquisition process are generated. The intrinsic and extrinsic parameter data are used throughout the spatial coordinate matching processing of thermal radiation information and environmental image information. During the acquisition process, the acquisition quality of thermal radiation information and environmental image information is continuously detected, and invalid acquisition data that is blurred or overexposed is removed.

7. The three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, In step S2, during the synchronous localization and map construction process, keyframe filtering is performed on the acquired continuous image frames to extract image frames with effective feature points as keyframes. Environmental pose data calculation and environmental map data construction are completed only based on keyframes. Simultaneously, local optimization and global optimization are performed on all keyframes. The optimization process minimizes the error between the coordinates of all map points projected onto the image and the actual coordinates of the detected feature points by adjusting the environmental pose data and map point coordinates.

8. The three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, In step S3, during the multi-source data fusion process, feature-level fusion of thermal radiation information, environmental image information, environmental pose data, and environmental map data is completed according to preset fusion rules. The preset fusion rules first unify the feature dimensions of each type of data, then normalize the feature data of the same dimension, and finally assign different fusion weights according to the data acquisition quality and scene adaptability. During the fusion process, the temperature features of thermal radiation information and the texture features of environmental image information are retained, while the spatial location features of environmental pose data and the topological structure features of environmental map data are integrated.

9. The three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, In step S4, after the 3D reconstruction model is generated, the thermal radiation information and environmental image information of the target environment are continuously collected in real time. The data collected in real time is processed according to the processing flow of steps S1 to S3 to obtain real-time fused environmental data. The environmental change features in the real-time fused environmental data are extracted. Based on the environmental change features, the 3D reconstruction model is incrementally updated. During the update process, only the areas that have changed are reconstructed, while the areas that have not changed retain the original data of the 3D reconstruction model.

10. A three-dimensional reconstruction method for environmental situation scanning based on infrared and SLAM according to claim 1, characterized in that, In step S4, during the construction of the 3D reconstruction model, the processing strategy for 3D reconstruction is adjusted according to the scene type of the target environment. For small object scenes, the density of feature point detection is increased and the accuracy of the densification processing of 3D point cloud data is optimized. For natural environment scenes, the filtering processing of point cloud data is strengthened to remove invalid point clouds caused by vegetation. For urban environment scenes, the construction rules of the 3D mesh model are optimized to restore the geometric structural features of buildings. The processing strategies for different scene types are adapted to the texture mapping processing flow to ensure that the texture restoration effect of the 3D reconstruction model matches the spatial structure restoration effect.

Images